{ "base": { "dir": "~/AI/local_models_exl/glm-4.6-exl3-3bpw", "bpw": 3.016001200732827 }, "alts": [ { "dir": "~/AI/local_models_exl/glm-4.6-exl3-4bpw", "bpw": 4.015793627432631 }, { "dir": "~/AI/local_models_exl/glm-4.6-exl3-5bpw", "bpw": 5.015586054132434 }, { "dir": "~/AI/local_models_exl/glm-4.6-exl3-6bpw", "bpw": 6.015378480832238 }, { "dir": "~/AI/local_models_exl/glm-4.6-exl3-8bpw", "bpw": 8.014963334231846 } ], "groups": [ { "idx": 0, "layers": [ "model.layers.0.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00013828780502081472, "dbits": 62914560 }, { "dkld": -8.892230689526437e-05, "dbits": 125829120 }, { "dkld": 2.608532086013915e-05, "dbits": 188743680 }, { "dkld": -0.00010220818221569339, "dbits": 314572800 } ] }, { "idx": 1, "layers": [ "model.layers.0.self_attn.k_proj", "model.layers.0.self_attn.v_proj" ], "candidates": [ { "dkld": -4.8393197357665674e-05, "dbits": 10485760 }, { "dkld": -5.9247855097060986e-05, "dbits": 20971520 }, { "dkld": 3.914535045623779e-05, "dbits": 31457280 }, { "dkld": -7.371120154858746e-05, "dbits": 52428800 } ] }, { "idx": 2, "layers": [ "model.layers.0.self_attn.o_proj" ], "candidates": [ { "dkld": -0.000403262116014963, "dbits": 62914560 }, { "dkld": -0.00022487062960863113, "dbits": 125829120 }, { "dkld": -0.00014930143952369967, "dbits": 188743680 }, { "dkld": -0.00023164171725512106, "dbits": 314572800 } ] }, { "idx": 3, "layers": [ "model.layers.0.mlp.gate_proj", "model.layers.0.mlp.up_proj" ], "candidates": [ { "dkld": -0.0004395037889480674, "dbits": 125829120 }, { "dkld": -0.0005234956741333036, "dbits": 251658240 }, { "dkld": -0.0005291301757097272, "dbits": 377487360 }, { "dkld": -0.0005393324419856182, "dbits": 629145600 } ] }, { "idx": 4, "layers": [ "model.layers.0.mlp.down_proj" ], "candidates": [ { "dkld": -0.00036360789090394974, "dbits": 62914560 }, { "dkld": -0.00018305480480194924, "dbits": 125829120 }, { "dkld": -0.0002387018874287633, "dbits": 188743680 }, { "dkld": -0.00028041424229741096, "dbits": 314572800 } ] }, { "idx": 5, "layers": [ "model.layers.1.self_attn.q_proj" ], "candidates": [ { "dkld": -7.69447535276524e-05, "dbits": 62914560 }, { "dkld": 4.411116242400426e-06, "dbits": 125829120 }, { "dkld": 7.813265547156056e-05, "dbits": 188743680 }, { "dkld": -2.723168581725277e-05, "dbits": 314572800 } ] }, { "idx": 6, "layers": [ "model.layers.1.self_attn.k_proj", "model.layers.1.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00011766962707043804, "dbits": 10485760 }, { "dkld": -0.0003419622778892545, "dbits": 20971520 }, { "dkld": -0.00022323504090310253, "dbits": 31457280 }, { "dkld": -0.00012209862470627386, "dbits": 52428800 } ] }, { "idx": 7, "layers": [ "model.layers.1.self_attn.o_proj" ], "candidates": [ { "dkld": 5.6345947086755555e-06, "dbits": 62914560 }, { "dkld": 6.732549518345676e-05, "dbits": 125829120 }, { "dkld": -0.00011621573939919749, "dbits": 188743680 }, { "dkld": 4.526600241652745e-06, "dbits": 314572800 } ] }, { "idx": 8, "layers": [ "model.layers.1.mlp.gate_proj", "model.layers.1.mlp.up_proj" ], "candidates": [ { "dkld": -0.00024302629753947258, "dbits": 125829120 }, { "dkld": -0.00036743637174369015, "dbits": 251658240 }, { "dkld": -0.00038665514439345516, "dbits": 377487360 }, { "dkld": -0.0004042137414217023, "dbits": 629145600 } ] }, { "idx": 9, "layers": [ "model.layers.1.mlp.down_proj" ], "candidates": [ { "dkld": 2.5338865816584843e-05, "dbits": 62914560 }, { "dkld": -5.4409541189681665e-05, "dbits": 125829120 }, { "dkld": -4.8652384430178386e-05, "dbits": 188743680 }, { "dkld": 4.316037520765703e-05, "dbits": 314572800 } ] }, { "idx": 10, "layers": [ "model.layers.2.self_attn.q_proj" ], "candidates": [ { "dkld": 1.0040868073699083e-05, "dbits": 62914560 }, { "dkld": -6.973668932915011e-05, "dbits": 125829120 }, { "dkld": -8.563026785851635e-05, "dbits": 188743680 }, { "dkld": -0.00010046176612377167, "dbits": 314572800 } ] }, { "idx": 11, "layers": [ "model.layers.2.self_attn.k_proj", "model.layers.2.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0006159825250506429, "dbits": 10485760 }, { "dkld": -0.00047235898673535226, "dbits": 20971520 }, { "dkld": -0.0003797250799834756, "dbits": 31457280 }, { "dkld": -0.0004572148434817819, "dbits": 52428800 } ] }, { "idx": 12, "layers": [ "model.layers.2.self_attn.o_proj" ], "candidates": [ { "dkld": -5.700569599867422e-05, "dbits": 62914560 }, { "dkld": 2.468880265950324e-05, "dbits": 125829120 }, { "dkld": 0.0001388624310493386, "dbits": 188743680 }, { "dkld": -0.00011958619579673091, "dbits": 314572800 } ] }, { "idx": 13, "layers": [ "model.layers.2.mlp.gate_proj", "model.layers.2.mlp.up_proj" ], "candidates": [ { "dkld": -8.232537657022754e-05, "dbits": 125829120 }, { "dkld": -0.0005809295922517832, "dbits": 251658240 }, { "dkld": -0.000638202670961624, "dbits": 377487360 }, { "dkld": -0.0006097804754972569, "dbits": 629145600 } ] }, { "idx": 14, "layers": [ "model.layers.2.mlp.down_proj" ], "candidates": [ { "dkld": -0.0006186797283589951, "dbits": 62914560 }, { "dkld": -0.0003680218011140879, "dbits": 125829120 }, { "dkld": -0.0002837544307112694, "dbits": 188743680 }, { "dkld": -0.0003535337746143341, "dbits": 314572800 } ] }, { "idx": 15, "layers": [ "model.layers.3.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002506915479898453, "dbits": 62914560 }, { "dkld": -0.00012434218078852255, "dbits": 125829120 }, { "dkld": -0.00027238717302680016, "dbits": 188743680 }, { "dkld": -0.00017387457191944677, "dbits": 314572800 } ] }, { "idx": 16, "layers": [ "model.layers.3.self_attn.k_proj", "model.layers.3.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0006001546978950528, "dbits": 10485760 }, { "dkld": -0.0007129720412194784, "dbits": 20971520 }, { "dkld": -0.0005572749301791247, "dbits": 31457280 }, { "dkld": -0.000468142889440068, "dbits": 52428800 } ] }, { "idx": 17, "layers": [ "model.layers.3.self_attn.o_proj" ], "candidates": [ { "dkld": -6.0727074742319975e-05, "dbits": 62914560 }, { "dkld": -7.068049162627099e-05, "dbits": 125829120 }, { "dkld": 4.2897649109308045e-06, "dbits": 188743680 }, { "dkld": 3.1332299113270845e-05, "dbits": 314572800 } ] }, { "idx": 18, "layers": [ "model.layers.3.mlp.shared_experts.gate_proj", "model.layers.3.mlp.shared_experts.up_proj", "model.layers.3.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0003118004649877437, "dbits": 23592960 }, { "dkld": 0.00011355206370353144, "dbits": 47185920 }, { "dkld": 0.00016428083181380115, "dbits": 70778880 }, { "dkld": 0.00014832010492682457, "dbits": 117964800 } ] }, { "idx": 19, "layers": [ "model.layers.3.mlp.experts.0.gate_proj", "model.layers.3.mlp.experts.1.gate_proj", "model.layers.3.mlp.experts.2.gate_proj", "model.layers.3.mlp.experts.3.gate_proj", "model.layers.3.mlp.experts.4.gate_proj", "model.layers.3.mlp.experts.5.gate_proj", "model.layers.3.mlp.experts.6.gate_proj", "model.layers.3.mlp.experts.7.gate_proj", "model.layers.3.mlp.experts.8.gate_proj", "model.layers.3.mlp.experts.9.gate_proj", "model.layers.3.mlp.experts.10.gate_proj", "model.layers.3.mlp.experts.11.gate_proj", "model.layers.3.mlp.experts.12.gate_proj", "model.layers.3.mlp.experts.13.gate_proj", "model.layers.3.mlp.experts.14.gate_proj", "model.layers.3.mlp.experts.15.gate_proj", "model.layers.3.mlp.experts.16.gate_proj", "model.layers.3.mlp.experts.17.gate_proj", "model.layers.3.mlp.experts.18.gate_proj", "model.layers.3.mlp.experts.19.gate_proj", "model.layers.3.mlp.experts.20.gate_proj", "model.layers.3.mlp.experts.21.gate_proj", "model.layers.3.mlp.experts.22.gate_proj", "model.layers.3.mlp.experts.23.gate_proj", "model.layers.3.mlp.experts.24.gate_proj", "model.layers.3.mlp.experts.25.gate_proj", "model.layers.3.mlp.experts.26.gate_proj", "model.layers.3.mlp.experts.27.gate_proj", "model.layers.3.mlp.experts.28.gate_proj", "model.layers.3.mlp.experts.29.gate_proj", "model.layers.3.mlp.experts.30.gate_proj", "model.layers.3.mlp.experts.31.gate_proj", "model.layers.3.mlp.experts.32.gate_proj", "model.layers.3.mlp.experts.33.gate_proj", "model.layers.3.mlp.experts.34.gate_proj", "model.layers.3.mlp.experts.35.gate_proj", "model.layers.3.mlp.experts.36.gate_proj", "model.layers.3.mlp.experts.37.gate_proj", "model.layers.3.mlp.experts.38.gate_proj", "model.layers.3.mlp.experts.39.gate_proj", "model.layers.3.mlp.experts.40.gate_proj", "model.layers.3.mlp.experts.41.gate_proj", "model.layers.3.mlp.experts.42.gate_proj", "model.layers.3.mlp.experts.43.gate_proj", "model.layers.3.mlp.experts.44.gate_proj", "model.layers.3.mlp.experts.45.gate_proj", "model.layers.3.mlp.experts.46.gate_proj", "model.layers.3.mlp.experts.47.gate_proj", "model.layers.3.mlp.experts.48.gate_proj", "model.layers.3.mlp.experts.49.gate_proj", "model.layers.3.mlp.experts.50.gate_proj", "model.layers.3.mlp.experts.51.gate_proj", "model.layers.3.mlp.experts.52.gate_proj", "model.layers.3.mlp.experts.53.gate_proj", "model.layers.3.mlp.experts.54.gate_proj", "model.layers.3.mlp.experts.55.gate_proj", "model.layers.3.mlp.experts.56.gate_proj", "model.layers.3.mlp.experts.57.gate_proj", "model.layers.3.mlp.experts.58.gate_proj", "model.layers.3.mlp.experts.59.gate_proj", "model.layers.3.mlp.experts.60.gate_proj", "model.layers.3.mlp.experts.61.gate_proj", "model.layers.3.mlp.experts.62.gate_proj", "model.layers.3.mlp.experts.63.gate_proj", "model.layers.3.mlp.experts.64.gate_proj", "model.layers.3.mlp.experts.65.gate_proj", "model.layers.3.mlp.experts.66.gate_proj", "model.layers.3.mlp.experts.67.gate_proj", "model.layers.3.mlp.experts.68.gate_proj", "model.layers.3.mlp.experts.69.gate_proj", "model.layers.3.mlp.experts.70.gate_proj", "model.layers.3.mlp.experts.71.gate_proj", "model.layers.3.mlp.experts.72.gate_proj", "model.layers.3.mlp.experts.73.gate_proj", "model.layers.3.mlp.experts.74.gate_proj", "model.layers.3.mlp.experts.75.gate_proj", "model.layers.3.mlp.experts.76.gate_proj", "model.layers.3.mlp.experts.77.gate_proj", "model.layers.3.mlp.experts.78.gate_proj", "model.layers.3.mlp.experts.79.gate_proj", "model.layers.3.mlp.experts.80.gate_proj", "model.layers.3.mlp.experts.81.gate_proj", "model.layers.3.mlp.experts.82.gate_proj", "model.layers.3.mlp.experts.83.gate_proj", "model.layers.3.mlp.experts.84.gate_proj", "model.layers.3.mlp.experts.85.gate_proj", "model.layers.3.mlp.experts.86.gate_proj", "model.layers.3.mlp.experts.87.gate_proj", "model.layers.3.mlp.experts.88.gate_proj", "model.layers.3.mlp.experts.89.gate_proj", "model.layers.3.mlp.experts.90.gate_proj", "model.layers.3.mlp.experts.91.gate_proj", "model.layers.3.mlp.experts.92.gate_proj", "model.layers.3.mlp.experts.93.gate_proj", "model.layers.3.mlp.experts.94.gate_proj", "model.layers.3.mlp.experts.95.gate_proj", "model.layers.3.mlp.experts.96.gate_proj", "model.layers.3.mlp.experts.97.gate_proj", "model.layers.3.mlp.experts.98.gate_proj", "model.layers.3.mlp.experts.99.gate_proj", "model.layers.3.mlp.experts.100.gate_proj", "model.layers.3.mlp.experts.101.gate_proj", "model.layers.3.mlp.experts.102.gate_proj", "model.layers.3.mlp.experts.103.gate_proj", "model.layers.3.mlp.experts.104.gate_proj", "model.layers.3.mlp.experts.105.gate_proj", "model.layers.3.mlp.experts.106.gate_proj", "model.layers.3.mlp.experts.107.gate_proj", "model.layers.3.mlp.experts.108.gate_proj", "model.layers.3.mlp.experts.109.gate_proj", "model.layers.3.mlp.experts.110.gate_proj", "model.layers.3.mlp.experts.111.gate_proj", "model.layers.3.mlp.experts.112.gate_proj", "model.layers.3.mlp.experts.113.gate_proj", "model.layers.3.mlp.experts.114.gate_proj", "model.layers.3.mlp.experts.115.gate_proj", "model.layers.3.mlp.experts.116.gate_proj", "model.layers.3.mlp.experts.117.gate_proj", "model.layers.3.mlp.experts.118.gate_proj", "model.layers.3.mlp.experts.119.gate_proj", "model.layers.3.mlp.experts.120.gate_proj", "model.layers.3.mlp.experts.121.gate_proj", "model.layers.3.mlp.experts.122.gate_proj", "model.layers.3.mlp.experts.123.gate_proj", "model.layers.3.mlp.experts.124.gate_proj", "model.layers.3.mlp.experts.125.gate_proj", "model.layers.3.mlp.experts.126.gate_proj", "model.layers.3.mlp.experts.127.gate_proj", "model.layers.3.mlp.experts.128.gate_proj", "model.layers.3.mlp.experts.129.gate_proj", "model.layers.3.mlp.experts.130.gate_proj", "model.layers.3.mlp.experts.131.gate_proj", "model.layers.3.mlp.experts.132.gate_proj", "model.layers.3.mlp.experts.133.gate_proj", "model.layers.3.mlp.experts.134.gate_proj", "model.layers.3.mlp.experts.135.gate_proj", "model.layers.3.mlp.experts.136.gate_proj", "model.layers.3.mlp.experts.137.gate_proj", "model.layers.3.mlp.experts.138.gate_proj", "model.layers.3.mlp.experts.139.gate_proj", "model.layers.3.mlp.experts.140.gate_proj", "model.layers.3.mlp.experts.141.gate_proj", "model.layers.3.mlp.experts.142.gate_proj", "model.layers.3.mlp.experts.143.gate_proj", "model.layers.3.mlp.experts.144.gate_proj", "model.layers.3.mlp.experts.145.gate_proj", "model.layers.3.mlp.experts.146.gate_proj", "model.layers.3.mlp.experts.147.gate_proj", "model.layers.3.mlp.experts.148.gate_proj", "model.layers.3.mlp.experts.149.gate_proj", "model.layers.3.mlp.experts.150.gate_proj", "model.layers.3.mlp.experts.151.gate_proj", "model.layers.3.mlp.experts.152.gate_proj", "model.layers.3.mlp.experts.153.gate_proj", "model.layers.3.mlp.experts.154.gate_proj", "model.layers.3.mlp.experts.155.gate_proj", "model.layers.3.mlp.experts.156.gate_proj", "model.layers.3.mlp.experts.157.gate_proj", "model.layers.3.mlp.experts.158.gate_proj", "model.layers.3.mlp.experts.159.gate_proj", "model.layers.3.mlp.experts.0.up_proj", "model.layers.3.mlp.experts.1.up_proj", "model.layers.3.mlp.experts.2.up_proj", "model.layers.3.mlp.experts.3.up_proj", "model.layers.3.mlp.experts.4.up_proj", "model.layers.3.mlp.experts.5.up_proj", "model.layers.3.mlp.experts.6.up_proj", "model.layers.3.mlp.experts.7.up_proj", "model.layers.3.mlp.experts.8.up_proj", "model.layers.3.mlp.experts.9.up_proj", "model.layers.3.mlp.experts.10.up_proj", "model.layers.3.mlp.experts.11.up_proj", "model.layers.3.mlp.experts.12.up_proj", "model.layers.3.mlp.experts.13.up_proj", "model.layers.3.mlp.experts.14.up_proj", "model.layers.3.mlp.experts.15.up_proj", "model.layers.3.mlp.experts.16.up_proj", "model.layers.3.mlp.experts.17.up_proj", "model.layers.3.mlp.experts.18.up_proj", "model.layers.3.mlp.experts.19.up_proj", "model.layers.3.mlp.experts.20.up_proj", "model.layers.3.mlp.experts.21.up_proj", "model.layers.3.mlp.experts.22.up_proj", "model.layers.3.mlp.experts.23.up_proj", "model.layers.3.mlp.experts.24.up_proj", "model.layers.3.mlp.experts.25.up_proj", "model.layers.3.mlp.experts.26.up_proj", "model.layers.3.mlp.experts.27.up_proj", "model.layers.3.mlp.experts.28.up_proj", "model.layers.3.mlp.experts.29.up_proj", "model.layers.3.mlp.experts.30.up_proj", "model.layers.3.mlp.experts.31.up_proj", "model.layers.3.mlp.experts.32.up_proj", "model.layers.3.mlp.experts.33.up_proj", "model.layers.3.mlp.experts.34.up_proj", "model.layers.3.mlp.experts.35.up_proj", "model.layers.3.mlp.experts.36.up_proj", "model.layers.3.mlp.experts.37.up_proj", "model.layers.3.mlp.experts.38.up_proj", "model.layers.3.mlp.experts.39.up_proj", "model.layers.3.mlp.experts.40.up_proj", "model.layers.3.mlp.experts.41.up_proj", "model.layers.3.mlp.experts.42.up_proj", "model.layers.3.mlp.experts.43.up_proj", "model.layers.3.mlp.experts.44.up_proj", "model.layers.3.mlp.experts.45.up_proj", "model.layers.3.mlp.experts.46.up_proj", "model.layers.3.mlp.experts.47.up_proj", "model.layers.3.mlp.experts.48.up_proj", "model.layers.3.mlp.experts.49.up_proj", "model.layers.3.mlp.experts.50.up_proj", "model.layers.3.mlp.experts.51.up_proj", "model.layers.3.mlp.experts.52.up_proj", "model.layers.3.mlp.experts.53.up_proj", "model.layers.3.mlp.experts.54.up_proj", "model.layers.3.mlp.experts.55.up_proj", "model.layers.3.mlp.experts.56.up_proj", "model.layers.3.mlp.experts.57.up_proj", "model.layers.3.mlp.experts.58.up_proj", "model.layers.3.mlp.experts.59.up_proj", "model.layers.3.mlp.experts.60.up_proj", "model.layers.3.mlp.experts.61.up_proj", "model.layers.3.mlp.experts.62.up_proj", "model.layers.3.mlp.experts.63.up_proj", "model.layers.3.mlp.experts.64.up_proj", "model.layers.3.mlp.experts.65.up_proj", "model.layers.3.mlp.experts.66.up_proj", "model.layers.3.mlp.experts.67.up_proj", "model.layers.3.mlp.experts.68.up_proj", "model.layers.3.mlp.experts.69.up_proj", "model.layers.3.mlp.experts.70.up_proj", "model.layers.3.mlp.experts.71.up_proj", "model.layers.3.mlp.experts.72.up_proj", "model.layers.3.mlp.experts.73.up_proj", "model.layers.3.mlp.experts.74.up_proj", "model.layers.3.mlp.experts.75.up_proj", "model.layers.3.mlp.experts.76.up_proj", "model.layers.3.mlp.experts.77.up_proj", "model.layers.3.mlp.experts.78.up_proj", "model.layers.3.mlp.experts.79.up_proj", "model.layers.3.mlp.experts.80.up_proj", "model.layers.3.mlp.experts.81.up_proj", "model.layers.3.mlp.experts.82.up_proj", "model.layers.3.mlp.experts.83.up_proj", "model.layers.3.mlp.experts.84.up_proj", "model.layers.3.mlp.experts.85.up_proj", "model.layers.3.mlp.experts.86.up_proj", "model.layers.3.mlp.experts.87.up_proj", "model.layers.3.mlp.experts.88.up_proj", "model.layers.3.mlp.experts.89.up_proj", "model.layers.3.mlp.experts.90.up_proj", "model.layers.3.mlp.experts.91.up_proj", "model.layers.3.mlp.experts.92.up_proj", "model.layers.3.mlp.experts.93.up_proj", "model.layers.3.mlp.experts.94.up_proj", "model.layers.3.mlp.experts.95.up_proj", "model.layers.3.mlp.experts.96.up_proj", "model.layers.3.mlp.experts.97.up_proj", "model.layers.3.mlp.experts.98.up_proj", "model.layers.3.mlp.experts.99.up_proj", "model.layers.3.mlp.experts.100.up_proj", "model.layers.3.mlp.experts.101.up_proj", "model.layers.3.mlp.experts.102.up_proj", "model.layers.3.mlp.experts.103.up_proj", "model.layers.3.mlp.experts.104.up_proj", "model.layers.3.mlp.experts.105.up_proj", "model.layers.3.mlp.experts.106.up_proj", "model.layers.3.mlp.experts.107.up_proj", "model.layers.3.mlp.experts.108.up_proj", "model.layers.3.mlp.experts.109.up_proj", "model.layers.3.mlp.experts.110.up_proj", "model.layers.3.mlp.experts.111.up_proj", "model.layers.3.mlp.experts.112.up_proj", "model.layers.3.mlp.experts.113.up_proj", "model.layers.3.mlp.experts.114.up_proj", "model.layers.3.mlp.experts.115.up_proj", "model.layers.3.mlp.experts.116.up_proj", "model.layers.3.mlp.experts.117.up_proj", "model.layers.3.mlp.experts.118.up_proj", "model.layers.3.mlp.experts.119.up_proj", "model.layers.3.mlp.experts.120.up_proj", "model.layers.3.mlp.experts.121.up_proj", "model.layers.3.mlp.experts.122.up_proj", "model.layers.3.mlp.experts.123.up_proj", "model.layers.3.mlp.experts.124.up_proj", "model.layers.3.mlp.experts.125.up_proj", "model.layers.3.mlp.experts.126.up_proj", "model.layers.3.mlp.experts.127.up_proj", "model.layers.3.mlp.experts.128.up_proj", "model.layers.3.mlp.experts.129.up_proj", "model.layers.3.mlp.experts.130.up_proj", "model.layers.3.mlp.experts.131.up_proj", "model.layers.3.mlp.experts.132.up_proj", "model.layers.3.mlp.experts.133.up_proj", "model.layers.3.mlp.experts.134.up_proj", "model.layers.3.mlp.experts.135.up_proj", "model.layers.3.mlp.experts.136.up_proj", "model.layers.3.mlp.experts.137.up_proj", "model.layers.3.mlp.experts.138.up_proj", "model.layers.3.mlp.experts.139.up_proj", "model.layers.3.mlp.experts.140.up_proj", "model.layers.3.mlp.experts.141.up_proj", "model.layers.3.mlp.experts.142.up_proj", "model.layers.3.mlp.experts.143.up_proj", "model.layers.3.mlp.experts.144.up_proj", "model.layers.3.mlp.experts.145.up_proj", "model.layers.3.mlp.experts.146.up_proj", "model.layers.3.mlp.experts.147.up_proj", "model.layers.3.mlp.experts.148.up_proj", "model.layers.3.mlp.experts.149.up_proj", "model.layers.3.mlp.experts.150.up_proj", "model.layers.3.mlp.experts.151.up_proj", "model.layers.3.mlp.experts.152.up_proj", "model.layers.3.mlp.experts.153.up_proj", "model.layers.3.mlp.experts.154.up_proj", "model.layers.3.mlp.experts.155.up_proj", "model.layers.3.mlp.experts.156.up_proj", "model.layers.3.mlp.experts.157.up_proj", "model.layers.3.mlp.experts.158.up_proj", "model.layers.3.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0004227037541568307, "dbits": 2516582400 }, { "dkld": -0.00026817135512829937, "dbits": 5033164800 }, { "dkld": -0.0003032566979527529, "dbits": 7549747200 }, { "dkld": -0.0003220466896891594, "dbits": 12582912000 } ] }, { "idx": 20, "layers": [ "model.layers.3.mlp.experts.0.down_proj", "model.layers.3.mlp.experts.1.down_proj", "model.layers.3.mlp.experts.2.down_proj", "model.layers.3.mlp.experts.3.down_proj", "model.layers.3.mlp.experts.4.down_proj", "model.layers.3.mlp.experts.5.down_proj", "model.layers.3.mlp.experts.6.down_proj", "model.layers.3.mlp.experts.7.down_proj", "model.layers.3.mlp.experts.8.down_proj", "model.layers.3.mlp.experts.9.down_proj", "model.layers.3.mlp.experts.10.down_proj", "model.layers.3.mlp.experts.11.down_proj", "model.layers.3.mlp.experts.12.down_proj", "model.layers.3.mlp.experts.13.down_proj", "model.layers.3.mlp.experts.14.down_proj", "model.layers.3.mlp.experts.15.down_proj", "model.layers.3.mlp.experts.16.down_proj", "model.layers.3.mlp.experts.17.down_proj", "model.layers.3.mlp.experts.18.down_proj", "model.layers.3.mlp.experts.19.down_proj", "model.layers.3.mlp.experts.20.down_proj", "model.layers.3.mlp.experts.21.down_proj", "model.layers.3.mlp.experts.22.down_proj", "model.layers.3.mlp.experts.23.down_proj", "model.layers.3.mlp.experts.24.down_proj", "model.layers.3.mlp.experts.25.down_proj", "model.layers.3.mlp.experts.26.down_proj", "model.layers.3.mlp.experts.27.down_proj", "model.layers.3.mlp.experts.28.down_proj", "model.layers.3.mlp.experts.29.down_proj", "model.layers.3.mlp.experts.30.down_proj", "model.layers.3.mlp.experts.31.down_proj", "model.layers.3.mlp.experts.32.down_proj", "model.layers.3.mlp.experts.33.down_proj", "model.layers.3.mlp.experts.34.down_proj", "model.layers.3.mlp.experts.35.down_proj", "model.layers.3.mlp.experts.36.down_proj", "model.layers.3.mlp.experts.37.down_proj", "model.layers.3.mlp.experts.38.down_proj", "model.layers.3.mlp.experts.39.down_proj", "model.layers.3.mlp.experts.40.down_proj", "model.layers.3.mlp.experts.41.down_proj", "model.layers.3.mlp.experts.42.down_proj", "model.layers.3.mlp.experts.43.down_proj", "model.layers.3.mlp.experts.44.down_proj", "model.layers.3.mlp.experts.45.down_proj", "model.layers.3.mlp.experts.46.down_proj", "model.layers.3.mlp.experts.47.down_proj", "model.layers.3.mlp.experts.48.down_proj", "model.layers.3.mlp.experts.49.down_proj", "model.layers.3.mlp.experts.50.down_proj", "model.layers.3.mlp.experts.51.down_proj", "model.layers.3.mlp.experts.52.down_proj", "model.layers.3.mlp.experts.53.down_proj", "model.layers.3.mlp.experts.54.down_proj", "model.layers.3.mlp.experts.55.down_proj", "model.layers.3.mlp.experts.56.down_proj", "model.layers.3.mlp.experts.57.down_proj", "model.layers.3.mlp.experts.58.down_proj", "model.layers.3.mlp.experts.59.down_proj", "model.layers.3.mlp.experts.60.down_proj", "model.layers.3.mlp.experts.61.down_proj", "model.layers.3.mlp.experts.62.down_proj", "model.layers.3.mlp.experts.63.down_proj", "model.layers.3.mlp.experts.64.down_proj", "model.layers.3.mlp.experts.65.down_proj", "model.layers.3.mlp.experts.66.down_proj", "model.layers.3.mlp.experts.67.down_proj", "model.layers.3.mlp.experts.68.down_proj", "model.layers.3.mlp.experts.69.down_proj", "model.layers.3.mlp.experts.70.down_proj", "model.layers.3.mlp.experts.71.down_proj", "model.layers.3.mlp.experts.72.down_proj", "model.layers.3.mlp.experts.73.down_proj", "model.layers.3.mlp.experts.74.down_proj", "model.layers.3.mlp.experts.75.down_proj", "model.layers.3.mlp.experts.76.down_proj", "model.layers.3.mlp.experts.77.down_proj", "model.layers.3.mlp.experts.78.down_proj", "model.layers.3.mlp.experts.79.down_proj", "model.layers.3.mlp.experts.80.down_proj", "model.layers.3.mlp.experts.81.down_proj", "model.layers.3.mlp.experts.82.down_proj", "model.layers.3.mlp.experts.83.down_proj", "model.layers.3.mlp.experts.84.down_proj", "model.layers.3.mlp.experts.85.down_proj", "model.layers.3.mlp.experts.86.down_proj", "model.layers.3.mlp.experts.87.down_proj", "model.layers.3.mlp.experts.88.down_proj", "model.layers.3.mlp.experts.89.down_proj", "model.layers.3.mlp.experts.90.down_proj", "model.layers.3.mlp.experts.91.down_proj", "model.layers.3.mlp.experts.92.down_proj", "model.layers.3.mlp.experts.93.down_proj", "model.layers.3.mlp.experts.94.down_proj", "model.layers.3.mlp.experts.95.down_proj", "model.layers.3.mlp.experts.96.down_proj", "model.layers.3.mlp.experts.97.down_proj", "model.layers.3.mlp.experts.98.down_proj", "model.layers.3.mlp.experts.99.down_proj", "model.layers.3.mlp.experts.100.down_proj", "model.layers.3.mlp.experts.101.down_proj", "model.layers.3.mlp.experts.102.down_proj", "model.layers.3.mlp.experts.103.down_proj", "model.layers.3.mlp.experts.104.down_proj", "model.layers.3.mlp.experts.105.down_proj", "model.layers.3.mlp.experts.106.down_proj", "model.layers.3.mlp.experts.107.down_proj", "model.layers.3.mlp.experts.108.down_proj", "model.layers.3.mlp.experts.109.down_proj", "model.layers.3.mlp.experts.110.down_proj", "model.layers.3.mlp.experts.111.down_proj", "model.layers.3.mlp.experts.112.down_proj", "model.layers.3.mlp.experts.113.down_proj", "model.layers.3.mlp.experts.114.down_proj", "model.layers.3.mlp.experts.115.down_proj", "model.layers.3.mlp.experts.116.down_proj", "model.layers.3.mlp.experts.117.down_proj", "model.layers.3.mlp.experts.118.down_proj", "model.layers.3.mlp.experts.119.down_proj", "model.layers.3.mlp.experts.120.down_proj", "model.layers.3.mlp.experts.121.down_proj", "model.layers.3.mlp.experts.122.down_proj", "model.layers.3.mlp.experts.123.down_proj", "model.layers.3.mlp.experts.124.down_proj", "model.layers.3.mlp.experts.125.down_proj", "model.layers.3.mlp.experts.126.down_proj", "model.layers.3.mlp.experts.127.down_proj", "model.layers.3.mlp.experts.128.down_proj", "model.layers.3.mlp.experts.129.down_proj", "model.layers.3.mlp.experts.130.down_proj", "model.layers.3.mlp.experts.131.down_proj", "model.layers.3.mlp.experts.132.down_proj", "model.layers.3.mlp.experts.133.down_proj", "model.layers.3.mlp.experts.134.down_proj", "model.layers.3.mlp.experts.135.down_proj", "model.layers.3.mlp.experts.136.down_proj", "model.layers.3.mlp.experts.137.down_proj", "model.layers.3.mlp.experts.138.down_proj", "model.layers.3.mlp.experts.139.down_proj", "model.layers.3.mlp.experts.140.down_proj", "model.layers.3.mlp.experts.141.down_proj", "model.layers.3.mlp.experts.142.down_proj", "model.layers.3.mlp.experts.143.down_proj", "model.layers.3.mlp.experts.144.down_proj", "model.layers.3.mlp.experts.145.down_proj", "model.layers.3.mlp.experts.146.down_proj", "model.layers.3.mlp.experts.147.down_proj", "model.layers.3.mlp.experts.148.down_proj", "model.layers.3.mlp.experts.149.down_proj", "model.layers.3.mlp.experts.150.down_proj", "model.layers.3.mlp.experts.151.down_proj", "model.layers.3.mlp.experts.152.down_proj", "model.layers.3.mlp.experts.153.down_proj", "model.layers.3.mlp.experts.154.down_proj", "model.layers.3.mlp.experts.155.down_proj", "model.layers.3.mlp.experts.156.down_proj", "model.layers.3.mlp.experts.157.down_proj", "model.layers.3.mlp.experts.158.down_proj", "model.layers.3.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0001924077048897771, "dbits": 1258291200 }, { "dkld": -0.00020047966390848715, "dbits": 2516582400 }, { "dkld": -0.00027710422873497564, "dbits": 3774873600 }, { "dkld": -0.00020298194140196124, "dbits": 6291456000 } ] }, { "idx": 21, "layers": [ "model.layers.4.self_attn.q_proj" ], "candidates": [ { "dkld": 5.7903677225107364e-05, "dbits": 62914560 }, { "dkld": 0.00013357829302548807, "dbits": 125829120 }, { "dkld": -3.287214785814563e-05, "dbits": 188743680 }, { "dkld": 0.0001152768731117193, "dbits": 314572800 } ] }, { "idx": 22, "layers": [ "model.layers.4.self_attn.k_proj", "model.layers.4.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00022472143173217496, "dbits": 10485760 }, { "dkld": 0.00011492436751722734, "dbits": 20971520 }, { "dkld": 8.895182982086858e-05, "dbits": 31457280 }, { "dkld": 0.0001406390219926834, "dbits": 52428800 } ] }, { "idx": 23, "layers": [ "model.layers.4.self_attn.o_proj" ], "candidates": [ { "dkld": -1.414828002453683e-05, "dbits": 62914560 }, { "dkld": 3.053434193134308e-05, "dbits": 125829120 }, { "dkld": 0.00014618970453739166, "dbits": 188743680 }, { "dkld": 9.04625281691468e-05, "dbits": 314572800 } ] }, { "idx": 24, "layers": [ "model.layers.4.mlp.shared_experts.gate_proj", "model.layers.4.mlp.shared_experts.up_proj", "model.layers.4.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0002575895749032525, "dbits": 23592960 }, { "dkld": -0.00036260392516851425, "dbits": 47185920 }, { "dkld": -0.00021550171077251712, "dbits": 70778880 }, { "dkld": -0.00040015745908023037, "dbits": 117964800 } ] }, { "idx": 25, "layers": [ "model.layers.4.mlp.experts.0.gate_proj", "model.layers.4.mlp.experts.1.gate_proj", "model.layers.4.mlp.experts.2.gate_proj", "model.layers.4.mlp.experts.3.gate_proj", "model.layers.4.mlp.experts.4.gate_proj", "model.layers.4.mlp.experts.5.gate_proj", "model.layers.4.mlp.experts.6.gate_proj", "model.layers.4.mlp.experts.7.gate_proj", "model.layers.4.mlp.experts.8.gate_proj", "model.layers.4.mlp.experts.9.gate_proj", "model.layers.4.mlp.experts.10.gate_proj", "model.layers.4.mlp.experts.11.gate_proj", "model.layers.4.mlp.experts.12.gate_proj", "model.layers.4.mlp.experts.13.gate_proj", "model.layers.4.mlp.experts.14.gate_proj", "model.layers.4.mlp.experts.15.gate_proj", "model.layers.4.mlp.experts.16.gate_proj", "model.layers.4.mlp.experts.17.gate_proj", "model.layers.4.mlp.experts.18.gate_proj", "model.layers.4.mlp.experts.19.gate_proj", "model.layers.4.mlp.experts.20.gate_proj", "model.layers.4.mlp.experts.21.gate_proj", "model.layers.4.mlp.experts.22.gate_proj", "model.layers.4.mlp.experts.23.gate_proj", "model.layers.4.mlp.experts.24.gate_proj", "model.layers.4.mlp.experts.25.gate_proj", "model.layers.4.mlp.experts.26.gate_proj", "model.layers.4.mlp.experts.27.gate_proj", "model.layers.4.mlp.experts.28.gate_proj", "model.layers.4.mlp.experts.29.gate_proj", "model.layers.4.mlp.experts.30.gate_proj", "model.layers.4.mlp.experts.31.gate_proj", "model.layers.4.mlp.experts.32.gate_proj", "model.layers.4.mlp.experts.33.gate_proj", "model.layers.4.mlp.experts.34.gate_proj", "model.layers.4.mlp.experts.35.gate_proj", "model.layers.4.mlp.experts.36.gate_proj", "model.layers.4.mlp.experts.37.gate_proj", "model.layers.4.mlp.experts.38.gate_proj", "model.layers.4.mlp.experts.39.gate_proj", "model.layers.4.mlp.experts.40.gate_proj", "model.layers.4.mlp.experts.41.gate_proj", "model.layers.4.mlp.experts.42.gate_proj", "model.layers.4.mlp.experts.43.gate_proj", "model.layers.4.mlp.experts.44.gate_proj", "model.layers.4.mlp.experts.45.gate_proj", "model.layers.4.mlp.experts.46.gate_proj", "model.layers.4.mlp.experts.47.gate_proj", "model.layers.4.mlp.experts.48.gate_proj", "model.layers.4.mlp.experts.49.gate_proj", "model.layers.4.mlp.experts.50.gate_proj", "model.layers.4.mlp.experts.51.gate_proj", "model.layers.4.mlp.experts.52.gate_proj", "model.layers.4.mlp.experts.53.gate_proj", "model.layers.4.mlp.experts.54.gate_proj", "model.layers.4.mlp.experts.55.gate_proj", "model.layers.4.mlp.experts.56.gate_proj", "model.layers.4.mlp.experts.57.gate_proj", "model.layers.4.mlp.experts.58.gate_proj", "model.layers.4.mlp.experts.59.gate_proj", "model.layers.4.mlp.experts.60.gate_proj", "model.layers.4.mlp.experts.61.gate_proj", "model.layers.4.mlp.experts.62.gate_proj", "model.layers.4.mlp.experts.63.gate_proj", "model.layers.4.mlp.experts.64.gate_proj", "model.layers.4.mlp.experts.65.gate_proj", "model.layers.4.mlp.experts.66.gate_proj", "model.layers.4.mlp.experts.67.gate_proj", "model.layers.4.mlp.experts.68.gate_proj", "model.layers.4.mlp.experts.69.gate_proj", "model.layers.4.mlp.experts.70.gate_proj", "model.layers.4.mlp.experts.71.gate_proj", "model.layers.4.mlp.experts.72.gate_proj", "model.layers.4.mlp.experts.73.gate_proj", "model.layers.4.mlp.experts.74.gate_proj", "model.layers.4.mlp.experts.75.gate_proj", "model.layers.4.mlp.experts.76.gate_proj", "model.layers.4.mlp.experts.77.gate_proj", "model.layers.4.mlp.experts.78.gate_proj", "model.layers.4.mlp.experts.79.gate_proj", "model.layers.4.mlp.experts.80.gate_proj", "model.layers.4.mlp.experts.81.gate_proj", "model.layers.4.mlp.experts.82.gate_proj", "model.layers.4.mlp.experts.83.gate_proj", "model.layers.4.mlp.experts.84.gate_proj", "model.layers.4.mlp.experts.85.gate_proj", "model.layers.4.mlp.experts.86.gate_proj", "model.layers.4.mlp.experts.87.gate_proj", "model.layers.4.mlp.experts.88.gate_proj", "model.layers.4.mlp.experts.89.gate_proj", "model.layers.4.mlp.experts.90.gate_proj", "model.layers.4.mlp.experts.91.gate_proj", "model.layers.4.mlp.experts.92.gate_proj", "model.layers.4.mlp.experts.93.gate_proj", "model.layers.4.mlp.experts.94.gate_proj", "model.layers.4.mlp.experts.95.gate_proj", "model.layers.4.mlp.experts.96.gate_proj", "model.layers.4.mlp.experts.97.gate_proj", "model.layers.4.mlp.experts.98.gate_proj", "model.layers.4.mlp.experts.99.gate_proj", "model.layers.4.mlp.experts.100.gate_proj", "model.layers.4.mlp.experts.101.gate_proj", "model.layers.4.mlp.experts.102.gate_proj", "model.layers.4.mlp.experts.103.gate_proj", "model.layers.4.mlp.experts.104.gate_proj", "model.layers.4.mlp.experts.105.gate_proj", "model.layers.4.mlp.experts.106.gate_proj", "model.layers.4.mlp.experts.107.gate_proj", "model.layers.4.mlp.experts.108.gate_proj", "model.layers.4.mlp.experts.109.gate_proj", "model.layers.4.mlp.experts.110.gate_proj", "model.layers.4.mlp.experts.111.gate_proj", "model.layers.4.mlp.experts.112.gate_proj", "model.layers.4.mlp.experts.113.gate_proj", "model.layers.4.mlp.experts.114.gate_proj", "model.layers.4.mlp.experts.115.gate_proj", "model.layers.4.mlp.experts.116.gate_proj", "model.layers.4.mlp.experts.117.gate_proj", "model.layers.4.mlp.experts.118.gate_proj", "model.layers.4.mlp.experts.119.gate_proj", "model.layers.4.mlp.experts.120.gate_proj", "model.layers.4.mlp.experts.121.gate_proj", "model.layers.4.mlp.experts.122.gate_proj", "model.layers.4.mlp.experts.123.gate_proj", "model.layers.4.mlp.experts.124.gate_proj", "model.layers.4.mlp.experts.125.gate_proj", "model.layers.4.mlp.experts.126.gate_proj", "model.layers.4.mlp.experts.127.gate_proj", "model.layers.4.mlp.experts.128.gate_proj", "model.layers.4.mlp.experts.129.gate_proj", "model.layers.4.mlp.experts.130.gate_proj", "model.layers.4.mlp.experts.131.gate_proj", "model.layers.4.mlp.experts.132.gate_proj", "model.layers.4.mlp.experts.133.gate_proj", "model.layers.4.mlp.experts.134.gate_proj", "model.layers.4.mlp.experts.135.gate_proj", "model.layers.4.mlp.experts.136.gate_proj", "model.layers.4.mlp.experts.137.gate_proj", "model.layers.4.mlp.experts.138.gate_proj", "model.layers.4.mlp.experts.139.gate_proj", "model.layers.4.mlp.experts.140.gate_proj", "model.layers.4.mlp.experts.141.gate_proj", "model.layers.4.mlp.experts.142.gate_proj", "model.layers.4.mlp.experts.143.gate_proj", "model.layers.4.mlp.experts.144.gate_proj", "model.layers.4.mlp.experts.145.gate_proj", "model.layers.4.mlp.experts.146.gate_proj", "model.layers.4.mlp.experts.147.gate_proj", "model.layers.4.mlp.experts.148.gate_proj", "model.layers.4.mlp.experts.149.gate_proj", "model.layers.4.mlp.experts.150.gate_proj", "model.layers.4.mlp.experts.151.gate_proj", "model.layers.4.mlp.experts.152.gate_proj", "model.layers.4.mlp.experts.153.gate_proj", "model.layers.4.mlp.experts.154.gate_proj", "model.layers.4.mlp.experts.155.gate_proj", "model.layers.4.mlp.experts.156.gate_proj", "model.layers.4.mlp.experts.157.gate_proj", "model.layers.4.mlp.experts.158.gate_proj", "model.layers.4.mlp.experts.159.gate_proj", "model.layers.4.mlp.experts.0.up_proj", "model.layers.4.mlp.experts.1.up_proj", "model.layers.4.mlp.experts.2.up_proj", "model.layers.4.mlp.experts.3.up_proj", "model.layers.4.mlp.experts.4.up_proj", "model.layers.4.mlp.experts.5.up_proj", "model.layers.4.mlp.experts.6.up_proj", "model.layers.4.mlp.experts.7.up_proj", "model.layers.4.mlp.experts.8.up_proj", "model.layers.4.mlp.experts.9.up_proj", "model.layers.4.mlp.experts.10.up_proj", "model.layers.4.mlp.experts.11.up_proj", "model.layers.4.mlp.experts.12.up_proj", "model.layers.4.mlp.experts.13.up_proj", "model.layers.4.mlp.experts.14.up_proj", "model.layers.4.mlp.experts.15.up_proj", "model.layers.4.mlp.experts.16.up_proj", "model.layers.4.mlp.experts.17.up_proj", "model.layers.4.mlp.experts.18.up_proj", "model.layers.4.mlp.experts.19.up_proj", "model.layers.4.mlp.experts.20.up_proj", "model.layers.4.mlp.experts.21.up_proj", "model.layers.4.mlp.experts.22.up_proj", "model.layers.4.mlp.experts.23.up_proj", "model.layers.4.mlp.experts.24.up_proj", "model.layers.4.mlp.experts.25.up_proj", "model.layers.4.mlp.experts.26.up_proj", "model.layers.4.mlp.experts.27.up_proj", "model.layers.4.mlp.experts.28.up_proj", "model.layers.4.mlp.experts.29.up_proj", "model.layers.4.mlp.experts.30.up_proj", "model.layers.4.mlp.experts.31.up_proj", "model.layers.4.mlp.experts.32.up_proj", "model.layers.4.mlp.experts.33.up_proj", "model.layers.4.mlp.experts.34.up_proj", "model.layers.4.mlp.experts.35.up_proj", "model.layers.4.mlp.experts.36.up_proj", "model.layers.4.mlp.experts.37.up_proj", "model.layers.4.mlp.experts.38.up_proj", "model.layers.4.mlp.experts.39.up_proj", "model.layers.4.mlp.experts.40.up_proj", "model.layers.4.mlp.experts.41.up_proj", "model.layers.4.mlp.experts.42.up_proj", "model.layers.4.mlp.experts.43.up_proj", "model.layers.4.mlp.experts.44.up_proj", "model.layers.4.mlp.experts.45.up_proj", "model.layers.4.mlp.experts.46.up_proj", "model.layers.4.mlp.experts.47.up_proj", "model.layers.4.mlp.experts.48.up_proj", "model.layers.4.mlp.experts.49.up_proj", "model.layers.4.mlp.experts.50.up_proj", "model.layers.4.mlp.experts.51.up_proj", "model.layers.4.mlp.experts.52.up_proj", "model.layers.4.mlp.experts.53.up_proj", "model.layers.4.mlp.experts.54.up_proj", "model.layers.4.mlp.experts.55.up_proj", "model.layers.4.mlp.experts.56.up_proj", "model.layers.4.mlp.experts.57.up_proj", "model.layers.4.mlp.experts.58.up_proj", "model.layers.4.mlp.experts.59.up_proj", "model.layers.4.mlp.experts.60.up_proj", "model.layers.4.mlp.experts.61.up_proj", "model.layers.4.mlp.experts.62.up_proj", "model.layers.4.mlp.experts.63.up_proj", "model.layers.4.mlp.experts.64.up_proj", "model.layers.4.mlp.experts.65.up_proj", "model.layers.4.mlp.experts.66.up_proj", "model.layers.4.mlp.experts.67.up_proj", "model.layers.4.mlp.experts.68.up_proj", "model.layers.4.mlp.experts.69.up_proj", "model.layers.4.mlp.experts.70.up_proj", "model.layers.4.mlp.experts.71.up_proj", "model.layers.4.mlp.experts.72.up_proj", "model.layers.4.mlp.experts.73.up_proj", "model.layers.4.mlp.experts.74.up_proj", "model.layers.4.mlp.experts.75.up_proj", "model.layers.4.mlp.experts.76.up_proj", "model.layers.4.mlp.experts.77.up_proj", "model.layers.4.mlp.experts.78.up_proj", "model.layers.4.mlp.experts.79.up_proj", "model.layers.4.mlp.experts.80.up_proj", "model.layers.4.mlp.experts.81.up_proj", "model.layers.4.mlp.experts.82.up_proj", "model.layers.4.mlp.experts.83.up_proj", "model.layers.4.mlp.experts.84.up_proj", "model.layers.4.mlp.experts.85.up_proj", "model.layers.4.mlp.experts.86.up_proj", "model.layers.4.mlp.experts.87.up_proj", "model.layers.4.mlp.experts.88.up_proj", "model.layers.4.mlp.experts.89.up_proj", "model.layers.4.mlp.experts.90.up_proj", "model.layers.4.mlp.experts.91.up_proj", "model.layers.4.mlp.experts.92.up_proj", "model.layers.4.mlp.experts.93.up_proj", "model.layers.4.mlp.experts.94.up_proj", "model.layers.4.mlp.experts.95.up_proj", "model.layers.4.mlp.experts.96.up_proj", "model.layers.4.mlp.experts.97.up_proj", "model.layers.4.mlp.experts.98.up_proj", "model.layers.4.mlp.experts.99.up_proj", "model.layers.4.mlp.experts.100.up_proj", "model.layers.4.mlp.experts.101.up_proj", "model.layers.4.mlp.experts.102.up_proj", "model.layers.4.mlp.experts.103.up_proj", "model.layers.4.mlp.experts.104.up_proj", "model.layers.4.mlp.experts.105.up_proj", "model.layers.4.mlp.experts.106.up_proj", "model.layers.4.mlp.experts.107.up_proj", "model.layers.4.mlp.experts.108.up_proj", "model.layers.4.mlp.experts.109.up_proj", "model.layers.4.mlp.experts.110.up_proj", "model.layers.4.mlp.experts.111.up_proj", "model.layers.4.mlp.experts.112.up_proj", "model.layers.4.mlp.experts.113.up_proj", "model.layers.4.mlp.experts.114.up_proj", "model.layers.4.mlp.experts.115.up_proj", "model.layers.4.mlp.experts.116.up_proj", "model.layers.4.mlp.experts.117.up_proj", "model.layers.4.mlp.experts.118.up_proj", "model.layers.4.mlp.experts.119.up_proj", "model.layers.4.mlp.experts.120.up_proj", "model.layers.4.mlp.experts.121.up_proj", "model.layers.4.mlp.experts.122.up_proj", "model.layers.4.mlp.experts.123.up_proj", "model.layers.4.mlp.experts.124.up_proj", "model.layers.4.mlp.experts.125.up_proj", "model.layers.4.mlp.experts.126.up_proj", "model.layers.4.mlp.experts.127.up_proj", "model.layers.4.mlp.experts.128.up_proj", "model.layers.4.mlp.experts.129.up_proj", "model.layers.4.mlp.experts.130.up_proj", "model.layers.4.mlp.experts.131.up_proj", "model.layers.4.mlp.experts.132.up_proj", "model.layers.4.mlp.experts.133.up_proj", "model.layers.4.mlp.experts.134.up_proj", "model.layers.4.mlp.experts.135.up_proj", "model.layers.4.mlp.experts.136.up_proj", "model.layers.4.mlp.experts.137.up_proj", "model.layers.4.mlp.experts.138.up_proj", "model.layers.4.mlp.experts.139.up_proj", "model.layers.4.mlp.experts.140.up_proj", "model.layers.4.mlp.experts.141.up_proj", "model.layers.4.mlp.experts.142.up_proj", "model.layers.4.mlp.experts.143.up_proj", "model.layers.4.mlp.experts.144.up_proj", "model.layers.4.mlp.experts.145.up_proj", "model.layers.4.mlp.experts.146.up_proj", "model.layers.4.mlp.experts.147.up_proj", "model.layers.4.mlp.experts.148.up_proj", "model.layers.4.mlp.experts.149.up_proj", "model.layers.4.mlp.experts.150.up_proj", "model.layers.4.mlp.experts.151.up_proj", "model.layers.4.mlp.experts.152.up_proj", "model.layers.4.mlp.experts.153.up_proj", "model.layers.4.mlp.experts.154.up_proj", "model.layers.4.mlp.experts.155.up_proj", "model.layers.4.mlp.experts.156.up_proj", "model.layers.4.mlp.experts.157.up_proj", "model.layers.4.mlp.experts.158.up_proj", "model.layers.4.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00017819106578827182, "dbits": 2516582400 }, { "dkld": -0.00030920598655939657, "dbits": 5033164800 }, { "dkld": -0.0004314039833843708, "dbits": 7549747200 }, { "dkld": -0.00032174978405237475, "dbits": 12582912000 } ] }, { "idx": 26, "layers": [ "model.layers.4.mlp.experts.0.down_proj", "model.layers.4.mlp.experts.1.down_proj", "model.layers.4.mlp.experts.2.down_proj", "model.layers.4.mlp.experts.3.down_proj", "model.layers.4.mlp.experts.4.down_proj", "model.layers.4.mlp.experts.5.down_proj", "model.layers.4.mlp.experts.6.down_proj", "model.layers.4.mlp.experts.7.down_proj", "model.layers.4.mlp.experts.8.down_proj", "model.layers.4.mlp.experts.9.down_proj", "model.layers.4.mlp.experts.10.down_proj", "model.layers.4.mlp.experts.11.down_proj", "model.layers.4.mlp.experts.12.down_proj", "model.layers.4.mlp.experts.13.down_proj", "model.layers.4.mlp.experts.14.down_proj", "model.layers.4.mlp.experts.15.down_proj", "model.layers.4.mlp.experts.16.down_proj", "model.layers.4.mlp.experts.17.down_proj", "model.layers.4.mlp.experts.18.down_proj", "model.layers.4.mlp.experts.19.down_proj", "model.layers.4.mlp.experts.20.down_proj", "model.layers.4.mlp.experts.21.down_proj", "model.layers.4.mlp.experts.22.down_proj", "model.layers.4.mlp.experts.23.down_proj", "model.layers.4.mlp.experts.24.down_proj", "model.layers.4.mlp.experts.25.down_proj", "model.layers.4.mlp.experts.26.down_proj", "model.layers.4.mlp.experts.27.down_proj", "model.layers.4.mlp.experts.28.down_proj", "model.layers.4.mlp.experts.29.down_proj", "model.layers.4.mlp.experts.30.down_proj", "model.layers.4.mlp.experts.31.down_proj", "model.layers.4.mlp.experts.32.down_proj", "model.layers.4.mlp.experts.33.down_proj", "model.layers.4.mlp.experts.34.down_proj", "model.layers.4.mlp.experts.35.down_proj", "model.layers.4.mlp.experts.36.down_proj", "model.layers.4.mlp.experts.37.down_proj", "model.layers.4.mlp.experts.38.down_proj", "model.layers.4.mlp.experts.39.down_proj", "model.layers.4.mlp.experts.40.down_proj", "model.layers.4.mlp.experts.41.down_proj", "model.layers.4.mlp.experts.42.down_proj", "model.layers.4.mlp.experts.43.down_proj", "model.layers.4.mlp.experts.44.down_proj", "model.layers.4.mlp.experts.45.down_proj", "model.layers.4.mlp.experts.46.down_proj", "model.layers.4.mlp.experts.47.down_proj", "model.layers.4.mlp.experts.48.down_proj", "model.layers.4.mlp.experts.49.down_proj", "model.layers.4.mlp.experts.50.down_proj", "model.layers.4.mlp.experts.51.down_proj", "model.layers.4.mlp.experts.52.down_proj", "model.layers.4.mlp.experts.53.down_proj", "model.layers.4.mlp.experts.54.down_proj", "model.layers.4.mlp.experts.55.down_proj", "model.layers.4.mlp.experts.56.down_proj", "model.layers.4.mlp.experts.57.down_proj", "model.layers.4.mlp.experts.58.down_proj", "model.layers.4.mlp.experts.59.down_proj", "model.layers.4.mlp.experts.60.down_proj", "model.layers.4.mlp.experts.61.down_proj", "model.layers.4.mlp.experts.62.down_proj", "model.layers.4.mlp.experts.63.down_proj", "model.layers.4.mlp.experts.64.down_proj", "model.layers.4.mlp.experts.65.down_proj", "model.layers.4.mlp.experts.66.down_proj", "model.layers.4.mlp.experts.67.down_proj", "model.layers.4.mlp.experts.68.down_proj", "model.layers.4.mlp.experts.69.down_proj", "model.layers.4.mlp.experts.70.down_proj", "model.layers.4.mlp.experts.71.down_proj", "model.layers.4.mlp.experts.72.down_proj", "model.layers.4.mlp.experts.73.down_proj", "model.layers.4.mlp.experts.74.down_proj", "model.layers.4.mlp.experts.75.down_proj", "model.layers.4.mlp.experts.76.down_proj", "model.layers.4.mlp.experts.77.down_proj", "model.layers.4.mlp.experts.78.down_proj", "model.layers.4.mlp.experts.79.down_proj", "model.layers.4.mlp.experts.80.down_proj", "model.layers.4.mlp.experts.81.down_proj", "model.layers.4.mlp.experts.82.down_proj", "model.layers.4.mlp.experts.83.down_proj", "model.layers.4.mlp.experts.84.down_proj", "model.layers.4.mlp.experts.85.down_proj", "model.layers.4.mlp.experts.86.down_proj", "model.layers.4.mlp.experts.87.down_proj", "model.layers.4.mlp.experts.88.down_proj", "model.layers.4.mlp.experts.89.down_proj", "model.layers.4.mlp.experts.90.down_proj", "model.layers.4.mlp.experts.91.down_proj", "model.layers.4.mlp.experts.92.down_proj", "model.layers.4.mlp.experts.93.down_proj", "model.layers.4.mlp.experts.94.down_proj", "model.layers.4.mlp.experts.95.down_proj", "model.layers.4.mlp.experts.96.down_proj", "model.layers.4.mlp.experts.97.down_proj", "model.layers.4.mlp.experts.98.down_proj", "model.layers.4.mlp.experts.99.down_proj", "model.layers.4.mlp.experts.100.down_proj", "model.layers.4.mlp.experts.101.down_proj", "model.layers.4.mlp.experts.102.down_proj", "model.layers.4.mlp.experts.103.down_proj", "model.layers.4.mlp.experts.104.down_proj", "model.layers.4.mlp.experts.105.down_proj", "model.layers.4.mlp.experts.106.down_proj", "model.layers.4.mlp.experts.107.down_proj", "model.layers.4.mlp.experts.108.down_proj", "model.layers.4.mlp.experts.109.down_proj", "model.layers.4.mlp.experts.110.down_proj", "model.layers.4.mlp.experts.111.down_proj", "model.layers.4.mlp.experts.112.down_proj", "model.layers.4.mlp.experts.113.down_proj", "model.layers.4.mlp.experts.114.down_proj", "model.layers.4.mlp.experts.115.down_proj", "model.layers.4.mlp.experts.116.down_proj", "model.layers.4.mlp.experts.117.down_proj", "model.layers.4.mlp.experts.118.down_proj", "model.layers.4.mlp.experts.119.down_proj", "model.layers.4.mlp.experts.120.down_proj", "model.layers.4.mlp.experts.121.down_proj", "model.layers.4.mlp.experts.122.down_proj", "model.layers.4.mlp.experts.123.down_proj", "model.layers.4.mlp.experts.124.down_proj", "model.layers.4.mlp.experts.125.down_proj", "model.layers.4.mlp.experts.126.down_proj", "model.layers.4.mlp.experts.127.down_proj", "model.layers.4.mlp.experts.128.down_proj", "model.layers.4.mlp.experts.129.down_proj", "model.layers.4.mlp.experts.130.down_proj", "model.layers.4.mlp.experts.131.down_proj", "model.layers.4.mlp.experts.132.down_proj", "model.layers.4.mlp.experts.133.down_proj", "model.layers.4.mlp.experts.134.down_proj", "model.layers.4.mlp.experts.135.down_proj", "model.layers.4.mlp.experts.136.down_proj", "model.layers.4.mlp.experts.137.down_proj", "model.layers.4.mlp.experts.138.down_proj", "model.layers.4.mlp.experts.139.down_proj", "model.layers.4.mlp.experts.140.down_proj", "model.layers.4.mlp.experts.141.down_proj", "model.layers.4.mlp.experts.142.down_proj", "model.layers.4.mlp.experts.143.down_proj", "model.layers.4.mlp.experts.144.down_proj", "model.layers.4.mlp.experts.145.down_proj", "model.layers.4.mlp.experts.146.down_proj", "model.layers.4.mlp.experts.147.down_proj", "model.layers.4.mlp.experts.148.down_proj", "model.layers.4.mlp.experts.149.down_proj", "model.layers.4.mlp.experts.150.down_proj", "model.layers.4.mlp.experts.151.down_proj", "model.layers.4.mlp.experts.152.down_proj", "model.layers.4.mlp.experts.153.down_proj", "model.layers.4.mlp.experts.154.down_proj", "model.layers.4.mlp.experts.155.down_proj", "model.layers.4.mlp.experts.156.down_proj", "model.layers.4.mlp.experts.157.down_proj", "model.layers.4.mlp.experts.158.down_proj", "model.layers.4.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00025744214653969366, "dbits": 1258291200 }, { "dkld": -0.00026075243949891247, "dbits": 2516582400 }, { "dkld": -0.0004022097215056475, "dbits": 3774873600 }, { "dkld": -0.0002971689216792611, "dbits": 6291456000 } ] }, { "idx": 27, "layers": [ "model.layers.5.self_attn.q_proj" ], "candidates": [ { "dkld": -9.347833693028329e-05, "dbits": 62914560 }, { "dkld": -8.468572050333856e-05, "dbits": 125829120 }, { "dkld": -0.00020385617390275557, "dbits": 188743680 }, { "dkld": -0.00014789067208767492, "dbits": 314572800 } ] }, { "idx": 28, "layers": [ "model.layers.5.self_attn.k_proj", "model.layers.5.self_attn.v_proj" ], "candidates": [ { "dkld": 6.976872682570856e-05, "dbits": 10485760 }, { "dkld": 0.00015066899359225672, "dbits": 20971520 }, { "dkld": 6.032912060617568e-05, "dbits": 31457280 }, { "dkld": 6.558205932377659e-05, "dbits": 52428800 } ] }, { "idx": 29, "layers": [ "model.layers.5.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002592211589217269, "dbits": 62914560 }, { "dkld": -0.00012726495042443553, "dbits": 125829120 }, { "dkld": -0.00028245411813260235, "dbits": 188743680 }, { "dkld": -0.0001957170665264213, "dbits": 314572800 } ] }, { "idx": 30, "layers": [ "model.layers.5.mlp.shared_experts.gate_proj", "model.layers.5.mlp.shared_experts.up_proj", "model.layers.5.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00047989217564464726, "dbits": 23592960 }, { "dkld": -0.0003436838276684312, "dbits": 47185920 }, { "dkld": -0.0003431381657719723, "dbits": 70778880 }, { "dkld": -0.0004902926273643998, "dbits": 117964800 } ] }, { "idx": 31, "layers": [ "model.layers.5.mlp.experts.0.gate_proj", "model.layers.5.mlp.experts.1.gate_proj", "model.layers.5.mlp.experts.2.gate_proj", "model.layers.5.mlp.experts.3.gate_proj", "model.layers.5.mlp.experts.4.gate_proj", "model.layers.5.mlp.experts.5.gate_proj", "model.layers.5.mlp.experts.6.gate_proj", "model.layers.5.mlp.experts.7.gate_proj", "model.layers.5.mlp.experts.8.gate_proj", "model.layers.5.mlp.experts.9.gate_proj", "model.layers.5.mlp.experts.10.gate_proj", "model.layers.5.mlp.experts.11.gate_proj", "model.layers.5.mlp.experts.12.gate_proj", "model.layers.5.mlp.experts.13.gate_proj", "model.layers.5.mlp.experts.14.gate_proj", "model.layers.5.mlp.experts.15.gate_proj", "model.layers.5.mlp.experts.16.gate_proj", "model.layers.5.mlp.experts.17.gate_proj", "model.layers.5.mlp.experts.18.gate_proj", "model.layers.5.mlp.experts.19.gate_proj", "model.layers.5.mlp.experts.20.gate_proj", "model.layers.5.mlp.experts.21.gate_proj", "model.layers.5.mlp.experts.22.gate_proj", "model.layers.5.mlp.experts.23.gate_proj", "model.layers.5.mlp.experts.24.gate_proj", "model.layers.5.mlp.experts.25.gate_proj", "model.layers.5.mlp.experts.26.gate_proj", "model.layers.5.mlp.experts.27.gate_proj", "model.layers.5.mlp.experts.28.gate_proj", "model.layers.5.mlp.experts.29.gate_proj", "model.layers.5.mlp.experts.30.gate_proj", "model.layers.5.mlp.experts.31.gate_proj", "model.layers.5.mlp.experts.32.gate_proj", "model.layers.5.mlp.experts.33.gate_proj", "model.layers.5.mlp.experts.34.gate_proj", "model.layers.5.mlp.experts.35.gate_proj", "model.layers.5.mlp.experts.36.gate_proj", "model.layers.5.mlp.experts.37.gate_proj", "model.layers.5.mlp.experts.38.gate_proj", "model.layers.5.mlp.experts.39.gate_proj", "model.layers.5.mlp.experts.40.gate_proj", "model.layers.5.mlp.experts.41.gate_proj", "model.layers.5.mlp.experts.42.gate_proj", "model.layers.5.mlp.experts.43.gate_proj", "model.layers.5.mlp.experts.44.gate_proj", "model.layers.5.mlp.experts.45.gate_proj", "model.layers.5.mlp.experts.46.gate_proj", "model.layers.5.mlp.experts.47.gate_proj", "model.layers.5.mlp.experts.48.gate_proj", "model.layers.5.mlp.experts.49.gate_proj", "model.layers.5.mlp.experts.50.gate_proj", "model.layers.5.mlp.experts.51.gate_proj", "model.layers.5.mlp.experts.52.gate_proj", "model.layers.5.mlp.experts.53.gate_proj", "model.layers.5.mlp.experts.54.gate_proj", "model.layers.5.mlp.experts.55.gate_proj", "model.layers.5.mlp.experts.56.gate_proj", "model.layers.5.mlp.experts.57.gate_proj", "model.layers.5.mlp.experts.58.gate_proj", "model.layers.5.mlp.experts.59.gate_proj", "model.layers.5.mlp.experts.60.gate_proj", "model.layers.5.mlp.experts.61.gate_proj", "model.layers.5.mlp.experts.62.gate_proj", "model.layers.5.mlp.experts.63.gate_proj", "model.layers.5.mlp.experts.64.gate_proj", "model.layers.5.mlp.experts.65.gate_proj", "model.layers.5.mlp.experts.66.gate_proj", "model.layers.5.mlp.experts.67.gate_proj", "model.layers.5.mlp.experts.68.gate_proj", "model.layers.5.mlp.experts.69.gate_proj", "model.layers.5.mlp.experts.70.gate_proj", "model.layers.5.mlp.experts.71.gate_proj", "model.layers.5.mlp.experts.72.gate_proj", "model.layers.5.mlp.experts.73.gate_proj", "model.layers.5.mlp.experts.74.gate_proj", "model.layers.5.mlp.experts.75.gate_proj", "model.layers.5.mlp.experts.76.gate_proj", "model.layers.5.mlp.experts.77.gate_proj", "model.layers.5.mlp.experts.78.gate_proj", "model.layers.5.mlp.experts.79.gate_proj", "model.layers.5.mlp.experts.80.gate_proj", "model.layers.5.mlp.experts.81.gate_proj", "model.layers.5.mlp.experts.82.gate_proj", "model.layers.5.mlp.experts.83.gate_proj", "model.layers.5.mlp.experts.84.gate_proj", "model.layers.5.mlp.experts.85.gate_proj", "model.layers.5.mlp.experts.86.gate_proj", "model.layers.5.mlp.experts.87.gate_proj", "model.layers.5.mlp.experts.88.gate_proj", "model.layers.5.mlp.experts.89.gate_proj", "model.layers.5.mlp.experts.90.gate_proj", "model.layers.5.mlp.experts.91.gate_proj", "model.layers.5.mlp.experts.92.gate_proj", "model.layers.5.mlp.experts.93.gate_proj", "model.layers.5.mlp.experts.94.gate_proj", "model.layers.5.mlp.experts.95.gate_proj", "model.layers.5.mlp.experts.96.gate_proj", "model.layers.5.mlp.experts.97.gate_proj", "model.layers.5.mlp.experts.98.gate_proj", "model.layers.5.mlp.experts.99.gate_proj", "model.layers.5.mlp.experts.100.gate_proj", "model.layers.5.mlp.experts.101.gate_proj", "model.layers.5.mlp.experts.102.gate_proj", "model.layers.5.mlp.experts.103.gate_proj", "model.layers.5.mlp.experts.104.gate_proj", "model.layers.5.mlp.experts.105.gate_proj", "model.layers.5.mlp.experts.106.gate_proj", "model.layers.5.mlp.experts.107.gate_proj", "model.layers.5.mlp.experts.108.gate_proj", "model.layers.5.mlp.experts.109.gate_proj", "model.layers.5.mlp.experts.110.gate_proj", "model.layers.5.mlp.experts.111.gate_proj", "model.layers.5.mlp.experts.112.gate_proj", "model.layers.5.mlp.experts.113.gate_proj", "model.layers.5.mlp.experts.114.gate_proj", "model.layers.5.mlp.experts.115.gate_proj", "model.layers.5.mlp.experts.116.gate_proj", "model.layers.5.mlp.experts.117.gate_proj", "model.layers.5.mlp.experts.118.gate_proj", "model.layers.5.mlp.experts.119.gate_proj", "model.layers.5.mlp.experts.120.gate_proj", "model.layers.5.mlp.experts.121.gate_proj", "model.layers.5.mlp.experts.122.gate_proj", "model.layers.5.mlp.experts.123.gate_proj", "model.layers.5.mlp.experts.124.gate_proj", "model.layers.5.mlp.experts.125.gate_proj", "model.layers.5.mlp.experts.126.gate_proj", "model.layers.5.mlp.experts.127.gate_proj", "model.layers.5.mlp.experts.128.gate_proj", "model.layers.5.mlp.experts.129.gate_proj", "model.layers.5.mlp.experts.130.gate_proj", "model.layers.5.mlp.experts.131.gate_proj", "model.layers.5.mlp.experts.132.gate_proj", "model.layers.5.mlp.experts.133.gate_proj", "model.layers.5.mlp.experts.134.gate_proj", "model.layers.5.mlp.experts.135.gate_proj", "model.layers.5.mlp.experts.136.gate_proj", "model.layers.5.mlp.experts.137.gate_proj", "model.layers.5.mlp.experts.138.gate_proj", "model.layers.5.mlp.experts.139.gate_proj", "model.layers.5.mlp.experts.140.gate_proj", "model.layers.5.mlp.experts.141.gate_proj", "model.layers.5.mlp.experts.142.gate_proj", "model.layers.5.mlp.experts.143.gate_proj", "model.layers.5.mlp.experts.144.gate_proj", "model.layers.5.mlp.experts.145.gate_proj", "model.layers.5.mlp.experts.146.gate_proj", "model.layers.5.mlp.experts.147.gate_proj", "model.layers.5.mlp.experts.148.gate_proj", "model.layers.5.mlp.experts.149.gate_proj", "model.layers.5.mlp.experts.150.gate_proj", "model.layers.5.mlp.experts.151.gate_proj", "model.layers.5.mlp.experts.152.gate_proj", "model.layers.5.mlp.experts.153.gate_proj", "model.layers.5.mlp.experts.154.gate_proj", "model.layers.5.mlp.experts.155.gate_proj", "model.layers.5.mlp.experts.156.gate_proj", "model.layers.5.mlp.experts.157.gate_proj", "model.layers.5.mlp.experts.158.gate_proj", "model.layers.5.mlp.experts.159.gate_proj", "model.layers.5.mlp.experts.0.up_proj", "model.layers.5.mlp.experts.1.up_proj", "model.layers.5.mlp.experts.2.up_proj", "model.layers.5.mlp.experts.3.up_proj", "model.layers.5.mlp.experts.4.up_proj", "model.layers.5.mlp.experts.5.up_proj", "model.layers.5.mlp.experts.6.up_proj", "model.layers.5.mlp.experts.7.up_proj", "model.layers.5.mlp.experts.8.up_proj", "model.layers.5.mlp.experts.9.up_proj", "model.layers.5.mlp.experts.10.up_proj", "model.layers.5.mlp.experts.11.up_proj", "model.layers.5.mlp.experts.12.up_proj", "model.layers.5.mlp.experts.13.up_proj", "model.layers.5.mlp.experts.14.up_proj", "model.layers.5.mlp.experts.15.up_proj", "model.layers.5.mlp.experts.16.up_proj", "model.layers.5.mlp.experts.17.up_proj", "model.layers.5.mlp.experts.18.up_proj", "model.layers.5.mlp.experts.19.up_proj", "model.layers.5.mlp.experts.20.up_proj", "model.layers.5.mlp.experts.21.up_proj", "model.layers.5.mlp.experts.22.up_proj", "model.layers.5.mlp.experts.23.up_proj", "model.layers.5.mlp.experts.24.up_proj", "model.layers.5.mlp.experts.25.up_proj", "model.layers.5.mlp.experts.26.up_proj", "model.layers.5.mlp.experts.27.up_proj", "model.layers.5.mlp.experts.28.up_proj", "model.layers.5.mlp.experts.29.up_proj", "model.layers.5.mlp.experts.30.up_proj", "model.layers.5.mlp.experts.31.up_proj", "model.layers.5.mlp.experts.32.up_proj", "model.layers.5.mlp.experts.33.up_proj", "model.layers.5.mlp.experts.34.up_proj", "model.layers.5.mlp.experts.35.up_proj", "model.layers.5.mlp.experts.36.up_proj", "model.layers.5.mlp.experts.37.up_proj", "model.layers.5.mlp.experts.38.up_proj", "model.layers.5.mlp.experts.39.up_proj", "model.layers.5.mlp.experts.40.up_proj", "model.layers.5.mlp.experts.41.up_proj", "model.layers.5.mlp.experts.42.up_proj", "model.layers.5.mlp.experts.43.up_proj", "model.layers.5.mlp.experts.44.up_proj", "model.layers.5.mlp.experts.45.up_proj", "model.layers.5.mlp.experts.46.up_proj", "model.layers.5.mlp.experts.47.up_proj", "model.layers.5.mlp.experts.48.up_proj", "model.layers.5.mlp.experts.49.up_proj", "model.layers.5.mlp.experts.50.up_proj", "model.layers.5.mlp.experts.51.up_proj", "model.layers.5.mlp.experts.52.up_proj", "model.layers.5.mlp.experts.53.up_proj", "model.layers.5.mlp.experts.54.up_proj", "model.layers.5.mlp.experts.55.up_proj", "model.layers.5.mlp.experts.56.up_proj", "model.layers.5.mlp.experts.57.up_proj", "model.layers.5.mlp.experts.58.up_proj", "model.layers.5.mlp.experts.59.up_proj", "model.layers.5.mlp.experts.60.up_proj", "model.layers.5.mlp.experts.61.up_proj", "model.layers.5.mlp.experts.62.up_proj", "model.layers.5.mlp.experts.63.up_proj", "model.layers.5.mlp.experts.64.up_proj", "model.layers.5.mlp.experts.65.up_proj", "model.layers.5.mlp.experts.66.up_proj", "model.layers.5.mlp.experts.67.up_proj", "model.layers.5.mlp.experts.68.up_proj", "model.layers.5.mlp.experts.69.up_proj", "model.layers.5.mlp.experts.70.up_proj", "model.layers.5.mlp.experts.71.up_proj", "model.layers.5.mlp.experts.72.up_proj", "model.layers.5.mlp.experts.73.up_proj", "model.layers.5.mlp.experts.74.up_proj", "model.layers.5.mlp.experts.75.up_proj", "model.layers.5.mlp.experts.76.up_proj", "model.layers.5.mlp.experts.77.up_proj", "model.layers.5.mlp.experts.78.up_proj", "model.layers.5.mlp.experts.79.up_proj", "model.layers.5.mlp.experts.80.up_proj", "model.layers.5.mlp.experts.81.up_proj", "model.layers.5.mlp.experts.82.up_proj", "model.layers.5.mlp.experts.83.up_proj", "model.layers.5.mlp.experts.84.up_proj", "model.layers.5.mlp.experts.85.up_proj", "model.layers.5.mlp.experts.86.up_proj", "model.layers.5.mlp.experts.87.up_proj", "model.layers.5.mlp.experts.88.up_proj", "model.layers.5.mlp.experts.89.up_proj", "model.layers.5.mlp.experts.90.up_proj", "model.layers.5.mlp.experts.91.up_proj", "model.layers.5.mlp.experts.92.up_proj", "model.layers.5.mlp.experts.93.up_proj", "model.layers.5.mlp.experts.94.up_proj", "model.layers.5.mlp.experts.95.up_proj", "model.layers.5.mlp.experts.96.up_proj", "model.layers.5.mlp.experts.97.up_proj", "model.layers.5.mlp.experts.98.up_proj", "model.layers.5.mlp.experts.99.up_proj", "model.layers.5.mlp.experts.100.up_proj", "model.layers.5.mlp.experts.101.up_proj", "model.layers.5.mlp.experts.102.up_proj", "model.layers.5.mlp.experts.103.up_proj", "model.layers.5.mlp.experts.104.up_proj", "model.layers.5.mlp.experts.105.up_proj", "model.layers.5.mlp.experts.106.up_proj", "model.layers.5.mlp.experts.107.up_proj", "model.layers.5.mlp.experts.108.up_proj", "model.layers.5.mlp.experts.109.up_proj", "model.layers.5.mlp.experts.110.up_proj", "model.layers.5.mlp.experts.111.up_proj", "model.layers.5.mlp.experts.112.up_proj", "model.layers.5.mlp.experts.113.up_proj", "model.layers.5.mlp.experts.114.up_proj", "model.layers.5.mlp.experts.115.up_proj", "model.layers.5.mlp.experts.116.up_proj", "model.layers.5.mlp.experts.117.up_proj", "model.layers.5.mlp.experts.118.up_proj", "model.layers.5.mlp.experts.119.up_proj", "model.layers.5.mlp.experts.120.up_proj", "model.layers.5.mlp.experts.121.up_proj", "model.layers.5.mlp.experts.122.up_proj", "model.layers.5.mlp.experts.123.up_proj", "model.layers.5.mlp.experts.124.up_proj", "model.layers.5.mlp.experts.125.up_proj", "model.layers.5.mlp.experts.126.up_proj", "model.layers.5.mlp.experts.127.up_proj", "model.layers.5.mlp.experts.128.up_proj", "model.layers.5.mlp.experts.129.up_proj", "model.layers.5.mlp.experts.130.up_proj", "model.layers.5.mlp.experts.131.up_proj", "model.layers.5.mlp.experts.132.up_proj", "model.layers.5.mlp.experts.133.up_proj", "model.layers.5.mlp.experts.134.up_proj", "model.layers.5.mlp.experts.135.up_proj", "model.layers.5.mlp.experts.136.up_proj", "model.layers.5.mlp.experts.137.up_proj", "model.layers.5.mlp.experts.138.up_proj", "model.layers.5.mlp.experts.139.up_proj", "model.layers.5.mlp.experts.140.up_proj", "model.layers.5.mlp.experts.141.up_proj", "model.layers.5.mlp.experts.142.up_proj", "model.layers.5.mlp.experts.143.up_proj", "model.layers.5.mlp.experts.144.up_proj", "model.layers.5.mlp.experts.145.up_proj", "model.layers.5.mlp.experts.146.up_proj", "model.layers.5.mlp.experts.147.up_proj", "model.layers.5.mlp.experts.148.up_proj", "model.layers.5.mlp.experts.149.up_proj", "model.layers.5.mlp.experts.150.up_proj", "model.layers.5.mlp.experts.151.up_proj", "model.layers.5.mlp.experts.152.up_proj", "model.layers.5.mlp.experts.153.up_proj", "model.layers.5.mlp.experts.154.up_proj", "model.layers.5.mlp.experts.155.up_proj", "model.layers.5.mlp.experts.156.up_proj", "model.layers.5.mlp.experts.157.up_proj", "model.layers.5.mlp.experts.158.up_proj", "model.layers.5.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00019888048991562046, "dbits": 2516582400 }, { "dkld": -0.00015246672555804808, "dbits": 5033164800 }, { "dkld": -0.00013362094759941656, "dbits": 7549747200 }, { "dkld": -0.00023739729076624472, "dbits": 12582912000 } ] }, { "idx": 32, "layers": [ "model.layers.5.mlp.experts.0.down_proj", "model.layers.5.mlp.experts.1.down_proj", "model.layers.5.mlp.experts.2.down_proj", "model.layers.5.mlp.experts.3.down_proj", "model.layers.5.mlp.experts.4.down_proj", "model.layers.5.mlp.experts.5.down_proj", "model.layers.5.mlp.experts.6.down_proj", "model.layers.5.mlp.experts.7.down_proj", "model.layers.5.mlp.experts.8.down_proj", "model.layers.5.mlp.experts.9.down_proj", "model.layers.5.mlp.experts.10.down_proj", "model.layers.5.mlp.experts.11.down_proj", "model.layers.5.mlp.experts.12.down_proj", "model.layers.5.mlp.experts.13.down_proj", "model.layers.5.mlp.experts.14.down_proj", "model.layers.5.mlp.experts.15.down_proj", "model.layers.5.mlp.experts.16.down_proj", "model.layers.5.mlp.experts.17.down_proj", "model.layers.5.mlp.experts.18.down_proj", "model.layers.5.mlp.experts.19.down_proj", "model.layers.5.mlp.experts.20.down_proj", "model.layers.5.mlp.experts.21.down_proj", "model.layers.5.mlp.experts.22.down_proj", "model.layers.5.mlp.experts.23.down_proj", "model.layers.5.mlp.experts.24.down_proj", "model.layers.5.mlp.experts.25.down_proj", "model.layers.5.mlp.experts.26.down_proj", "model.layers.5.mlp.experts.27.down_proj", "model.layers.5.mlp.experts.28.down_proj", "model.layers.5.mlp.experts.29.down_proj", "model.layers.5.mlp.experts.30.down_proj", "model.layers.5.mlp.experts.31.down_proj", "model.layers.5.mlp.experts.32.down_proj", "model.layers.5.mlp.experts.33.down_proj", "model.layers.5.mlp.experts.34.down_proj", "model.layers.5.mlp.experts.35.down_proj", "model.layers.5.mlp.experts.36.down_proj", "model.layers.5.mlp.experts.37.down_proj", "model.layers.5.mlp.experts.38.down_proj", "model.layers.5.mlp.experts.39.down_proj", "model.layers.5.mlp.experts.40.down_proj", "model.layers.5.mlp.experts.41.down_proj", "model.layers.5.mlp.experts.42.down_proj", "model.layers.5.mlp.experts.43.down_proj", "model.layers.5.mlp.experts.44.down_proj", "model.layers.5.mlp.experts.45.down_proj", "model.layers.5.mlp.experts.46.down_proj", "model.layers.5.mlp.experts.47.down_proj", "model.layers.5.mlp.experts.48.down_proj", "model.layers.5.mlp.experts.49.down_proj", "model.layers.5.mlp.experts.50.down_proj", "model.layers.5.mlp.experts.51.down_proj", "model.layers.5.mlp.experts.52.down_proj", "model.layers.5.mlp.experts.53.down_proj", "model.layers.5.mlp.experts.54.down_proj", "model.layers.5.mlp.experts.55.down_proj", "model.layers.5.mlp.experts.56.down_proj", "model.layers.5.mlp.experts.57.down_proj", "model.layers.5.mlp.experts.58.down_proj", "model.layers.5.mlp.experts.59.down_proj", "model.layers.5.mlp.experts.60.down_proj", "model.layers.5.mlp.experts.61.down_proj", "model.layers.5.mlp.experts.62.down_proj", "model.layers.5.mlp.experts.63.down_proj", "model.layers.5.mlp.experts.64.down_proj", "model.layers.5.mlp.experts.65.down_proj", "model.layers.5.mlp.experts.66.down_proj", "model.layers.5.mlp.experts.67.down_proj", "model.layers.5.mlp.experts.68.down_proj", "model.layers.5.mlp.experts.69.down_proj", "model.layers.5.mlp.experts.70.down_proj", "model.layers.5.mlp.experts.71.down_proj", "model.layers.5.mlp.experts.72.down_proj", "model.layers.5.mlp.experts.73.down_proj", "model.layers.5.mlp.experts.74.down_proj", "model.layers.5.mlp.experts.75.down_proj", "model.layers.5.mlp.experts.76.down_proj", "model.layers.5.mlp.experts.77.down_proj", "model.layers.5.mlp.experts.78.down_proj", "model.layers.5.mlp.experts.79.down_proj", "model.layers.5.mlp.experts.80.down_proj", "model.layers.5.mlp.experts.81.down_proj", "model.layers.5.mlp.experts.82.down_proj", "model.layers.5.mlp.experts.83.down_proj", "model.layers.5.mlp.experts.84.down_proj", "model.layers.5.mlp.experts.85.down_proj", "model.layers.5.mlp.experts.86.down_proj", "model.layers.5.mlp.experts.87.down_proj", "model.layers.5.mlp.experts.88.down_proj", "model.layers.5.mlp.experts.89.down_proj", "model.layers.5.mlp.experts.90.down_proj", "model.layers.5.mlp.experts.91.down_proj", "model.layers.5.mlp.experts.92.down_proj", "model.layers.5.mlp.experts.93.down_proj", "model.layers.5.mlp.experts.94.down_proj", "model.layers.5.mlp.experts.95.down_proj", "model.layers.5.mlp.experts.96.down_proj", "model.layers.5.mlp.experts.97.down_proj", "model.layers.5.mlp.experts.98.down_proj", "model.layers.5.mlp.experts.99.down_proj", "model.layers.5.mlp.experts.100.down_proj", "model.layers.5.mlp.experts.101.down_proj", "model.layers.5.mlp.experts.102.down_proj", "model.layers.5.mlp.experts.103.down_proj", "model.layers.5.mlp.experts.104.down_proj", "model.layers.5.mlp.experts.105.down_proj", "model.layers.5.mlp.experts.106.down_proj", "model.layers.5.mlp.experts.107.down_proj", "model.layers.5.mlp.experts.108.down_proj", "model.layers.5.mlp.experts.109.down_proj", "model.layers.5.mlp.experts.110.down_proj", "model.layers.5.mlp.experts.111.down_proj", "model.layers.5.mlp.experts.112.down_proj", "model.layers.5.mlp.experts.113.down_proj", "model.layers.5.mlp.experts.114.down_proj", "model.layers.5.mlp.experts.115.down_proj", "model.layers.5.mlp.experts.116.down_proj", "model.layers.5.mlp.experts.117.down_proj", "model.layers.5.mlp.experts.118.down_proj", "model.layers.5.mlp.experts.119.down_proj", "model.layers.5.mlp.experts.120.down_proj", "model.layers.5.mlp.experts.121.down_proj", "model.layers.5.mlp.experts.122.down_proj", "model.layers.5.mlp.experts.123.down_proj", "model.layers.5.mlp.experts.124.down_proj", "model.layers.5.mlp.experts.125.down_proj", "model.layers.5.mlp.experts.126.down_proj", "model.layers.5.mlp.experts.127.down_proj", "model.layers.5.mlp.experts.128.down_proj", "model.layers.5.mlp.experts.129.down_proj", "model.layers.5.mlp.experts.130.down_proj", "model.layers.5.mlp.experts.131.down_proj", "model.layers.5.mlp.experts.132.down_proj", "model.layers.5.mlp.experts.133.down_proj", "model.layers.5.mlp.experts.134.down_proj", "model.layers.5.mlp.experts.135.down_proj", "model.layers.5.mlp.experts.136.down_proj", "model.layers.5.mlp.experts.137.down_proj", "model.layers.5.mlp.experts.138.down_proj", "model.layers.5.mlp.experts.139.down_proj", "model.layers.5.mlp.experts.140.down_proj", "model.layers.5.mlp.experts.141.down_proj", "model.layers.5.mlp.experts.142.down_proj", "model.layers.5.mlp.experts.143.down_proj", "model.layers.5.mlp.experts.144.down_proj", "model.layers.5.mlp.experts.145.down_proj", "model.layers.5.mlp.experts.146.down_proj", "model.layers.5.mlp.experts.147.down_proj", "model.layers.5.mlp.experts.148.down_proj", "model.layers.5.mlp.experts.149.down_proj", "model.layers.5.mlp.experts.150.down_proj", "model.layers.5.mlp.experts.151.down_proj", "model.layers.5.mlp.experts.152.down_proj", "model.layers.5.mlp.experts.153.down_proj", "model.layers.5.mlp.experts.154.down_proj", "model.layers.5.mlp.experts.155.down_proj", "model.layers.5.mlp.experts.156.down_proj", "model.layers.5.mlp.experts.157.down_proj", "model.layers.5.mlp.experts.158.down_proj", "model.layers.5.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00021082349121570587, "dbits": 1258291200 }, { "dkld": -0.00024714125320315916, "dbits": 2516582400 }, { "dkld": -0.0002193515188992079, "dbits": 3774873600 }, { "dkld": -0.0003047892823815346, "dbits": 6291456000 } ] }, { "idx": 33, "layers": [ "model.layers.6.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00010098982602357864, "dbits": 62914560 }, { "dkld": 0.00015635071322321614, "dbits": 125829120 }, { "dkld": 0.00016430001705884656, "dbits": 188743680 }, { "dkld": 0.00014759637415408255, "dbits": 314572800 } ] }, { "idx": 34, "layers": [ "model.layers.6.self_attn.k_proj", "model.layers.6.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0001849040389061002, "dbits": 10485760 }, { "dkld": -0.00023479908704758523, "dbits": 20971520 }, { "dkld": -0.0002001577988267056, "dbits": 31457280 }, { "dkld": -0.000123617425560954, "dbits": 52428800 } ] }, { "idx": 35, "layers": [ "model.layers.6.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003033781424164883, "dbits": 62914560 }, { "dkld": -0.0003146762028336525, "dbits": 125829120 }, { "dkld": -0.0003046177327633015, "dbits": 188743680 }, { "dkld": -0.00037820665165783085, "dbits": 314572800 } ] }, { "idx": 36, "layers": [ "model.layers.6.mlp.shared_experts.gate_proj", "model.layers.6.mlp.shared_experts.up_proj", "model.layers.6.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -9.480388835073073e-05, "dbits": 23592960 }, { "dkld": -0.000224141497164973, "dbits": 47185920 }, { "dkld": -0.0003746224567294232, "dbits": 70778880 }, { "dkld": -0.00030741309747100987, "dbits": 117964800 } ] }, { "idx": 37, "layers": [ "model.layers.6.mlp.experts.0.gate_proj", "model.layers.6.mlp.experts.1.gate_proj", "model.layers.6.mlp.experts.2.gate_proj", "model.layers.6.mlp.experts.3.gate_proj", "model.layers.6.mlp.experts.4.gate_proj", "model.layers.6.mlp.experts.5.gate_proj", "model.layers.6.mlp.experts.6.gate_proj", "model.layers.6.mlp.experts.7.gate_proj", "model.layers.6.mlp.experts.8.gate_proj", "model.layers.6.mlp.experts.9.gate_proj", "model.layers.6.mlp.experts.10.gate_proj", "model.layers.6.mlp.experts.11.gate_proj", "model.layers.6.mlp.experts.12.gate_proj", "model.layers.6.mlp.experts.13.gate_proj", "model.layers.6.mlp.experts.14.gate_proj", "model.layers.6.mlp.experts.15.gate_proj", "model.layers.6.mlp.experts.16.gate_proj", "model.layers.6.mlp.experts.17.gate_proj", "model.layers.6.mlp.experts.18.gate_proj", "model.layers.6.mlp.experts.19.gate_proj", "model.layers.6.mlp.experts.20.gate_proj", "model.layers.6.mlp.experts.21.gate_proj", "model.layers.6.mlp.experts.22.gate_proj", "model.layers.6.mlp.experts.23.gate_proj", "model.layers.6.mlp.experts.24.gate_proj", "model.layers.6.mlp.experts.25.gate_proj", "model.layers.6.mlp.experts.26.gate_proj", "model.layers.6.mlp.experts.27.gate_proj", "model.layers.6.mlp.experts.28.gate_proj", "model.layers.6.mlp.experts.29.gate_proj", "model.layers.6.mlp.experts.30.gate_proj", "model.layers.6.mlp.experts.31.gate_proj", "model.layers.6.mlp.experts.32.gate_proj", "model.layers.6.mlp.experts.33.gate_proj", "model.layers.6.mlp.experts.34.gate_proj", "model.layers.6.mlp.experts.35.gate_proj", "model.layers.6.mlp.experts.36.gate_proj", "model.layers.6.mlp.experts.37.gate_proj", "model.layers.6.mlp.experts.38.gate_proj", "model.layers.6.mlp.experts.39.gate_proj", "model.layers.6.mlp.experts.40.gate_proj", "model.layers.6.mlp.experts.41.gate_proj", "model.layers.6.mlp.experts.42.gate_proj", "model.layers.6.mlp.experts.43.gate_proj", "model.layers.6.mlp.experts.44.gate_proj", "model.layers.6.mlp.experts.45.gate_proj", "model.layers.6.mlp.experts.46.gate_proj", "model.layers.6.mlp.experts.47.gate_proj", "model.layers.6.mlp.experts.48.gate_proj", "model.layers.6.mlp.experts.49.gate_proj", "model.layers.6.mlp.experts.50.gate_proj", "model.layers.6.mlp.experts.51.gate_proj", "model.layers.6.mlp.experts.52.gate_proj", "model.layers.6.mlp.experts.53.gate_proj", "model.layers.6.mlp.experts.54.gate_proj", "model.layers.6.mlp.experts.55.gate_proj", "model.layers.6.mlp.experts.56.gate_proj", "model.layers.6.mlp.experts.57.gate_proj", "model.layers.6.mlp.experts.58.gate_proj", "model.layers.6.mlp.experts.59.gate_proj", "model.layers.6.mlp.experts.60.gate_proj", "model.layers.6.mlp.experts.61.gate_proj", "model.layers.6.mlp.experts.62.gate_proj", "model.layers.6.mlp.experts.63.gate_proj", "model.layers.6.mlp.experts.64.gate_proj", "model.layers.6.mlp.experts.65.gate_proj", "model.layers.6.mlp.experts.66.gate_proj", "model.layers.6.mlp.experts.67.gate_proj", "model.layers.6.mlp.experts.68.gate_proj", "model.layers.6.mlp.experts.69.gate_proj", "model.layers.6.mlp.experts.70.gate_proj", "model.layers.6.mlp.experts.71.gate_proj", "model.layers.6.mlp.experts.72.gate_proj", "model.layers.6.mlp.experts.73.gate_proj", "model.layers.6.mlp.experts.74.gate_proj", "model.layers.6.mlp.experts.75.gate_proj", "model.layers.6.mlp.experts.76.gate_proj", "model.layers.6.mlp.experts.77.gate_proj", "model.layers.6.mlp.experts.78.gate_proj", "model.layers.6.mlp.experts.79.gate_proj", "model.layers.6.mlp.experts.80.gate_proj", "model.layers.6.mlp.experts.81.gate_proj", "model.layers.6.mlp.experts.82.gate_proj", "model.layers.6.mlp.experts.83.gate_proj", "model.layers.6.mlp.experts.84.gate_proj", "model.layers.6.mlp.experts.85.gate_proj", "model.layers.6.mlp.experts.86.gate_proj", "model.layers.6.mlp.experts.87.gate_proj", "model.layers.6.mlp.experts.88.gate_proj", "model.layers.6.mlp.experts.89.gate_proj", "model.layers.6.mlp.experts.90.gate_proj", "model.layers.6.mlp.experts.91.gate_proj", "model.layers.6.mlp.experts.92.gate_proj", "model.layers.6.mlp.experts.93.gate_proj", "model.layers.6.mlp.experts.94.gate_proj", "model.layers.6.mlp.experts.95.gate_proj", "model.layers.6.mlp.experts.96.gate_proj", "model.layers.6.mlp.experts.97.gate_proj", "model.layers.6.mlp.experts.98.gate_proj", "model.layers.6.mlp.experts.99.gate_proj", "model.layers.6.mlp.experts.100.gate_proj", "model.layers.6.mlp.experts.101.gate_proj", "model.layers.6.mlp.experts.102.gate_proj", "model.layers.6.mlp.experts.103.gate_proj", "model.layers.6.mlp.experts.104.gate_proj", "model.layers.6.mlp.experts.105.gate_proj", "model.layers.6.mlp.experts.106.gate_proj", "model.layers.6.mlp.experts.107.gate_proj", "model.layers.6.mlp.experts.108.gate_proj", "model.layers.6.mlp.experts.109.gate_proj", "model.layers.6.mlp.experts.110.gate_proj", "model.layers.6.mlp.experts.111.gate_proj", "model.layers.6.mlp.experts.112.gate_proj", "model.layers.6.mlp.experts.113.gate_proj", "model.layers.6.mlp.experts.114.gate_proj", "model.layers.6.mlp.experts.115.gate_proj", "model.layers.6.mlp.experts.116.gate_proj", "model.layers.6.mlp.experts.117.gate_proj", "model.layers.6.mlp.experts.118.gate_proj", "model.layers.6.mlp.experts.119.gate_proj", "model.layers.6.mlp.experts.120.gate_proj", "model.layers.6.mlp.experts.121.gate_proj", "model.layers.6.mlp.experts.122.gate_proj", "model.layers.6.mlp.experts.123.gate_proj", "model.layers.6.mlp.experts.124.gate_proj", "model.layers.6.mlp.experts.125.gate_proj", "model.layers.6.mlp.experts.126.gate_proj", "model.layers.6.mlp.experts.127.gate_proj", "model.layers.6.mlp.experts.128.gate_proj", "model.layers.6.mlp.experts.129.gate_proj", "model.layers.6.mlp.experts.130.gate_proj", "model.layers.6.mlp.experts.131.gate_proj", "model.layers.6.mlp.experts.132.gate_proj", "model.layers.6.mlp.experts.133.gate_proj", "model.layers.6.mlp.experts.134.gate_proj", "model.layers.6.mlp.experts.135.gate_proj", "model.layers.6.mlp.experts.136.gate_proj", "model.layers.6.mlp.experts.137.gate_proj", "model.layers.6.mlp.experts.138.gate_proj", "model.layers.6.mlp.experts.139.gate_proj", "model.layers.6.mlp.experts.140.gate_proj", "model.layers.6.mlp.experts.141.gate_proj", "model.layers.6.mlp.experts.142.gate_proj", "model.layers.6.mlp.experts.143.gate_proj", "model.layers.6.mlp.experts.144.gate_proj", "model.layers.6.mlp.experts.145.gate_proj", "model.layers.6.mlp.experts.146.gate_proj", "model.layers.6.mlp.experts.147.gate_proj", "model.layers.6.mlp.experts.148.gate_proj", "model.layers.6.mlp.experts.149.gate_proj", "model.layers.6.mlp.experts.150.gate_proj", "model.layers.6.mlp.experts.151.gate_proj", "model.layers.6.mlp.experts.152.gate_proj", "model.layers.6.mlp.experts.153.gate_proj", "model.layers.6.mlp.experts.154.gate_proj", "model.layers.6.mlp.experts.155.gate_proj", "model.layers.6.mlp.experts.156.gate_proj", "model.layers.6.mlp.experts.157.gate_proj", "model.layers.6.mlp.experts.158.gate_proj", "model.layers.6.mlp.experts.159.gate_proj", "model.layers.6.mlp.experts.0.up_proj", "model.layers.6.mlp.experts.1.up_proj", "model.layers.6.mlp.experts.2.up_proj", "model.layers.6.mlp.experts.3.up_proj", "model.layers.6.mlp.experts.4.up_proj", "model.layers.6.mlp.experts.5.up_proj", "model.layers.6.mlp.experts.6.up_proj", "model.layers.6.mlp.experts.7.up_proj", "model.layers.6.mlp.experts.8.up_proj", "model.layers.6.mlp.experts.9.up_proj", "model.layers.6.mlp.experts.10.up_proj", "model.layers.6.mlp.experts.11.up_proj", "model.layers.6.mlp.experts.12.up_proj", "model.layers.6.mlp.experts.13.up_proj", "model.layers.6.mlp.experts.14.up_proj", "model.layers.6.mlp.experts.15.up_proj", "model.layers.6.mlp.experts.16.up_proj", "model.layers.6.mlp.experts.17.up_proj", "model.layers.6.mlp.experts.18.up_proj", "model.layers.6.mlp.experts.19.up_proj", "model.layers.6.mlp.experts.20.up_proj", "model.layers.6.mlp.experts.21.up_proj", "model.layers.6.mlp.experts.22.up_proj", "model.layers.6.mlp.experts.23.up_proj", "model.layers.6.mlp.experts.24.up_proj", "model.layers.6.mlp.experts.25.up_proj", "model.layers.6.mlp.experts.26.up_proj", "model.layers.6.mlp.experts.27.up_proj", "model.layers.6.mlp.experts.28.up_proj", "model.layers.6.mlp.experts.29.up_proj", "model.layers.6.mlp.experts.30.up_proj", "model.layers.6.mlp.experts.31.up_proj", "model.layers.6.mlp.experts.32.up_proj", "model.layers.6.mlp.experts.33.up_proj", "model.layers.6.mlp.experts.34.up_proj", "model.layers.6.mlp.experts.35.up_proj", "model.layers.6.mlp.experts.36.up_proj", "model.layers.6.mlp.experts.37.up_proj", "model.layers.6.mlp.experts.38.up_proj", "model.layers.6.mlp.experts.39.up_proj", "model.layers.6.mlp.experts.40.up_proj", "model.layers.6.mlp.experts.41.up_proj", "model.layers.6.mlp.experts.42.up_proj", "model.layers.6.mlp.experts.43.up_proj", "model.layers.6.mlp.experts.44.up_proj", "model.layers.6.mlp.experts.45.up_proj", "model.layers.6.mlp.experts.46.up_proj", "model.layers.6.mlp.experts.47.up_proj", "model.layers.6.mlp.experts.48.up_proj", "model.layers.6.mlp.experts.49.up_proj", "model.layers.6.mlp.experts.50.up_proj", "model.layers.6.mlp.experts.51.up_proj", "model.layers.6.mlp.experts.52.up_proj", "model.layers.6.mlp.experts.53.up_proj", "model.layers.6.mlp.experts.54.up_proj", "model.layers.6.mlp.experts.55.up_proj", "model.layers.6.mlp.experts.56.up_proj", "model.layers.6.mlp.experts.57.up_proj", "model.layers.6.mlp.experts.58.up_proj", "model.layers.6.mlp.experts.59.up_proj", "model.layers.6.mlp.experts.60.up_proj", "model.layers.6.mlp.experts.61.up_proj", "model.layers.6.mlp.experts.62.up_proj", "model.layers.6.mlp.experts.63.up_proj", "model.layers.6.mlp.experts.64.up_proj", "model.layers.6.mlp.experts.65.up_proj", "model.layers.6.mlp.experts.66.up_proj", "model.layers.6.mlp.experts.67.up_proj", "model.layers.6.mlp.experts.68.up_proj", "model.layers.6.mlp.experts.69.up_proj", "model.layers.6.mlp.experts.70.up_proj", "model.layers.6.mlp.experts.71.up_proj", "model.layers.6.mlp.experts.72.up_proj", "model.layers.6.mlp.experts.73.up_proj", "model.layers.6.mlp.experts.74.up_proj", "model.layers.6.mlp.experts.75.up_proj", "model.layers.6.mlp.experts.76.up_proj", "model.layers.6.mlp.experts.77.up_proj", "model.layers.6.mlp.experts.78.up_proj", "model.layers.6.mlp.experts.79.up_proj", "model.layers.6.mlp.experts.80.up_proj", "model.layers.6.mlp.experts.81.up_proj", "model.layers.6.mlp.experts.82.up_proj", "model.layers.6.mlp.experts.83.up_proj", "model.layers.6.mlp.experts.84.up_proj", "model.layers.6.mlp.experts.85.up_proj", "model.layers.6.mlp.experts.86.up_proj", "model.layers.6.mlp.experts.87.up_proj", "model.layers.6.mlp.experts.88.up_proj", "model.layers.6.mlp.experts.89.up_proj", "model.layers.6.mlp.experts.90.up_proj", "model.layers.6.mlp.experts.91.up_proj", "model.layers.6.mlp.experts.92.up_proj", "model.layers.6.mlp.experts.93.up_proj", "model.layers.6.mlp.experts.94.up_proj", "model.layers.6.mlp.experts.95.up_proj", "model.layers.6.mlp.experts.96.up_proj", "model.layers.6.mlp.experts.97.up_proj", "model.layers.6.mlp.experts.98.up_proj", "model.layers.6.mlp.experts.99.up_proj", "model.layers.6.mlp.experts.100.up_proj", "model.layers.6.mlp.experts.101.up_proj", "model.layers.6.mlp.experts.102.up_proj", "model.layers.6.mlp.experts.103.up_proj", "model.layers.6.mlp.experts.104.up_proj", "model.layers.6.mlp.experts.105.up_proj", "model.layers.6.mlp.experts.106.up_proj", "model.layers.6.mlp.experts.107.up_proj", "model.layers.6.mlp.experts.108.up_proj", "model.layers.6.mlp.experts.109.up_proj", "model.layers.6.mlp.experts.110.up_proj", "model.layers.6.mlp.experts.111.up_proj", "model.layers.6.mlp.experts.112.up_proj", "model.layers.6.mlp.experts.113.up_proj", "model.layers.6.mlp.experts.114.up_proj", "model.layers.6.mlp.experts.115.up_proj", "model.layers.6.mlp.experts.116.up_proj", "model.layers.6.mlp.experts.117.up_proj", "model.layers.6.mlp.experts.118.up_proj", "model.layers.6.mlp.experts.119.up_proj", "model.layers.6.mlp.experts.120.up_proj", "model.layers.6.mlp.experts.121.up_proj", "model.layers.6.mlp.experts.122.up_proj", "model.layers.6.mlp.experts.123.up_proj", "model.layers.6.mlp.experts.124.up_proj", "model.layers.6.mlp.experts.125.up_proj", "model.layers.6.mlp.experts.126.up_proj", "model.layers.6.mlp.experts.127.up_proj", "model.layers.6.mlp.experts.128.up_proj", "model.layers.6.mlp.experts.129.up_proj", "model.layers.6.mlp.experts.130.up_proj", "model.layers.6.mlp.experts.131.up_proj", "model.layers.6.mlp.experts.132.up_proj", "model.layers.6.mlp.experts.133.up_proj", "model.layers.6.mlp.experts.134.up_proj", "model.layers.6.mlp.experts.135.up_proj", "model.layers.6.mlp.experts.136.up_proj", "model.layers.6.mlp.experts.137.up_proj", "model.layers.6.mlp.experts.138.up_proj", "model.layers.6.mlp.experts.139.up_proj", "model.layers.6.mlp.experts.140.up_proj", "model.layers.6.mlp.experts.141.up_proj", "model.layers.6.mlp.experts.142.up_proj", "model.layers.6.mlp.experts.143.up_proj", "model.layers.6.mlp.experts.144.up_proj", "model.layers.6.mlp.experts.145.up_proj", "model.layers.6.mlp.experts.146.up_proj", "model.layers.6.mlp.experts.147.up_proj", "model.layers.6.mlp.experts.148.up_proj", "model.layers.6.mlp.experts.149.up_proj", "model.layers.6.mlp.experts.150.up_proj", "model.layers.6.mlp.experts.151.up_proj", "model.layers.6.mlp.experts.152.up_proj", "model.layers.6.mlp.experts.153.up_proj", "model.layers.6.mlp.experts.154.up_proj", "model.layers.6.mlp.experts.155.up_proj", "model.layers.6.mlp.experts.156.up_proj", "model.layers.6.mlp.experts.157.up_proj", "model.layers.6.mlp.experts.158.up_proj", "model.layers.6.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00027203857898712436, "dbits": 2516582400 }, { "dkld": -0.00024212282150984366, "dbits": 5033164800 }, { "dkld": -0.00036374013870954514, "dbits": 7549747200 }, { "dkld": -0.00022652009502054093, "dbits": 12582912000 } ] }, { "idx": 38, "layers": [ "model.layers.6.mlp.experts.0.down_proj", "model.layers.6.mlp.experts.1.down_proj", "model.layers.6.mlp.experts.2.down_proj", "model.layers.6.mlp.experts.3.down_proj", "model.layers.6.mlp.experts.4.down_proj", "model.layers.6.mlp.experts.5.down_proj", "model.layers.6.mlp.experts.6.down_proj", "model.layers.6.mlp.experts.7.down_proj", "model.layers.6.mlp.experts.8.down_proj", "model.layers.6.mlp.experts.9.down_proj", "model.layers.6.mlp.experts.10.down_proj", "model.layers.6.mlp.experts.11.down_proj", "model.layers.6.mlp.experts.12.down_proj", "model.layers.6.mlp.experts.13.down_proj", "model.layers.6.mlp.experts.14.down_proj", "model.layers.6.mlp.experts.15.down_proj", "model.layers.6.mlp.experts.16.down_proj", "model.layers.6.mlp.experts.17.down_proj", "model.layers.6.mlp.experts.18.down_proj", "model.layers.6.mlp.experts.19.down_proj", "model.layers.6.mlp.experts.20.down_proj", "model.layers.6.mlp.experts.21.down_proj", "model.layers.6.mlp.experts.22.down_proj", "model.layers.6.mlp.experts.23.down_proj", "model.layers.6.mlp.experts.24.down_proj", "model.layers.6.mlp.experts.25.down_proj", "model.layers.6.mlp.experts.26.down_proj", "model.layers.6.mlp.experts.27.down_proj", "model.layers.6.mlp.experts.28.down_proj", "model.layers.6.mlp.experts.29.down_proj", "model.layers.6.mlp.experts.30.down_proj", "model.layers.6.mlp.experts.31.down_proj", "model.layers.6.mlp.experts.32.down_proj", "model.layers.6.mlp.experts.33.down_proj", "model.layers.6.mlp.experts.34.down_proj", "model.layers.6.mlp.experts.35.down_proj", "model.layers.6.mlp.experts.36.down_proj", "model.layers.6.mlp.experts.37.down_proj", "model.layers.6.mlp.experts.38.down_proj", "model.layers.6.mlp.experts.39.down_proj", "model.layers.6.mlp.experts.40.down_proj", "model.layers.6.mlp.experts.41.down_proj", "model.layers.6.mlp.experts.42.down_proj", "model.layers.6.mlp.experts.43.down_proj", "model.layers.6.mlp.experts.44.down_proj", "model.layers.6.mlp.experts.45.down_proj", "model.layers.6.mlp.experts.46.down_proj", "model.layers.6.mlp.experts.47.down_proj", "model.layers.6.mlp.experts.48.down_proj", "model.layers.6.mlp.experts.49.down_proj", "model.layers.6.mlp.experts.50.down_proj", "model.layers.6.mlp.experts.51.down_proj", "model.layers.6.mlp.experts.52.down_proj", "model.layers.6.mlp.experts.53.down_proj", "model.layers.6.mlp.experts.54.down_proj", "model.layers.6.mlp.experts.55.down_proj", "model.layers.6.mlp.experts.56.down_proj", "model.layers.6.mlp.experts.57.down_proj", "model.layers.6.mlp.experts.58.down_proj", "model.layers.6.mlp.experts.59.down_proj", "model.layers.6.mlp.experts.60.down_proj", "model.layers.6.mlp.experts.61.down_proj", "model.layers.6.mlp.experts.62.down_proj", "model.layers.6.mlp.experts.63.down_proj", "model.layers.6.mlp.experts.64.down_proj", "model.layers.6.mlp.experts.65.down_proj", "model.layers.6.mlp.experts.66.down_proj", "model.layers.6.mlp.experts.67.down_proj", "model.layers.6.mlp.experts.68.down_proj", "model.layers.6.mlp.experts.69.down_proj", "model.layers.6.mlp.experts.70.down_proj", "model.layers.6.mlp.experts.71.down_proj", "model.layers.6.mlp.experts.72.down_proj", "model.layers.6.mlp.experts.73.down_proj", "model.layers.6.mlp.experts.74.down_proj", "model.layers.6.mlp.experts.75.down_proj", "model.layers.6.mlp.experts.76.down_proj", "model.layers.6.mlp.experts.77.down_proj", "model.layers.6.mlp.experts.78.down_proj", "model.layers.6.mlp.experts.79.down_proj", "model.layers.6.mlp.experts.80.down_proj", "model.layers.6.mlp.experts.81.down_proj", "model.layers.6.mlp.experts.82.down_proj", "model.layers.6.mlp.experts.83.down_proj", "model.layers.6.mlp.experts.84.down_proj", "model.layers.6.mlp.experts.85.down_proj", "model.layers.6.mlp.experts.86.down_proj", "model.layers.6.mlp.experts.87.down_proj", "model.layers.6.mlp.experts.88.down_proj", "model.layers.6.mlp.experts.89.down_proj", "model.layers.6.mlp.experts.90.down_proj", "model.layers.6.mlp.experts.91.down_proj", "model.layers.6.mlp.experts.92.down_proj", "model.layers.6.mlp.experts.93.down_proj", "model.layers.6.mlp.experts.94.down_proj", "model.layers.6.mlp.experts.95.down_proj", "model.layers.6.mlp.experts.96.down_proj", "model.layers.6.mlp.experts.97.down_proj", "model.layers.6.mlp.experts.98.down_proj", "model.layers.6.mlp.experts.99.down_proj", "model.layers.6.mlp.experts.100.down_proj", "model.layers.6.mlp.experts.101.down_proj", "model.layers.6.mlp.experts.102.down_proj", "model.layers.6.mlp.experts.103.down_proj", "model.layers.6.mlp.experts.104.down_proj", "model.layers.6.mlp.experts.105.down_proj", "model.layers.6.mlp.experts.106.down_proj", "model.layers.6.mlp.experts.107.down_proj", "model.layers.6.mlp.experts.108.down_proj", "model.layers.6.mlp.experts.109.down_proj", "model.layers.6.mlp.experts.110.down_proj", "model.layers.6.mlp.experts.111.down_proj", "model.layers.6.mlp.experts.112.down_proj", "model.layers.6.mlp.experts.113.down_proj", "model.layers.6.mlp.experts.114.down_proj", "model.layers.6.mlp.experts.115.down_proj", "model.layers.6.mlp.experts.116.down_proj", "model.layers.6.mlp.experts.117.down_proj", "model.layers.6.mlp.experts.118.down_proj", "model.layers.6.mlp.experts.119.down_proj", "model.layers.6.mlp.experts.120.down_proj", "model.layers.6.mlp.experts.121.down_proj", "model.layers.6.mlp.experts.122.down_proj", "model.layers.6.mlp.experts.123.down_proj", "model.layers.6.mlp.experts.124.down_proj", "model.layers.6.mlp.experts.125.down_proj", "model.layers.6.mlp.experts.126.down_proj", "model.layers.6.mlp.experts.127.down_proj", "model.layers.6.mlp.experts.128.down_proj", "model.layers.6.mlp.experts.129.down_proj", "model.layers.6.mlp.experts.130.down_proj", "model.layers.6.mlp.experts.131.down_proj", "model.layers.6.mlp.experts.132.down_proj", "model.layers.6.mlp.experts.133.down_proj", "model.layers.6.mlp.experts.134.down_proj", "model.layers.6.mlp.experts.135.down_proj", "model.layers.6.mlp.experts.136.down_proj", "model.layers.6.mlp.experts.137.down_proj", "model.layers.6.mlp.experts.138.down_proj", "model.layers.6.mlp.experts.139.down_proj", "model.layers.6.mlp.experts.140.down_proj", "model.layers.6.mlp.experts.141.down_proj", "model.layers.6.mlp.experts.142.down_proj", "model.layers.6.mlp.experts.143.down_proj", "model.layers.6.mlp.experts.144.down_proj", "model.layers.6.mlp.experts.145.down_proj", "model.layers.6.mlp.experts.146.down_proj", "model.layers.6.mlp.experts.147.down_proj", "model.layers.6.mlp.experts.148.down_proj", "model.layers.6.mlp.experts.149.down_proj", "model.layers.6.mlp.experts.150.down_proj", "model.layers.6.mlp.experts.151.down_proj", "model.layers.6.mlp.experts.152.down_proj", "model.layers.6.mlp.experts.153.down_proj", "model.layers.6.mlp.experts.154.down_proj", "model.layers.6.mlp.experts.155.down_proj", "model.layers.6.mlp.experts.156.down_proj", "model.layers.6.mlp.experts.157.down_proj", "model.layers.6.mlp.experts.158.down_proj", "model.layers.6.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00012872442603112377, "dbits": 1258291200 }, { "dkld": -0.00042999340221286375, "dbits": 2516582400 }, { "dkld": -0.0003457723185420064, "dbits": 3774873600 }, { "dkld": -0.000345354899764061, "dbits": 6291456000 } ] }, { "idx": 39, "layers": [ "model.layers.7.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002893224358558738, "dbits": 62914560 }, { "dkld": -0.00034349504858255386, "dbits": 125829120 }, { "dkld": -0.00031540114432573874, "dbits": 188743680 }, { "dkld": -0.0003008143976330757, "dbits": 314572800 } ] }, { "idx": 40, "layers": [ "model.layers.7.self_attn.k_proj", "model.layers.7.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00027521448209882893, "dbits": 10485760 }, { "dkld": -4.370855167508403e-05, "dbits": 20971520 }, { "dkld": -0.00012599024921655655, "dbits": 31457280 }, { "dkld": -0.00014449851587415574, "dbits": 52428800 } ] }, { "idx": 41, "layers": [ "model.layers.7.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00010888082906604368, "dbits": 62914560 }, { "dkld": -0.00023113824427128948, "dbits": 125829120 }, { "dkld": -0.00024301474913955567, "dbits": 188743680 }, { "dkld": -0.00031124372035265524, "dbits": 314572800 } ] }, { "idx": 42, "layers": [ "model.layers.7.mlp.shared_experts.gate_proj", "model.layers.7.mlp.shared_experts.up_proj", "model.layers.7.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 9.852759540079914e-05, "dbits": 23592960 }, { "dkld": 0.0004529677331447518, "dbits": 47185920 }, { "dkld": 0.0002845930866897106, "dbits": 70778880 }, { "dkld": 0.0003181221894919817, "dbits": 117964800 } ] }, { "idx": 43, "layers": [ "model.layers.7.mlp.experts.0.gate_proj", "model.layers.7.mlp.experts.1.gate_proj", "model.layers.7.mlp.experts.2.gate_proj", "model.layers.7.mlp.experts.3.gate_proj", "model.layers.7.mlp.experts.4.gate_proj", "model.layers.7.mlp.experts.5.gate_proj", "model.layers.7.mlp.experts.6.gate_proj", "model.layers.7.mlp.experts.7.gate_proj", "model.layers.7.mlp.experts.8.gate_proj", "model.layers.7.mlp.experts.9.gate_proj", "model.layers.7.mlp.experts.10.gate_proj", "model.layers.7.mlp.experts.11.gate_proj", "model.layers.7.mlp.experts.12.gate_proj", "model.layers.7.mlp.experts.13.gate_proj", "model.layers.7.mlp.experts.14.gate_proj", "model.layers.7.mlp.experts.15.gate_proj", "model.layers.7.mlp.experts.16.gate_proj", "model.layers.7.mlp.experts.17.gate_proj", "model.layers.7.mlp.experts.18.gate_proj", "model.layers.7.mlp.experts.19.gate_proj", "model.layers.7.mlp.experts.20.gate_proj", "model.layers.7.mlp.experts.21.gate_proj", "model.layers.7.mlp.experts.22.gate_proj", "model.layers.7.mlp.experts.23.gate_proj", "model.layers.7.mlp.experts.24.gate_proj", "model.layers.7.mlp.experts.25.gate_proj", "model.layers.7.mlp.experts.26.gate_proj", "model.layers.7.mlp.experts.27.gate_proj", "model.layers.7.mlp.experts.28.gate_proj", "model.layers.7.mlp.experts.29.gate_proj", "model.layers.7.mlp.experts.30.gate_proj", "model.layers.7.mlp.experts.31.gate_proj", "model.layers.7.mlp.experts.32.gate_proj", "model.layers.7.mlp.experts.33.gate_proj", "model.layers.7.mlp.experts.34.gate_proj", "model.layers.7.mlp.experts.35.gate_proj", "model.layers.7.mlp.experts.36.gate_proj", "model.layers.7.mlp.experts.37.gate_proj", "model.layers.7.mlp.experts.38.gate_proj", "model.layers.7.mlp.experts.39.gate_proj", "model.layers.7.mlp.experts.40.gate_proj", "model.layers.7.mlp.experts.41.gate_proj", "model.layers.7.mlp.experts.42.gate_proj", "model.layers.7.mlp.experts.43.gate_proj", "model.layers.7.mlp.experts.44.gate_proj", "model.layers.7.mlp.experts.45.gate_proj", "model.layers.7.mlp.experts.46.gate_proj", "model.layers.7.mlp.experts.47.gate_proj", "model.layers.7.mlp.experts.48.gate_proj", "model.layers.7.mlp.experts.49.gate_proj", "model.layers.7.mlp.experts.50.gate_proj", "model.layers.7.mlp.experts.51.gate_proj", "model.layers.7.mlp.experts.52.gate_proj", "model.layers.7.mlp.experts.53.gate_proj", "model.layers.7.mlp.experts.54.gate_proj", "model.layers.7.mlp.experts.55.gate_proj", "model.layers.7.mlp.experts.56.gate_proj", "model.layers.7.mlp.experts.57.gate_proj", "model.layers.7.mlp.experts.58.gate_proj", "model.layers.7.mlp.experts.59.gate_proj", "model.layers.7.mlp.experts.60.gate_proj", "model.layers.7.mlp.experts.61.gate_proj", "model.layers.7.mlp.experts.62.gate_proj", "model.layers.7.mlp.experts.63.gate_proj", "model.layers.7.mlp.experts.64.gate_proj", "model.layers.7.mlp.experts.65.gate_proj", "model.layers.7.mlp.experts.66.gate_proj", "model.layers.7.mlp.experts.67.gate_proj", "model.layers.7.mlp.experts.68.gate_proj", "model.layers.7.mlp.experts.69.gate_proj", "model.layers.7.mlp.experts.70.gate_proj", "model.layers.7.mlp.experts.71.gate_proj", "model.layers.7.mlp.experts.72.gate_proj", "model.layers.7.mlp.experts.73.gate_proj", "model.layers.7.mlp.experts.74.gate_proj", "model.layers.7.mlp.experts.75.gate_proj", "model.layers.7.mlp.experts.76.gate_proj", "model.layers.7.mlp.experts.77.gate_proj", "model.layers.7.mlp.experts.78.gate_proj", "model.layers.7.mlp.experts.79.gate_proj", "model.layers.7.mlp.experts.80.gate_proj", "model.layers.7.mlp.experts.81.gate_proj", "model.layers.7.mlp.experts.82.gate_proj", "model.layers.7.mlp.experts.83.gate_proj", "model.layers.7.mlp.experts.84.gate_proj", "model.layers.7.mlp.experts.85.gate_proj", "model.layers.7.mlp.experts.86.gate_proj", "model.layers.7.mlp.experts.87.gate_proj", "model.layers.7.mlp.experts.88.gate_proj", "model.layers.7.mlp.experts.89.gate_proj", "model.layers.7.mlp.experts.90.gate_proj", "model.layers.7.mlp.experts.91.gate_proj", "model.layers.7.mlp.experts.92.gate_proj", "model.layers.7.mlp.experts.93.gate_proj", "model.layers.7.mlp.experts.94.gate_proj", "model.layers.7.mlp.experts.95.gate_proj", "model.layers.7.mlp.experts.96.gate_proj", "model.layers.7.mlp.experts.97.gate_proj", "model.layers.7.mlp.experts.98.gate_proj", "model.layers.7.mlp.experts.99.gate_proj", "model.layers.7.mlp.experts.100.gate_proj", "model.layers.7.mlp.experts.101.gate_proj", "model.layers.7.mlp.experts.102.gate_proj", "model.layers.7.mlp.experts.103.gate_proj", "model.layers.7.mlp.experts.104.gate_proj", "model.layers.7.mlp.experts.105.gate_proj", "model.layers.7.mlp.experts.106.gate_proj", "model.layers.7.mlp.experts.107.gate_proj", "model.layers.7.mlp.experts.108.gate_proj", "model.layers.7.mlp.experts.109.gate_proj", "model.layers.7.mlp.experts.110.gate_proj", "model.layers.7.mlp.experts.111.gate_proj", "model.layers.7.mlp.experts.112.gate_proj", "model.layers.7.mlp.experts.113.gate_proj", "model.layers.7.mlp.experts.114.gate_proj", "model.layers.7.mlp.experts.115.gate_proj", "model.layers.7.mlp.experts.116.gate_proj", "model.layers.7.mlp.experts.117.gate_proj", "model.layers.7.mlp.experts.118.gate_proj", "model.layers.7.mlp.experts.119.gate_proj", "model.layers.7.mlp.experts.120.gate_proj", "model.layers.7.mlp.experts.121.gate_proj", "model.layers.7.mlp.experts.122.gate_proj", "model.layers.7.mlp.experts.123.gate_proj", "model.layers.7.mlp.experts.124.gate_proj", "model.layers.7.mlp.experts.125.gate_proj", "model.layers.7.mlp.experts.126.gate_proj", "model.layers.7.mlp.experts.127.gate_proj", "model.layers.7.mlp.experts.128.gate_proj", "model.layers.7.mlp.experts.129.gate_proj", "model.layers.7.mlp.experts.130.gate_proj", "model.layers.7.mlp.experts.131.gate_proj", "model.layers.7.mlp.experts.132.gate_proj", "model.layers.7.mlp.experts.133.gate_proj", "model.layers.7.mlp.experts.134.gate_proj", "model.layers.7.mlp.experts.135.gate_proj", "model.layers.7.mlp.experts.136.gate_proj", "model.layers.7.mlp.experts.137.gate_proj", "model.layers.7.mlp.experts.138.gate_proj", "model.layers.7.mlp.experts.139.gate_proj", "model.layers.7.mlp.experts.140.gate_proj", "model.layers.7.mlp.experts.141.gate_proj", "model.layers.7.mlp.experts.142.gate_proj", "model.layers.7.mlp.experts.143.gate_proj", "model.layers.7.mlp.experts.144.gate_proj", "model.layers.7.mlp.experts.145.gate_proj", "model.layers.7.mlp.experts.146.gate_proj", "model.layers.7.mlp.experts.147.gate_proj", "model.layers.7.mlp.experts.148.gate_proj", "model.layers.7.mlp.experts.149.gate_proj", "model.layers.7.mlp.experts.150.gate_proj", "model.layers.7.mlp.experts.151.gate_proj", "model.layers.7.mlp.experts.152.gate_proj", "model.layers.7.mlp.experts.153.gate_proj", "model.layers.7.mlp.experts.154.gate_proj", "model.layers.7.mlp.experts.155.gate_proj", "model.layers.7.mlp.experts.156.gate_proj", "model.layers.7.mlp.experts.157.gate_proj", "model.layers.7.mlp.experts.158.gate_proj", "model.layers.7.mlp.experts.159.gate_proj", "model.layers.7.mlp.experts.0.up_proj", "model.layers.7.mlp.experts.1.up_proj", "model.layers.7.mlp.experts.2.up_proj", "model.layers.7.mlp.experts.3.up_proj", "model.layers.7.mlp.experts.4.up_proj", "model.layers.7.mlp.experts.5.up_proj", "model.layers.7.mlp.experts.6.up_proj", "model.layers.7.mlp.experts.7.up_proj", "model.layers.7.mlp.experts.8.up_proj", "model.layers.7.mlp.experts.9.up_proj", "model.layers.7.mlp.experts.10.up_proj", "model.layers.7.mlp.experts.11.up_proj", "model.layers.7.mlp.experts.12.up_proj", "model.layers.7.mlp.experts.13.up_proj", "model.layers.7.mlp.experts.14.up_proj", "model.layers.7.mlp.experts.15.up_proj", "model.layers.7.mlp.experts.16.up_proj", "model.layers.7.mlp.experts.17.up_proj", "model.layers.7.mlp.experts.18.up_proj", "model.layers.7.mlp.experts.19.up_proj", "model.layers.7.mlp.experts.20.up_proj", "model.layers.7.mlp.experts.21.up_proj", "model.layers.7.mlp.experts.22.up_proj", "model.layers.7.mlp.experts.23.up_proj", "model.layers.7.mlp.experts.24.up_proj", "model.layers.7.mlp.experts.25.up_proj", "model.layers.7.mlp.experts.26.up_proj", "model.layers.7.mlp.experts.27.up_proj", "model.layers.7.mlp.experts.28.up_proj", "model.layers.7.mlp.experts.29.up_proj", "model.layers.7.mlp.experts.30.up_proj", "model.layers.7.mlp.experts.31.up_proj", "model.layers.7.mlp.experts.32.up_proj", "model.layers.7.mlp.experts.33.up_proj", "model.layers.7.mlp.experts.34.up_proj", "model.layers.7.mlp.experts.35.up_proj", "model.layers.7.mlp.experts.36.up_proj", "model.layers.7.mlp.experts.37.up_proj", "model.layers.7.mlp.experts.38.up_proj", "model.layers.7.mlp.experts.39.up_proj", "model.layers.7.mlp.experts.40.up_proj", "model.layers.7.mlp.experts.41.up_proj", "model.layers.7.mlp.experts.42.up_proj", "model.layers.7.mlp.experts.43.up_proj", "model.layers.7.mlp.experts.44.up_proj", "model.layers.7.mlp.experts.45.up_proj", "model.layers.7.mlp.experts.46.up_proj", "model.layers.7.mlp.experts.47.up_proj", "model.layers.7.mlp.experts.48.up_proj", "model.layers.7.mlp.experts.49.up_proj", "model.layers.7.mlp.experts.50.up_proj", "model.layers.7.mlp.experts.51.up_proj", "model.layers.7.mlp.experts.52.up_proj", "model.layers.7.mlp.experts.53.up_proj", "model.layers.7.mlp.experts.54.up_proj", "model.layers.7.mlp.experts.55.up_proj", "model.layers.7.mlp.experts.56.up_proj", "model.layers.7.mlp.experts.57.up_proj", "model.layers.7.mlp.experts.58.up_proj", "model.layers.7.mlp.experts.59.up_proj", "model.layers.7.mlp.experts.60.up_proj", "model.layers.7.mlp.experts.61.up_proj", "model.layers.7.mlp.experts.62.up_proj", "model.layers.7.mlp.experts.63.up_proj", "model.layers.7.mlp.experts.64.up_proj", "model.layers.7.mlp.experts.65.up_proj", "model.layers.7.mlp.experts.66.up_proj", "model.layers.7.mlp.experts.67.up_proj", "model.layers.7.mlp.experts.68.up_proj", "model.layers.7.mlp.experts.69.up_proj", "model.layers.7.mlp.experts.70.up_proj", "model.layers.7.mlp.experts.71.up_proj", "model.layers.7.mlp.experts.72.up_proj", "model.layers.7.mlp.experts.73.up_proj", "model.layers.7.mlp.experts.74.up_proj", "model.layers.7.mlp.experts.75.up_proj", "model.layers.7.mlp.experts.76.up_proj", "model.layers.7.mlp.experts.77.up_proj", "model.layers.7.mlp.experts.78.up_proj", "model.layers.7.mlp.experts.79.up_proj", "model.layers.7.mlp.experts.80.up_proj", "model.layers.7.mlp.experts.81.up_proj", "model.layers.7.mlp.experts.82.up_proj", "model.layers.7.mlp.experts.83.up_proj", "model.layers.7.mlp.experts.84.up_proj", "model.layers.7.mlp.experts.85.up_proj", "model.layers.7.mlp.experts.86.up_proj", "model.layers.7.mlp.experts.87.up_proj", "model.layers.7.mlp.experts.88.up_proj", "model.layers.7.mlp.experts.89.up_proj", "model.layers.7.mlp.experts.90.up_proj", "model.layers.7.mlp.experts.91.up_proj", "model.layers.7.mlp.experts.92.up_proj", "model.layers.7.mlp.experts.93.up_proj", "model.layers.7.mlp.experts.94.up_proj", "model.layers.7.mlp.experts.95.up_proj", "model.layers.7.mlp.experts.96.up_proj", "model.layers.7.mlp.experts.97.up_proj", "model.layers.7.mlp.experts.98.up_proj", "model.layers.7.mlp.experts.99.up_proj", "model.layers.7.mlp.experts.100.up_proj", "model.layers.7.mlp.experts.101.up_proj", "model.layers.7.mlp.experts.102.up_proj", "model.layers.7.mlp.experts.103.up_proj", "model.layers.7.mlp.experts.104.up_proj", "model.layers.7.mlp.experts.105.up_proj", "model.layers.7.mlp.experts.106.up_proj", "model.layers.7.mlp.experts.107.up_proj", "model.layers.7.mlp.experts.108.up_proj", "model.layers.7.mlp.experts.109.up_proj", "model.layers.7.mlp.experts.110.up_proj", "model.layers.7.mlp.experts.111.up_proj", "model.layers.7.mlp.experts.112.up_proj", "model.layers.7.mlp.experts.113.up_proj", "model.layers.7.mlp.experts.114.up_proj", "model.layers.7.mlp.experts.115.up_proj", "model.layers.7.mlp.experts.116.up_proj", "model.layers.7.mlp.experts.117.up_proj", "model.layers.7.mlp.experts.118.up_proj", "model.layers.7.mlp.experts.119.up_proj", "model.layers.7.mlp.experts.120.up_proj", "model.layers.7.mlp.experts.121.up_proj", "model.layers.7.mlp.experts.122.up_proj", "model.layers.7.mlp.experts.123.up_proj", "model.layers.7.mlp.experts.124.up_proj", "model.layers.7.mlp.experts.125.up_proj", "model.layers.7.mlp.experts.126.up_proj", "model.layers.7.mlp.experts.127.up_proj", "model.layers.7.mlp.experts.128.up_proj", "model.layers.7.mlp.experts.129.up_proj", "model.layers.7.mlp.experts.130.up_proj", "model.layers.7.mlp.experts.131.up_proj", "model.layers.7.mlp.experts.132.up_proj", "model.layers.7.mlp.experts.133.up_proj", "model.layers.7.mlp.experts.134.up_proj", "model.layers.7.mlp.experts.135.up_proj", "model.layers.7.mlp.experts.136.up_proj", "model.layers.7.mlp.experts.137.up_proj", "model.layers.7.mlp.experts.138.up_proj", "model.layers.7.mlp.experts.139.up_proj", "model.layers.7.mlp.experts.140.up_proj", "model.layers.7.mlp.experts.141.up_proj", "model.layers.7.mlp.experts.142.up_proj", "model.layers.7.mlp.experts.143.up_proj", "model.layers.7.mlp.experts.144.up_proj", "model.layers.7.mlp.experts.145.up_proj", "model.layers.7.mlp.experts.146.up_proj", "model.layers.7.mlp.experts.147.up_proj", "model.layers.7.mlp.experts.148.up_proj", "model.layers.7.mlp.experts.149.up_proj", "model.layers.7.mlp.experts.150.up_proj", "model.layers.7.mlp.experts.151.up_proj", "model.layers.7.mlp.experts.152.up_proj", "model.layers.7.mlp.experts.153.up_proj", "model.layers.7.mlp.experts.154.up_proj", "model.layers.7.mlp.experts.155.up_proj", "model.layers.7.mlp.experts.156.up_proj", "model.layers.7.mlp.experts.157.up_proj", "model.layers.7.mlp.experts.158.up_proj", "model.layers.7.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00027795908972622074, "dbits": 2516582400 }, { "dkld": -0.0003251033835113132, "dbits": 5033164800 }, { "dkld": -0.000463888980448246, "dbits": 7549747200 }, { "dkld": -0.000494949333369743, "dbits": 12582912000 } ] }, { "idx": 44, "layers": [ "model.layers.7.mlp.experts.0.down_proj", "model.layers.7.mlp.experts.1.down_proj", "model.layers.7.mlp.experts.2.down_proj", "model.layers.7.mlp.experts.3.down_proj", "model.layers.7.mlp.experts.4.down_proj", "model.layers.7.mlp.experts.5.down_proj", "model.layers.7.mlp.experts.6.down_proj", "model.layers.7.mlp.experts.7.down_proj", "model.layers.7.mlp.experts.8.down_proj", "model.layers.7.mlp.experts.9.down_proj", "model.layers.7.mlp.experts.10.down_proj", "model.layers.7.mlp.experts.11.down_proj", "model.layers.7.mlp.experts.12.down_proj", "model.layers.7.mlp.experts.13.down_proj", "model.layers.7.mlp.experts.14.down_proj", "model.layers.7.mlp.experts.15.down_proj", "model.layers.7.mlp.experts.16.down_proj", "model.layers.7.mlp.experts.17.down_proj", "model.layers.7.mlp.experts.18.down_proj", "model.layers.7.mlp.experts.19.down_proj", "model.layers.7.mlp.experts.20.down_proj", "model.layers.7.mlp.experts.21.down_proj", "model.layers.7.mlp.experts.22.down_proj", "model.layers.7.mlp.experts.23.down_proj", "model.layers.7.mlp.experts.24.down_proj", "model.layers.7.mlp.experts.25.down_proj", "model.layers.7.mlp.experts.26.down_proj", "model.layers.7.mlp.experts.27.down_proj", "model.layers.7.mlp.experts.28.down_proj", "model.layers.7.mlp.experts.29.down_proj", "model.layers.7.mlp.experts.30.down_proj", "model.layers.7.mlp.experts.31.down_proj", "model.layers.7.mlp.experts.32.down_proj", "model.layers.7.mlp.experts.33.down_proj", "model.layers.7.mlp.experts.34.down_proj", "model.layers.7.mlp.experts.35.down_proj", "model.layers.7.mlp.experts.36.down_proj", "model.layers.7.mlp.experts.37.down_proj", "model.layers.7.mlp.experts.38.down_proj", "model.layers.7.mlp.experts.39.down_proj", "model.layers.7.mlp.experts.40.down_proj", "model.layers.7.mlp.experts.41.down_proj", "model.layers.7.mlp.experts.42.down_proj", "model.layers.7.mlp.experts.43.down_proj", "model.layers.7.mlp.experts.44.down_proj", "model.layers.7.mlp.experts.45.down_proj", "model.layers.7.mlp.experts.46.down_proj", "model.layers.7.mlp.experts.47.down_proj", "model.layers.7.mlp.experts.48.down_proj", "model.layers.7.mlp.experts.49.down_proj", "model.layers.7.mlp.experts.50.down_proj", "model.layers.7.mlp.experts.51.down_proj", "model.layers.7.mlp.experts.52.down_proj", "model.layers.7.mlp.experts.53.down_proj", "model.layers.7.mlp.experts.54.down_proj", "model.layers.7.mlp.experts.55.down_proj", "model.layers.7.mlp.experts.56.down_proj", "model.layers.7.mlp.experts.57.down_proj", "model.layers.7.mlp.experts.58.down_proj", "model.layers.7.mlp.experts.59.down_proj", "model.layers.7.mlp.experts.60.down_proj", "model.layers.7.mlp.experts.61.down_proj", "model.layers.7.mlp.experts.62.down_proj", "model.layers.7.mlp.experts.63.down_proj", "model.layers.7.mlp.experts.64.down_proj", "model.layers.7.mlp.experts.65.down_proj", "model.layers.7.mlp.experts.66.down_proj", "model.layers.7.mlp.experts.67.down_proj", "model.layers.7.mlp.experts.68.down_proj", "model.layers.7.mlp.experts.69.down_proj", "model.layers.7.mlp.experts.70.down_proj", "model.layers.7.mlp.experts.71.down_proj", "model.layers.7.mlp.experts.72.down_proj", "model.layers.7.mlp.experts.73.down_proj", "model.layers.7.mlp.experts.74.down_proj", "model.layers.7.mlp.experts.75.down_proj", "model.layers.7.mlp.experts.76.down_proj", "model.layers.7.mlp.experts.77.down_proj", "model.layers.7.mlp.experts.78.down_proj", "model.layers.7.mlp.experts.79.down_proj", "model.layers.7.mlp.experts.80.down_proj", "model.layers.7.mlp.experts.81.down_proj", "model.layers.7.mlp.experts.82.down_proj", "model.layers.7.mlp.experts.83.down_proj", "model.layers.7.mlp.experts.84.down_proj", "model.layers.7.mlp.experts.85.down_proj", "model.layers.7.mlp.experts.86.down_proj", "model.layers.7.mlp.experts.87.down_proj", "model.layers.7.mlp.experts.88.down_proj", "model.layers.7.mlp.experts.89.down_proj", "model.layers.7.mlp.experts.90.down_proj", "model.layers.7.mlp.experts.91.down_proj", "model.layers.7.mlp.experts.92.down_proj", "model.layers.7.mlp.experts.93.down_proj", "model.layers.7.mlp.experts.94.down_proj", "model.layers.7.mlp.experts.95.down_proj", "model.layers.7.mlp.experts.96.down_proj", "model.layers.7.mlp.experts.97.down_proj", "model.layers.7.mlp.experts.98.down_proj", "model.layers.7.mlp.experts.99.down_proj", "model.layers.7.mlp.experts.100.down_proj", "model.layers.7.mlp.experts.101.down_proj", "model.layers.7.mlp.experts.102.down_proj", "model.layers.7.mlp.experts.103.down_proj", "model.layers.7.mlp.experts.104.down_proj", "model.layers.7.mlp.experts.105.down_proj", "model.layers.7.mlp.experts.106.down_proj", "model.layers.7.mlp.experts.107.down_proj", "model.layers.7.mlp.experts.108.down_proj", "model.layers.7.mlp.experts.109.down_proj", "model.layers.7.mlp.experts.110.down_proj", "model.layers.7.mlp.experts.111.down_proj", "model.layers.7.mlp.experts.112.down_proj", "model.layers.7.mlp.experts.113.down_proj", "model.layers.7.mlp.experts.114.down_proj", "model.layers.7.mlp.experts.115.down_proj", "model.layers.7.mlp.experts.116.down_proj", "model.layers.7.mlp.experts.117.down_proj", "model.layers.7.mlp.experts.118.down_proj", "model.layers.7.mlp.experts.119.down_proj", "model.layers.7.mlp.experts.120.down_proj", "model.layers.7.mlp.experts.121.down_proj", "model.layers.7.mlp.experts.122.down_proj", "model.layers.7.mlp.experts.123.down_proj", "model.layers.7.mlp.experts.124.down_proj", "model.layers.7.mlp.experts.125.down_proj", "model.layers.7.mlp.experts.126.down_proj", "model.layers.7.mlp.experts.127.down_proj", "model.layers.7.mlp.experts.128.down_proj", "model.layers.7.mlp.experts.129.down_proj", "model.layers.7.mlp.experts.130.down_proj", "model.layers.7.mlp.experts.131.down_proj", "model.layers.7.mlp.experts.132.down_proj", "model.layers.7.mlp.experts.133.down_proj", "model.layers.7.mlp.experts.134.down_proj", "model.layers.7.mlp.experts.135.down_proj", "model.layers.7.mlp.experts.136.down_proj", "model.layers.7.mlp.experts.137.down_proj", "model.layers.7.mlp.experts.138.down_proj", "model.layers.7.mlp.experts.139.down_proj", "model.layers.7.mlp.experts.140.down_proj", "model.layers.7.mlp.experts.141.down_proj", "model.layers.7.mlp.experts.142.down_proj", "model.layers.7.mlp.experts.143.down_proj", "model.layers.7.mlp.experts.144.down_proj", "model.layers.7.mlp.experts.145.down_proj", "model.layers.7.mlp.experts.146.down_proj", "model.layers.7.mlp.experts.147.down_proj", "model.layers.7.mlp.experts.148.down_proj", "model.layers.7.mlp.experts.149.down_proj", "model.layers.7.mlp.experts.150.down_proj", "model.layers.7.mlp.experts.151.down_proj", "model.layers.7.mlp.experts.152.down_proj", "model.layers.7.mlp.experts.153.down_proj", "model.layers.7.mlp.experts.154.down_proj", "model.layers.7.mlp.experts.155.down_proj", "model.layers.7.mlp.experts.156.down_proj", "model.layers.7.mlp.experts.157.down_proj", "model.layers.7.mlp.experts.158.down_proj", "model.layers.7.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00039677368476986885, "dbits": 1258291200 }, { "dkld": -0.0003865726292133359, "dbits": 2516582400 }, { "dkld": -0.0003521127626299886, "dbits": 3774873600 }, { "dkld": -0.00036147814244032184, "dbits": 6291456000 } ] }, { "idx": 45, "layers": [ "model.layers.8.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00016594296321273405, "dbits": 62914560 }, { "dkld": -0.00026046307757497666, "dbits": 125829120 }, { "dkld": -0.00026713935658335963, "dbits": 188743680 }, { "dkld": -0.00018874621018767634, "dbits": 314572800 } ] }, { "idx": 46, "layers": [ "model.layers.8.self_attn.k_proj", "model.layers.8.self_attn.v_proj" ], "candidates": [ { "dkld": 0.000859662331640712, "dbits": 10485760 }, { "dkld": 0.0003103368915617355, "dbits": 20971520 }, { "dkld": 0.0002727684564888394, "dbits": 31457280 }, { "dkld": 0.0002974383533000946, "dbits": 52428800 } ] }, { "idx": 47, "layers": [ "model.layers.8.self_attn.o_proj" ], "candidates": [ { "dkld": -0.001276416145265108, "dbits": 62914560 }, { "dkld": -0.0010513480752706583, "dbits": 125829120 }, { "dkld": -0.0010937627404928318, "dbits": 188743680 }, { "dkld": -0.0010093452408909825, "dbits": 314572800 } ] }, { "idx": 48, "layers": [ "model.layers.8.mlp.shared_experts.gate_proj", "model.layers.8.mlp.shared_experts.up_proj", "model.layers.8.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00019113505259155117, "dbits": 23592960 }, { "dkld": 0.00012982990592717847, "dbits": 47185920 }, { "dkld": 5.0811935216180104e-05, "dbits": 70778880 }, { "dkld": 0.00012551406398414455, "dbits": 117964800 } ] }, { "idx": 49, "layers": [ "model.layers.8.mlp.experts.0.gate_proj", "model.layers.8.mlp.experts.1.gate_proj", "model.layers.8.mlp.experts.2.gate_proj", "model.layers.8.mlp.experts.3.gate_proj", "model.layers.8.mlp.experts.4.gate_proj", "model.layers.8.mlp.experts.5.gate_proj", "model.layers.8.mlp.experts.6.gate_proj", "model.layers.8.mlp.experts.7.gate_proj", "model.layers.8.mlp.experts.8.gate_proj", "model.layers.8.mlp.experts.9.gate_proj", "model.layers.8.mlp.experts.10.gate_proj", "model.layers.8.mlp.experts.11.gate_proj", "model.layers.8.mlp.experts.12.gate_proj", "model.layers.8.mlp.experts.13.gate_proj", "model.layers.8.mlp.experts.14.gate_proj", "model.layers.8.mlp.experts.15.gate_proj", "model.layers.8.mlp.experts.16.gate_proj", "model.layers.8.mlp.experts.17.gate_proj", "model.layers.8.mlp.experts.18.gate_proj", "model.layers.8.mlp.experts.19.gate_proj", "model.layers.8.mlp.experts.20.gate_proj", "model.layers.8.mlp.experts.21.gate_proj", "model.layers.8.mlp.experts.22.gate_proj", "model.layers.8.mlp.experts.23.gate_proj", "model.layers.8.mlp.experts.24.gate_proj", "model.layers.8.mlp.experts.25.gate_proj", "model.layers.8.mlp.experts.26.gate_proj", "model.layers.8.mlp.experts.27.gate_proj", "model.layers.8.mlp.experts.28.gate_proj", "model.layers.8.mlp.experts.29.gate_proj", "model.layers.8.mlp.experts.30.gate_proj", "model.layers.8.mlp.experts.31.gate_proj", "model.layers.8.mlp.experts.32.gate_proj", "model.layers.8.mlp.experts.33.gate_proj", "model.layers.8.mlp.experts.34.gate_proj", "model.layers.8.mlp.experts.35.gate_proj", "model.layers.8.mlp.experts.36.gate_proj", "model.layers.8.mlp.experts.37.gate_proj", "model.layers.8.mlp.experts.38.gate_proj", "model.layers.8.mlp.experts.39.gate_proj", "model.layers.8.mlp.experts.40.gate_proj", "model.layers.8.mlp.experts.41.gate_proj", "model.layers.8.mlp.experts.42.gate_proj", "model.layers.8.mlp.experts.43.gate_proj", "model.layers.8.mlp.experts.44.gate_proj", "model.layers.8.mlp.experts.45.gate_proj", "model.layers.8.mlp.experts.46.gate_proj", "model.layers.8.mlp.experts.47.gate_proj", "model.layers.8.mlp.experts.48.gate_proj", "model.layers.8.mlp.experts.49.gate_proj", "model.layers.8.mlp.experts.50.gate_proj", "model.layers.8.mlp.experts.51.gate_proj", "model.layers.8.mlp.experts.52.gate_proj", "model.layers.8.mlp.experts.53.gate_proj", "model.layers.8.mlp.experts.54.gate_proj", "model.layers.8.mlp.experts.55.gate_proj", "model.layers.8.mlp.experts.56.gate_proj", "model.layers.8.mlp.experts.57.gate_proj", "model.layers.8.mlp.experts.58.gate_proj", "model.layers.8.mlp.experts.59.gate_proj", "model.layers.8.mlp.experts.60.gate_proj", "model.layers.8.mlp.experts.61.gate_proj", "model.layers.8.mlp.experts.62.gate_proj", "model.layers.8.mlp.experts.63.gate_proj", "model.layers.8.mlp.experts.64.gate_proj", "model.layers.8.mlp.experts.65.gate_proj", "model.layers.8.mlp.experts.66.gate_proj", "model.layers.8.mlp.experts.67.gate_proj", "model.layers.8.mlp.experts.68.gate_proj", "model.layers.8.mlp.experts.69.gate_proj", "model.layers.8.mlp.experts.70.gate_proj", "model.layers.8.mlp.experts.71.gate_proj", "model.layers.8.mlp.experts.72.gate_proj", "model.layers.8.mlp.experts.73.gate_proj", "model.layers.8.mlp.experts.74.gate_proj", "model.layers.8.mlp.experts.75.gate_proj", "model.layers.8.mlp.experts.76.gate_proj", "model.layers.8.mlp.experts.77.gate_proj", "model.layers.8.mlp.experts.78.gate_proj", "model.layers.8.mlp.experts.79.gate_proj", "model.layers.8.mlp.experts.80.gate_proj", "model.layers.8.mlp.experts.81.gate_proj", "model.layers.8.mlp.experts.82.gate_proj", "model.layers.8.mlp.experts.83.gate_proj", "model.layers.8.mlp.experts.84.gate_proj", "model.layers.8.mlp.experts.85.gate_proj", "model.layers.8.mlp.experts.86.gate_proj", "model.layers.8.mlp.experts.87.gate_proj", "model.layers.8.mlp.experts.88.gate_proj", "model.layers.8.mlp.experts.89.gate_proj", "model.layers.8.mlp.experts.90.gate_proj", "model.layers.8.mlp.experts.91.gate_proj", "model.layers.8.mlp.experts.92.gate_proj", "model.layers.8.mlp.experts.93.gate_proj", "model.layers.8.mlp.experts.94.gate_proj", "model.layers.8.mlp.experts.95.gate_proj", "model.layers.8.mlp.experts.96.gate_proj", "model.layers.8.mlp.experts.97.gate_proj", "model.layers.8.mlp.experts.98.gate_proj", "model.layers.8.mlp.experts.99.gate_proj", "model.layers.8.mlp.experts.100.gate_proj", "model.layers.8.mlp.experts.101.gate_proj", "model.layers.8.mlp.experts.102.gate_proj", "model.layers.8.mlp.experts.103.gate_proj", "model.layers.8.mlp.experts.104.gate_proj", "model.layers.8.mlp.experts.105.gate_proj", "model.layers.8.mlp.experts.106.gate_proj", "model.layers.8.mlp.experts.107.gate_proj", "model.layers.8.mlp.experts.108.gate_proj", "model.layers.8.mlp.experts.109.gate_proj", "model.layers.8.mlp.experts.110.gate_proj", "model.layers.8.mlp.experts.111.gate_proj", "model.layers.8.mlp.experts.112.gate_proj", "model.layers.8.mlp.experts.113.gate_proj", "model.layers.8.mlp.experts.114.gate_proj", "model.layers.8.mlp.experts.115.gate_proj", "model.layers.8.mlp.experts.116.gate_proj", "model.layers.8.mlp.experts.117.gate_proj", "model.layers.8.mlp.experts.118.gate_proj", "model.layers.8.mlp.experts.119.gate_proj", "model.layers.8.mlp.experts.120.gate_proj", "model.layers.8.mlp.experts.121.gate_proj", "model.layers.8.mlp.experts.122.gate_proj", "model.layers.8.mlp.experts.123.gate_proj", "model.layers.8.mlp.experts.124.gate_proj", "model.layers.8.mlp.experts.125.gate_proj", "model.layers.8.mlp.experts.126.gate_proj", "model.layers.8.mlp.experts.127.gate_proj", "model.layers.8.mlp.experts.128.gate_proj", "model.layers.8.mlp.experts.129.gate_proj", "model.layers.8.mlp.experts.130.gate_proj", "model.layers.8.mlp.experts.131.gate_proj", "model.layers.8.mlp.experts.132.gate_proj", "model.layers.8.mlp.experts.133.gate_proj", "model.layers.8.mlp.experts.134.gate_proj", "model.layers.8.mlp.experts.135.gate_proj", "model.layers.8.mlp.experts.136.gate_proj", "model.layers.8.mlp.experts.137.gate_proj", "model.layers.8.mlp.experts.138.gate_proj", "model.layers.8.mlp.experts.139.gate_proj", "model.layers.8.mlp.experts.140.gate_proj", "model.layers.8.mlp.experts.141.gate_proj", "model.layers.8.mlp.experts.142.gate_proj", "model.layers.8.mlp.experts.143.gate_proj", "model.layers.8.mlp.experts.144.gate_proj", "model.layers.8.mlp.experts.145.gate_proj", "model.layers.8.mlp.experts.146.gate_proj", "model.layers.8.mlp.experts.147.gate_proj", "model.layers.8.mlp.experts.148.gate_proj", "model.layers.8.mlp.experts.149.gate_proj", "model.layers.8.mlp.experts.150.gate_proj", "model.layers.8.mlp.experts.151.gate_proj", "model.layers.8.mlp.experts.152.gate_proj", "model.layers.8.mlp.experts.153.gate_proj", "model.layers.8.mlp.experts.154.gate_proj", "model.layers.8.mlp.experts.155.gate_proj", "model.layers.8.mlp.experts.156.gate_proj", "model.layers.8.mlp.experts.157.gate_proj", "model.layers.8.mlp.experts.158.gate_proj", "model.layers.8.mlp.experts.159.gate_proj", "model.layers.8.mlp.experts.0.up_proj", "model.layers.8.mlp.experts.1.up_proj", "model.layers.8.mlp.experts.2.up_proj", "model.layers.8.mlp.experts.3.up_proj", "model.layers.8.mlp.experts.4.up_proj", "model.layers.8.mlp.experts.5.up_proj", "model.layers.8.mlp.experts.6.up_proj", "model.layers.8.mlp.experts.7.up_proj", "model.layers.8.mlp.experts.8.up_proj", "model.layers.8.mlp.experts.9.up_proj", "model.layers.8.mlp.experts.10.up_proj", "model.layers.8.mlp.experts.11.up_proj", "model.layers.8.mlp.experts.12.up_proj", "model.layers.8.mlp.experts.13.up_proj", "model.layers.8.mlp.experts.14.up_proj", "model.layers.8.mlp.experts.15.up_proj", "model.layers.8.mlp.experts.16.up_proj", "model.layers.8.mlp.experts.17.up_proj", "model.layers.8.mlp.experts.18.up_proj", "model.layers.8.mlp.experts.19.up_proj", "model.layers.8.mlp.experts.20.up_proj", "model.layers.8.mlp.experts.21.up_proj", "model.layers.8.mlp.experts.22.up_proj", "model.layers.8.mlp.experts.23.up_proj", "model.layers.8.mlp.experts.24.up_proj", "model.layers.8.mlp.experts.25.up_proj", "model.layers.8.mlp.experts.26.up_proj", "model.layers.8.mlp.experts.27.up_proj", "model.layers.8.mlp.experts.28.up_proj", "model.layers.8.mlp.experts.29.up_proj", "model.layers.8.mlp.experts.30.up_proj", "model.layers.8.mlp.experts.31.up_proj", "model.layers.8.mlp.experts.32.up_proj", "model.layers.8.mlp.experts.33.up_proj", "model.layers.8.mlp.experts.34.up_proj", "model.layers.8.mlp.experts.35.up_proj", "model.layers.8.mlp.experts.36.up_proj", "model.layers.8.mlp.experts.37.up_proj", "model.layers.8.mlp.experts.38.up_proj", "model.layers.8.mlp.experts.39.up_proj", "model.layers.8.mlp.experts.40.up_proj", "model.layers.8.mlp.experts.41.up_proj", "model.layers.8.mlp.experts.42.up_proj", "model.layers.8.mlp.experts.43.up_proj", "model.layers.8.mlp.experts.44.up_proj", "model.layers.8.mlp.experts.45.up_proj", "model.layers.8.mlp.experts.46.up_proj", "model.layers.8.mlp.experts.47.up_proj", "model.layers.8.mlp.experts.48.up_proj", "model.layers.8.mlp.experts.49.up_proj", "model.layers.8.mlp.experts.50.up_proj", "model.layers.8.mlp.experts.51.up_proj", "model.layers.8.mlp.experts.52.up_proj", "model.layers.8.mlp.experts.53.up_proj", "model.layers.8.mlp.experts.54.up_proj", "model.layers.8.mlp.experts.55.up_proj", "model.layers.8.mlp.experts.56.up_proj", "model.layers.8.mlp.experts.57.up_proj", "model.layers.8.mlp.experts.58.up_proj", "model.layers.8.mlp.experts.59.up_proj", "model.layers.8.mlp.experts.60.up_proj", "model.layers.8.mlp.experts.61.up_proj", "model.layers.8.mlp.experts.62.up_proj", "model.layers.8.mlp.experts.63.up_proj", "model.layers.8.mlp.experts.64.up_proj", "model.layers.8.mlp.experts.65.up_proj", "model.layers.8.mlp.experts.66.up_proj", "model.layers.8.mlp.experts.67.up_proj", "model.layers.8.mlp.experts.68.up_proj", "model.layers.8.mlp.experts.69.up_proj", "model.layers.8.mlp.experts.70.up_proj", "model.layers.8.mlp.experts.71.up_proj", "model.layers.8.mlp.experts.72.up_proj", "model.layers.8.mlp.experts.73.up_proj", "model.layers.8.mlp.experts.74.up_proj", "model.layers.8.mlp.experts.75.up_proj", "model.layers.8.mlp.experts.76.up_proj", "model.layers.8.mlp.experts.77.up_proj", "model.layers.8.mlp.experts.78.up_proj", "model.layers.8.mlp.experts.79.up_proj", "model.layers.8.mlp.experts.80.up_proj", "model.layers.8.mlp.experts.81.up_proj", "model.layers.8.mlp.experts.82.up_proj", "model.layers.8.mlp.experts.83.up_proj", "model.layers.8.mlp.experts.84.up_proj", "model.layers.8.mlp.experts.85.up_proj", "model.layers.8.mlp.experts.86.up_proj", "model.layers.8.mlp.experts.87.up_proj", "model.layers.8.mlp.experts.88.up_proj", "model.layers.8.mlp.experts.89.up_proj", "model.layers.8.mlp.experts.90.up_proj", "model.layers.8.mlp.experts.91.up_proj", "model.layers.8.mlp.experts.92.up_proj", "model.layers.8.mlp.experts.93.up_proj", "model.layers.8.mlp.experts.94.up_proj", "model.layers.8.mlp.experts.95.up_proj", "model.layers.8.mlp.experts.96.up_proj", "model.layers.8.mlp.experts.97.up_proj", "model.layers.8.mlp.experts.98.up_proj", "model.layers.8.mlp.experts.99.up_proj", "model.layers.8.mlp.experts.100.up_proj", "model.layers.8.mlp.experts.101.up_proj", "model.layers.8.mlp.experts.102.up_proj", "model.layers.8.mlp.experts.103.up_proj", "model.layers.8.mlp.experts.104.up_proj", "model.layers.8.mlp.experts.105.up_proj", "model.layers.8.mlp.experts.106.up_proj", "model.layers.8.mlp.experts.107.up_proj", "model.layers.8.mlp.experts.108.up_proj", "model.layers.8.mlp.experts.109.up_proj", "model.layers.8.mlp.experts.110.up_proj", "model.layers.8.mlp.experts.111.up_proj", "model.layers.8.mlp.experts.112.up_proj", "model.layers.8.mlp.experts.113.up_proj", "model.layers.8.mlp.experts.114.up_proj", "model.layers.8.mlp.experts.115.up_proj", "model.layers.8.mlp.experts.116.up_proj", "model.layers.8.mlp.experts.117.up_proj", "model.layers.8.mlp.experts.118.up_proj", "model.layers.8.mlp.experts.119.up_proj", "model.layers.8.mlp.experts.120.up_proj", "model.layers.8.mlp.experts.121.up_proj", "model.layers.8.mlp.experts.122.up_proj", "model.layers.8.mlp.experts.123.up_proj", "model.layers.8.mlp.experts.124.up_proj", "model.layers.8.mlp.experts.125.up_proj", "model.layers.8.mlp.experts.126.up_proj", "model.layers.8.mlp.experts.127.up_proj", "model.layers.8.mlp.experts.128.up_proj", "model.layers.8.mlp.experts.129.up_proj", "model.layers.8.mlp.experts.130.up_proj", "model.layers.8.mlp.experts.131.up_proj", "model.layers.8.mlp.experts.132.up_proj", "model.layers.8.mlp.experts.133.up_proj", "model.layers.8.mlp.experts.134.up_proj", "model.layers.8.mlp.experts.135.up_proj", "model.layers.8.mlp.experts.136.up_proj", "model.layers.8.mlp.experts.137.up_proj", "model.layers.8.mlp.experts.138.up_proj", "model.layers.8.mlp.experts.139.up_proj", "model.layers.8.mlp.experts.140.up_proj", "model.layers.8.mlp.experts.141.up_proj", "model.layers.8.mlp.experts.142.up_proj", "model.layers.8.mlp.experts.143.up_proj", "model.layers.8.mlp.experts.144.up_proj", "model.layers.8.mlp.experts.145.up_proj", "model.layers.8.mlp.experts.146.up_proj", "model.layers.8.mlp.experts.147.up_proj", "model.layers.8.mlp.experts.148.up_proj", "model.layers.8.mlp.experts.149.up_proj", "model.layers.8.mlp.experts.150.up_proj", "model.layers.8.mlp.experts.151.up_proj", "model.layers.8.mlp.experts.152.up_proj", "model.layers.8.mlp.experts.153.up_proj", "model.layers.8.mlp.experts.154.up_proj", "model.layers.8.mlp.experts.155.up_proj", "model.layers.8.mlp.experts.156.up_proj", "model.layers.8.mlp.experts.157.up_proj", "model.layers.8.mlp.experts.158.up_proj", "model.layers.8.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00037158876657486517, "dbits": 2516582400 }, { "dkld": -0.00023330859839916507, "dbits": 5033164800 }, { "dkld": -0.0003645703196525629, "dbits": 7549747200 }, { "dkld": -0.0002839850261807525, "dbits": 12582912000 } ] }, { "idx": 50, "layers": [ "model.layers.8.mlp.experts.0.down_proj", "model.layers.8.mlp.experts.1.down_proj", "model.layers.8.mlp.experts.2.down_proj", "model.layers.8.mlp.experts.3.down_proj", "model.layers.8.mlp.experts.4.down_proj", "model.layers.8.mlp.experts.5.down_proj", "model.layers.8.mlp.experts.6.down_proj", "model.layers.8.mlp.experts.7.down_proj", "model.layers.8.mlp.experts.8.down_proj", "model.layers.8.mlp.experts.9.down_proj", "model.layers.8.mlp.experts.10.down_proj", "model.layers.8.mlp.experts.11.down_proj", "model.layers.8.mlp.experts.12.down_proj", "model.layers.8.mlp.experts.13.down_proj", "model.layers.8.mlp.experts.14.down_proj", "model.layers.8.mlp.experts.15.down_proj", "model.layers.8.mlp.experts.16.down_proj", "model.layers.8.mlp.experts.17.down_proj", "model.layers.8.mlp.experts.18.down_proj", "model.layers.8.mlp.experts.19.down_proj", "model.layers.8.mlp.experts.20.down_proj", "model.layers.8.mlp.experts.21.down_proj", "model.layers.8.mlp.experts.22.down_proj", "model.layers.8.mlp.experts.23.down_proj", "model.layers.8.mlp.experts.24.down_proj", "model.layers.8.mlp.experts.25.down_proj", "model.layers.8.mlp.experts.26.down_proj", "model.layers.8.mlp.experts.27.down_proj", "model.layers.8.mlp.experts.28.down_proj", "model.layers.8.mlp.experts.29.down_proj", "model.layers.8.mlp.experts.30.down_proj", "model.layers.8.mlp.experts.31.down_proj", "model.layers.8.mlp.experts.32.down_proj", "model.layers.8.mlp.experts.33.down_proj", "model.layers.8.mlp.experts.34.down_proj", "model.layers.8.mlp.experts.35.down_proj", "model.layers.8.mlp.experts.36.down_proj", "model.layers.8.mlp.experts.37.down_proj", "model.layers.8.mlp.experts.38.down_proj", "model.layers.8.mlp.experts.39.down_proj", "model.layers.8.mlp.experts.40.down_proj", "model.layers.8.mlp.experts.41.down_proj", "model.layers.8.mlp.experts.42.down_proj", "model.layers.8.mlp.experts.43.down_proj", "model.layers.8.mlp.experts.44.down_proj", "model.layers.8.mlp.experts.45.down_proj", "model.layers.8.mlp.experts.46.down_proj", "model.layers.8.mlp.experts.47.down_proj", "model.layers.8.mlp.experts.48.down_proj", "model.layers.8.mlp.experts.49.down_proj", "model.layers.8.mlp.experts.50.down_proj", "model.layers.8.mlp.experts.51.down_proj", "model.layers.8.mlp.experts.52.down_proj", "model.layers.8.mlp.experts.53.down_proj", "model.layers.8.mlp.experts.54.down_proj", "model.layers.8.mlp.experts.55.down_proj", "model.layers.8.mlp.experts.56.down_proj", "model.layers.8.mlp.experts.57.down_proj", "model.layers.8.mlp.experts.58.down_proj", "model.layers.8.mlp.experts.59.down_proj", "model.layers.8.mlp.experts.60.down_proj", "model.layers.8.mlp.experts.61.down_proj", "model.layers.8.mlp.experts.62.down_proj", "model.layers.8.mlp.experts.63.down_proj", "model.layers.8.mlp.experts.64.down_proj", "model.layers.8.mlp.experts.65.down_proj", "model.layers.8.mlp.experts.66.down_proj", "model.layers.8.mlp.experts.67.down_proj", "model.layers.8.mlp.experts.68.down_proj", "model.layers.8.mlp.experts.69.down_proj", "model.layers.8.mlp.experts.70.down_proj", "model.layers.8.mlp.experts.71.down_proj", "model.layers.8.mlp.experts.72.down_proj", "model.layers.8.mlp.experts.73.down_proj", "model.layers.8.mlp.experts.74.down_proj", "model.layers.8.mlp.experts.75.down_proj", "model.layers.8.mlp.experts.76.down_proj", "model.layers.8.mlp.experts.77.down_proj", "model.layers.8.mlp.experts.78.down_proj", "model.layers.8.mlp.experts.79.down_proj", "model.layers.8.mlp.experts.80.down_proj", "model.layers.8.mlp.experts.81.down_proj", "model.layers.8.mlp.experts.82.down_proj", "model.layers.8.mlp.experts.83.down_proj", "model.layers.8.mlp.experts.84.down_proj", "model.layers.8.mlp.experts.85.down_proj", "model.layers.8.mlp.experts.86.down_proj", "model.layers.8.mlp.experts.87.down_proj", "model.layers.8.mlp.experts.88.down_proj", "model.layers.8.mlp.experts.89.down_proj", "model.layers.8.mlp.experts.90.down_proj", "model.layers.8.mlp.experts.91.down_proj", "model.layers.8.mlp.experts.92.down_proj", "model.layers.8.mlp.experts.93.down_proj", "model.layers.8.mlp.experts.94.down_proj", "model.layers.8.mlp.experts.95.down_proj", "model.layers.8.mlp.experts.96.down_proj", "model.layers.8.mlp.experts.97.down_proj", "model.layers.8.mlp.experts.98.down_proj", "model.layers.8.mlp.experts.99.down_proj", "model.layers.8.mlp.experts.100.down_proj", "model.layers.8.mlp.experts.101.down_proj", "model.layers.8.mlp.experts.102.down_proj", "model.layers.8.mlp.experts.103.down_proj", "model.layers.8.mlp.experts.104.down_proj", "model.layers.8.mlp.experts.105.down_proj", "model.layers.8.mlp.experts.106.down_proj", "model.layers.8.mlp.experts.107.down_proj", "model.layers.8.mlp.experts.108.down_proj", "model.layers.8.mlp.experts.109.down_proj", "model.layers.8.mlp.experts.110.down_proj", "model.layers.8.mlp.experts.111.down_proj", "model.layers.8.mlp.experts.112.down_proj", "model.layers.8.mlp.experts.113.down_proj", "model.layers.8.mlp.experts.114.down_proj", "model.layers.8.mlp.experts.115.down_proj", "model.layers.8.mlp.experts.116.down_proj", "model.layers.8.mlp.experts.117.down_proj", "model.layers.8.mlp.experts.118.down_proj", "model.layers.8.mlp.experts.119.down_proj", "model.layers.8.mlp.experts.120.down_proj", "model.layers.8.mlp.experts.121.down_proj", "model.layers.8.mlp.experts.122.down_proj", "model.layers.8.mlp.experts.123.down_proj", "model.layers.8.mlp.experts.124.down_proj", "model.layers.8.mlp.experts.125.down_proj", "model.layers.8.mlp.experts.126.down_proj", "model.layers.8.mlp.experts.127.down_proj", "model.layers.8.mlp.experts.128.down_proj", "model.layers.8.mlp.experts.129.down_proj", "model.layers.8.mlp.experts.130.down_proj", "model.layers.8.mlp.experts.131.down_proj", "model.layers.8.mlp.experts.132.down_proj", "model.layers.8.mlp.experts.133.down_proj", "model.layers.8.mlp.experts.134.down_proj", "model.layers.8.mlp.experts.135.down_proj", "model.layers.8.mlp.experts.136.down_proj", "model.layers.8.mlp.experts.137.down_proj", "model.layers.8.mlp.experts.138.down_proj", "model.layers.8.mlp.experts.139.down_proj", "model.layers.8.mlp.experts.140.down_proj", "model.layers.8.mlp.experts.141.down_proj", "model.layers.8.mlp.experts.142.down_proj", "model.layers.8.mlp.experts.143.down_proj", "model.layers.8.mlp.experts.144.down_proj", "model.layers.8.mlp.experts.145.down_proj", "model.layers.8.mlp.experts.146.down_proj", "model.layers.8.mlp.experts.147.down_proj", "model.layers.8.mlp.experts.148.down_proj", "model.layers.8.mlp.experts.149.down_proj", "model.layers.8.mlp.experts.150.down_proj", "model.layers.8.mlp.experts.151.down_proj", "model.layers.8.mlp.experts.152.down_proj", "model.layers.8.mlp.experts.153.down_proj", "model.layers.8.mlp.experts.154.down_proj", "model.layers.8.mlp.experts.155.down_proj", "model.layers.8.mlp.experts.156.down_proj", "model.layers.8.mlp.experts.157.down_proj", "model.layers.8.mlp.experts.158.down_proj", "model.layers.8.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002638109959662016, "dbits": 1258291200 }, { "dkld": -0.0003043592907488457, "dbits": 2516582400 }, { "dkld": -0.00031046932563186247, "dbits": 3774873600 }, { "dkld": -0.0003249408677220428, "dbits": 6291456000 } ] }, { "idx": 51, "layers": [ "model.layers.9.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00022493302822113037, "dbits": 62914560 }, { "dkld": -0.00017484119161964573, "dbits": 125829120 }, { "dkld": -0.00028562992811204113, "dbits": 188743680 }, { "dkld": -0.00033493414521218456, "dbits": 314572800 } ] }, { "idx": 52, "layers": [ "model.layers.9.self_attn.k_proj", "model.layers.9.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007009593769908007, "dbits": 10485760 }, { "dkld": -0.000234160199761399, "dbits": 20971520 }, { "dkld": -0.000471746362745773, "dbits": 31457280 }, { "dkld": -0.0005248092114925468, "dbits": 52428800 } ] }, { "idx": 53, "layers": [ "model.layers.9.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002992817200720421, "dbits": 62914560 }, { "dkld": -0.0002594418823719108, "dbits": 125829120 }, { "dkld": -0.0003222662955522648, "dbits": 188743680 }, { "dkld": -0.00033514760434627533, "dbits": 314572800 } ] }, { "idx": 54, "layers": [ "model.layers.9.mlp.shared_experts.gate_proj", "model.layers.9.mlp.shared_experts.up_proj", "model.layers.9.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0002648031339049367, "dbits": 23592960 }, { "dkld": 7.220916450023651e-05, "dbits": 47185920 }, { "dkld": 0.00011021625250577649, "dbits": 70778880 }, { "dkld": 0.00011518709361552082, "dbits": 117964800 } ] }, { "idx": 55, "layers": [ "model.layers.9.mlp.experts.0.gate_proj", "model.layers.9.mlp.experts.1.gate_proj", "model.layers.9.mlp.experts.2.gate_proj", "model.layers.9.mlp.experts.3.gate_proj", "model.layers.9.mlp.experts.4.gate_proj", "model.layers.9.mlp.experts.5.gate_proj", "model.layers.9.mlp.experts.6.gate_proj", "model.layers.9.mlp.experts.7.gate_proj", "model.layers.9.mlp.experts.8.gate_proj", "model.layers.9.mlp.experts.9.gate_proj", "model.layers.9.mlp.experts.10.gate_proj", "model.layers.9.mlp.experts.11.gate_proj", "model.layers.9.mlp.experts.12.gate_proj", "model.layers.9.mlp.experts.13.gate_proj", "model.layers.9.mlp.experts.14.gate_proj", "model.layers.9.mlp.experts.15.gate_proj", "model.layers.9.mlp.experts.16.gate_proj", "model.layers.9.mlp.experts.17.gate_proj", "model.layers.9.mlp.experts.18.gate_proj", "model.layers.9.mlp.experts.19.gate_proj", "model.layers.9.mlp.experts.20.gate_proj", "model.layers.9.mlp.experts.21.gate_proj", "model.layers.9.mlp.experts.22.gate_proj", "model.layers.9.mlp.experts.23.gate_proj", "model.layers.9.mlp.experts.24.gate_proj", "model.layers.9.mlp.experts.25.gate_proj", "model.layers.9.mlp.experts.26.gate_proj", "model.layers.9.mlp.experts.27.gate_proj", "model.layers.9.mlp.experts.28.gate_proj", "model.layers.9.mlp.experts.29.gate_proj", "model.layers.9.mlp.experts.30.gate_proj", "model.layers.9.mlp.experts.31.gate_proj", "model.layers.9.mlp.experts.32.gate_proj", "model.layers.9.mlp.experts.33.gate_proj", "model.layers.9.mlp.experts.34.gate_proj", "model.layers.9.mlp.experts.35.gate_proj", "model.layers.9.mlp.experts.36.gate_proj", "model.layers.9.mlp.experts.37.gate_proj", "model.layers.9.mlp.experts.38.gate_proj", "model.layers.9.mlp.experts.39.gate_proj", "model.layers.9.mlp.experts.40.gate_proj", "model.layers.9.mlp.experts.41.gate_proj", "model.layers.9.mlp.experts.42.gate_proj", "model.layers.9.mlp.experts.43.gate_proj", "model.layers.9.mlp.experts.44.gate_proj", "model.layers.9.mlp.experts.45.gate_proj", "model.layers.9.mlp.experts.46.gate_proj", "model.layers.9.mlp.experts.47.gate_proj", "model.layers.9.mlp.experts.48.gate_proj", "model.layers.9.mlp.experts.49.gate_proj", "model.layers.9.mlp.experts.50.gate_proj", "model.layers.9.mlp.experts.51.gate_proj", "model.layers.9.mlp.experts.52.gate_proj", "model.layers.9.mlp.experts.53.gate_proj", "model.layers.9.mlp.experts.54.gate_proj", "model.layers.9.mlp.experts.55.gate_proj", "model.layers.9.mlp.experts.56.gate_proj", "model.layers.9.mlp.experts.57.gate_proj", "model.layers.9.mlp.experts.58.gate_proj", "model.layers.9.mlp.experts.59.gate_proj", "model.layers.9.mlp.experts.60.gate_proj", "model.layers.9.mlp.experts.61.gate_proj", "model.layers.9.mlp.experts.62.gate_proj", "model.layers.9.mlp.experts.63.gate_proj", "model.layers.9.mlp.experts.64.gate_proj", "model.layers.9.mlp.experts.65.gate_proj", "model.layers.9.mlp.experts.66.gate_proj", "model.layers.9.mlp.experts.67.gate_proj", "model.layers.9.mlp.experts.68.gate_proj", "model.layers.9.mlp.experts.69.gate_proj", "model.layers.9.mlp.experts.70.gate_proj", "model.layers.9.mlp.experts.71.gate_proj", "model.layers.9.mlp.experts.72.gate_proj", "model.layers.9.mlp.experts.73.gate_proj", "model.layers.9.mlp.experts.74.gate_proj", "model.layers.9.mlp.experts.75.gate_proj", "model.layers.9.mlp.experts.76.gate_proj", "model.layers.9.mlp.experts.77.gate_proj", "model.layers.9.mlp.experts.78.gate_proj", "model.layers.9.mlp.experts.79.gate_proj", "model.layers.9.mlp.experts.80.gate_proj", "model.layers.9.mlp.experts.81.gate_proj", "model.layers.9.mlp.experts.82.gate_proj", "model.layers.9.mlp.experts.83.gate_proj", "model.layers.9.mlp.experts.84.gate_proj", "model.layers.9.mlp.experts.85.gate_proj", "model.layers.9.mlp.experts.86.gate_proj", "model.layers.9.mlp.experts.87.gate_proj", "model.layers.9.mlp.experts.88.gate_proj", "model.layers.9.mlp.experts.89.gate_proj", "model.layers.9.mlp.experts.90.gate_proj", "model.layers.9.mlp.experts.91.gate_proj", "model.layers.9.mlp.experts.92.gate_proj", "model.layers.9.mlp.experts.93.gate_proj", "model.layers.9.mlp.experts.94.gate_proj", "model.layers.9.mlp.experts.95.gate_proj", "model.layers.9.mlp.experts.96.gate_proj", "model.layers.9.mlp.experts.97.gate_proj", "model.layers.9.mlp.experts.98.gate_proj", "model.layers.9.mlp.experts.99.gate_proj", "model.layers.9.mlp.experts.100.gate_proj", "model.layers.9.mlp.experts.101.gate_proj", "model.layers.9.mlp.experts.102.gate_proj", "model.layers.9.mlp.experts.103.gate_proj", "model.layers.9.mlp.experts.104.gate_proj", "model.layers.9.mlp.experts.105.gate_proj", "model.layers.9.mlp.experts.106.gate_proj", "model.layers.9.mlp.experts.107.gate_proj", "model.layers.9.mlp.experts.108.gate_proj", "model.layers.9.mlp.experts.109.gate_proj", "model.layers.9.mlp.experts.110.gate_proj", "model.layers.9.mlp.experts.111.gate_proj", "model.layers.9.mlp.experts.112.gate_proj", "model.layers.9.mlp.experts.113.gate_proj", "model.layers.9.mlp.experts.114.gate_proj", "model.layers.9.mlp.experts.115.gate_proj", "model.layers.9.mlp.experts.116.gate_proj", "model.layers.9.mlp.experts.117.gate_proj", "model.layers.9.mlp.experts.118.gate_proj", "model.layers.9.mlp.experts.119.gate_proj", "model.layers.9.mlp.experts.120.gate_proj", "model.layers.9.mlp.experts.121.gate_proj", "model.layers.9.mlp.experts.122.gate_proj", "model.layers.9.mlp.experts.123.gate_proj", "model.layers.9.mlp.experts.124.gate_proj", "model.layers.9.mlp.experts.125.gate_proj", "model.layers.9.mlp.experts.126.gate_proj", "model.layers.9.mlp.experts.127.gate_proj", "model.layers.9.mlp.experts.128.gate_proj", "model.layers.9.mlp.experts.129.gate_proj", "model.layers.9.mlp.experts.130.gate_proj", "model.layers.9.mlp.experts.131.gate_proj", "model.layers.9.mlp.experts.132.gate_proj", "model.layers.9.mlp.experts.133.gate_proj", "model.layers.9.mlp.experts.134.gate_proj", "model.layers.9.mlp.experts.135.gate_proj", "model.layers.9.mlp.experts.136.gate_proj", "model.layers.9.mlp.experts.137.gate_proj", "model.layers.9.mlp.experts.138.gate_proj", "model.layers.9.mlp.experts.139.gate_proj", "model.layers.9.mlp.experts.140.gate_proj", "model.layers.9.mlp.experts.141.gate_proj", "model.layers.9.mlp.experts.142.gate_proj", "model.layers.9.mlp.experts.143.gate_proj", "model.layers.9.mlp.experts.144.gate_proj", "model.layers.9.mlp.experts.145.gate_proj", "model.layers.9.mlp.experts.146.gate_proj", "model.layers.9.mlp.experts.147.gate_proj", "model.layers.9.mlp.experts.148.gate_proj", "model.layers.9.mlp.experts.149.gate_proj", "model.layers.9.mlp.experts.150.gate_proj", "model.layers.9.mlp.experts.151.gate_proj", "model.layers.9.mlp.experts.152.gate_proj", "model.layers.9.mlp.experts.153.gate_proj", "model.layers.9.mlp.experts.154.gate_proj", "model.layers.9.mlp.experts.155.gate_proj", "model.layers.9.mlp.experts.156.gate_proj", "model.layers.9.mlp.experts.157.gate_proj", "model.layers.9.mlp.experts.158.gate_proj", "model.layers.9.mlp.experts.159.gate_proj", "model.layers.9.mlp.experts.0.up_proj", "model.layers.9.mlp.experts.1.up_proj", "model.layers.9.mlp.experts.2.up_proj", "model.layers.9.mlp.experts.3.up_proj", "model.layers.9.mlp.experts.4.up_proj", "model.layers.9.mlp.experts.5.up_proj", "model.layers.9.mlp.experts.6.up_proj", "model.layers.9.mlp.experts.7.up_proj", "model.layers.9.mlp.experts.8.up_proj", "model.layers.9.mlp.experts.9.up_proj", "model.layers.9.mlp.experts.10.up_proj", "model.layers.9.mlp.experts.11.up_proj", "model.layers.9.mlp.experts.12.up_proj", "model.layers.9.mlp.experts.13.up_proj", "model.layers.9.mlp.experts.14.up_proj", "model.layers.9.mlp.experts.15.up_proj", "model.layers.9.mlp.experts.16.up_proj", "model.layers.9.mlp.experts.17.up_proj", "model.layers.9.mlp.experts.18.up_proj", "model.layers.9.mlp.experts.19.up_proj", "model.layers.9.mlp.experts.20.up_proj", "model.layers.9.mlp.experts.21.up_proj", "model.layers.9.mlp.experts.22.up_proj", "model.layers.9.mlp.experts.23.up_proj", "model.layers.9.mlp.experts.24.up_proj", "model.layers.9.mlp.experts.25.up_proj", "model.layers.9.mlp.experts.26.up_proj", "model.layers.9.mlp.experts.27.up_proj", "model.layers.9.mlp.experts.28.up_proj", "model.layers.9.mlp.experts.29.up_proj", "model.layers.9.mlp.experts.30.up_proj", "model.layers.9.mlp.experts.31.up_proj", "model.layers.9.mlp.experts.32.up_proj", "model.layers.9.mlp.experts.33.up_proj", "model.layers.9.mlp.experts.34.up_proj", "model.layers.9.mlp.experts.35.up_proj", "model.layers.9.mlp.experts.36.up_proj", "model.layers.9.mlp.experts.37.up_proj", "model.layers.9.mlp.experts.38.up_proj", "model.layers.9.mlp.experts.39.up_proj", "model.layers.9.mlp.experts.40.up_proj", "model.layers.9.mlp.experts.41.up_proj", "model.layers.9.mlp.experts.42.up_proj", "model.layers.9.mlp.experts.43.up_proj", "model.layers.9.mlp.experts.44.up_proj", "model.layers.9.mlp.experts.45.up_proj", "model.layers.9.mlp.experts.46.up_proj", "model.layers.9.mlp.experts.47.up_proj", "model.layers.9.mlp.experts.48.up_proj", "model.layers.9.mlp.experts.49.up_proj", "model.layers.9.mlp.experts.50.up_proj", "model.layers.9.mlp.experts.51.up_proj", "model.layers.9.mlp.experts.52.up_proj", "model.layers.9.mlp.experts.53.up_proj", "model.layers.9.mlp.experts.54.up_proj", "model.layers.9.mlp.experts.55.up_proj", "model.layers.9.mlp.experts.56.up_proj", "model.layers.9.mlp.experts.57.up_proj", "model.layers.9.mlp.experts.58.up_proj", "model.layers.9.mlp.experts.59.up_proj", "model.layers.9.mlp.experts.60.up_proj", "model.layers.9.mlp.experts.61.up_proj", "model.layers.9.mlp.experts.62.up_proj", "model.layers.9.mlp.experts.63.up_proj", "model.layers.9.mlp.experts.64.up_proj", "model.layers.9.mlp.experts.65.up_proj", "model.layers.9.mlp.experts.66.up_proj", "model.layers.9.mlp.experts.67.up_proj", "model.layers.9.mlp.experts.68.up_proj", "model.layers.9.mlp.experts.69.up_proj", "model.layers.9.mlp.experts.70.up_proj", "model.layers.9.mlp.experts.71.up_proj", "model.layers.9.mlp.experts.72.up_proj", "model.layers.9.mlp.experts.73.up_proj", "model.layers.9.mlp.experts.74.up_proj", "model.layers.9.mlp.experts.75.up_proj", "model.layers.9.mlp.experts.76.up_proj", "model.layers.9.mlp.experts.77.up_proj", "model.layers.9.mlp.experts.78.up_proj", "model.layers.9.mlp.experts.79.up_proj", "model.layers.9.mlp.experts.80.up_proj", "model.layers.9.mlp.experts.81.up_proj", "model.layers.9.mlp.experts.82.up_proj", "model.layers.9.mlp.experts.83.up_proj", "model.layers.9.mlp.experts.84.up_proj", "model.layers.9.mlp.experts.85.up_proj", "model.layers.9.mlp.experts.86.up_proj", "model.layers.9.mlp.experts.87.up_proj", "model.layers.9.mlp.experts.88.up_proj", "model.layers.9.mlp.experts.89.up_proj", "model.layers.9.mlp.experts.90.up_proj", "model.layers.9.mlp.experts.91.up_proj", "model.layers.9.mlp.experts.92.up_proj", "model.layers.9.mlp.experts.93.up_proj", "model.layers.9.mlp.experts.94.up_proj", "model.layers.9.mlp.experts.95.up_proj", "model.layers.9.mlp.experts.96.up_proj", "model.layers.9.mlp.experts.97.up_proj", "model.layers.9.mlp.experts.98.up_proj", "model.layers.9.mlp.experts.99.up_proj", "model.layers.9.mlp.experts.100.up_proj", "model.layers.9.mlp.experts.101.up_proj", "model.layers.9.mlp.experts.102.up_proj", "model.layers.9.mlp.experts.103.up_proj", "model.layers.9.mlp.experts.104.up_proj", "model.layers.9.mlp.experts.105.up_proj", "model.layers.9.mlp.experts.106.up_proj", "model.layers.9.mlp.experts.107.up_proj", "model.layers.9.mlp.experts.108.up_proj", "model.layers.9.mlp.experts.109.up_proj", "model.layers.9.mlp.experts.110.up_proj", "model.layers.9.mlp.experts.111.up_proj", "model.layers.9.mlp.experts.112.up_proj", "model.layers.9.mlp.experts.113.up_proj", "model.layers.9.mlp.experts.114.up_proj", "model.layers.9.mlp.experts.115.up_proj", "model.layers.9.mlp.experts.116.up_proj", "model.layers.9.mlp.experts.117.up_proj", "model.layers.9.mlp.experts.118.up_proj", "model.layers.9.mlp.experts.119.up_proj", "model.layers.9.mlp.experts.120.up_proj", "model.layers.9.mlp.experts.121.up_proj", "model.layers.9.mlp.experts.122.up_proj", "model.layers.9.mlp.experts.123.up_proj", "model.layers.9.mlp.experts.124.up_proj", "model.layers.9.mlp.experts.125.up_proj", "model.layers.9.mlp.experts.126.up_proj", "model.layers.9.mlp.experts.127.up_proj", "model.layers.9.mlp.experts.128.up_proj", "model.layers.9.mlp.experts.129.up_proj", "model.layers.9.mlp.experts.130.up_proj", "model.layers.9.mlp.experts.131.up_proj", "model.layers.9.mlp.experts.132.up_proj", "model.layers.9.mlp.experts.133.up_proj", "model.layers.9.mlp.experts.134.up_proj", "model.layers.9.mlp.experts.135.up_proj", "model.layers.9.mlp.experts.136.up_proj", "model.layers.9.mlp.experts.137.up_proj", "model.layers.9.mlp.experts.138.up_proj", "model.layers.9.mlp.experts.139.up_proj", "model.layers.9.mlp.experts.140.up_proj", "model.layers.9.mlp.experts.141.up_proj", "model.layers.9.mlp.experts.142.up_proj", "model.layers.9.mlp.experts.143.up_proj", "model.layers.9.mlp.experts.144.up_proj", "model.layers.9.mlp.experts.145.up_proj", "model.layers.9.mlp.experts.146.up_proj", "model.layers.9.mlp.experts.147.up_proj", "model.layers.9.mlp.experts.148.up_proj", "model.layers.9.mlp.experts.149.up_proj", "model.layers.9.mlp.experts.150.up_proj", "model.layers.9.mlp.experts.151.up_proj", "model.layers.9.mlp.experts.152.up_proj", "model.layers.9.mlp.experts.153.up_proj", "model.layers.9.mlp.experts.154.up_proj", "model.layers.9.mlp.experts.155.up_proj", "model.layers.9.mlp.experts.156.up_proj", "model.layers.9.mlp.experts.157.up_proj", "model.layers.9.mlp.experts.158.up_proj", "model.layers.9.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00046886708587408343, "dbits": 2516582400 }, { "dkld": -0.0003671668469905909, "dbits": 5033164800 }, { "dkld": -0.00043684765696526406, "dbits": 7549747200 }, { "dkld": -0.0003936843015253655, "dbits": 12582912000 } ] }, { "idx": 56, "layers": [ "model.layers.9.mlp.experts.0.down_proj", "model.layers.9.mlp.experts.1.down_proj", "model.layers.9.mlp.experts.2.down_proj", "model.layers.9.mlp.experts.3.down_proj", "model.layers.9.mlp.experts.4.down_proj", "model.layers.9.mlp.experts.5.down_proj", "model.layers.9.mlp.experts.6.down_proj", "model.layers.9.mlp.experts.7.down_proj", "model.layers.9.mlp.experts.8.down_proj", "model.layers.9.mlp.experts.9.down_proj", "model.layers.9.mlp.experts.10.down_proj", "model.layers.9.mlp.experts.11.down_proj", "model.layers.9.mlp.experts.12.down_proj", "model.layers.9.mlp.experts.13.down_proj", "model.layers.9.mlp.experts.14.down_proj", "model.layers.9.mlp.experts.15.down_proj", "model.layers.9.mlp.experts.16.down_proj", "model.layers.9.mlp.experts.17.down_proj", "model.layers.9.mlp.experts.18.down_proj", "model.layers.9.mlp.experts.19.down_proj", "model.layers.9.mlp.experts.20.down_proj", "model.layers.9.mlp.experts.21.down_proj", "model.layers.9.mlp.experts.22.down_proj", "model.layers.9.mlp.experts.23.down_proj", "model.layers.9.mlp.experts.24.down_proj", "model.layers.9.mlp.experts.25.down_proj", "model.layers.9.mlp.experts.26.down_proj", "model.layers.9.mlp.experts.27.down_proj", "model.layers.9.mlp.experts.28.down_proj", "model.layers.9.mlp.experts.29.down_proj", "model.layers.9.mlp.experts.30.down_proj", "model.layers.9.mlp.experts.31.down_proj", "model.layers.9.mlp.experts.32.down_proj", "model.layers.9.mlp.experts.33.down_proj", "model.layers.9.mlp.experts.34.down_proj", "model.layers.9.mlp.experts.35.down_proj", "model.layers.9.mlp.experts.36.down_proj", "model.layers.9.mlp.experts.37.down_proj", "model.layers.9.mlp.experts.38.down_proj", "model.layers.9.mlp.experts.39.down_proj", "model.layers.9.mlp.experts.40.down_proj", "model.layers.9.mlp.experts.41.down_proj", "model.layers.9.mlp.experts.42.down_proj", "model.layers.9.mlp.experts.43.down_proj", "model.layers.9.mlp.experts.44.down_proj", "model.layers.9.mlp.experts.45.down_proj", "model.layers.9.mlp.experts.46.down_proj", "model.layers.9.mlp.experts.47.down_proj", "model.layers.9.mlp.experts.48.down_proj", "model.layers.9.mlp.experts.49.down_proj", "model.layers.9.mlp.experts.50.down_proj", "model.layers.9.mlp.experts.51.down_proj", "model.layers.9.mlp.experts.52.down_proj", "model.layers.9.mlp.experts.53.down_proj", "model.layers.9.mlp.experts.54.down_proj", "model.layers.9.mlp.experts.55.down_proj", "model.layers.9.mlp.experts.56.down_proj", "model.layers.9.mlp.experts.57.down_proj", "model.layers.9.mlp.experts.58.down_proj", "model.layers.9.mlp.experts.59.down_proj", "model.layers.9.mlp.experts.60.down_proj", "model.layers.9.mlp.experts.61.down_proj", "model.layers.9.mlp.experts.62.down_proj", "model.layers.9.mlp.experts.63.down_proj", "model.layers.9.mlp.experts.64.down_proj", "model.layers.9.mlp.experts.65.down_proj", "model.layers.9.mlp.experts.66.down_proj", "model.layers.9.mlp.experts.67.down_proj", "model.layers.9.mlp.experts.68.down_proj", "model.layers.9.mlp.experts.69.down_proj", "model.layers.9.mlp.experts.70.down_proj", "model.layers.9.mlp.experts.71.down_proj", "model.layers.9.mlp.experts.72.down_proj", "model.layers.9.mlp.experts.73.down_proj", "model.layers.9.mlp.experts.74.down_proj", "model.layers.9.mlp.experts.75.down_proj", "model.layers.9.mlp.experts.76.down_proj", "model.layers.9.mlp.experts.77.down_proj", "model.layers.9.mlp.experts.78.down_proj", "model.layers.9.mlp.experts.79.down_proj", "model.layers.9.mlp.experts.80.down_proj", "model.layers.9.mlp.experts.81.down_proj", "model.layers.9.mlp.experts.82.down_proj", "model.layers.9.mlp.experts.83.down_proj", "model.layers.9.mlp.experts.84.down_proj", "model.layers.9.mlp.experts.85.down_proj", "model.layers.9.mlp.experts.86.down_proj", "model.layers.9.mlp.experts.87.down_proj", "model.layers.9.mlp.experts.88.down_proj", "model.layers.9.mlp.experts.89.down_proj", "model.layers.9.mlp.experts.90.down_proj", "model.layers.9.mlp.experts.91.down_proj", "model.layers.9.mlp.experts.92.down_proj", "model.layers.9.mlp.experts.93.down_proj", "model.layers.9.mlp.experts.94.down_proj", "model.layers.9.mlp.experts.95.down_proj", "model.layers.9.mlp.experts.96.down_proj", "model.layers.9.mlp.experts.97.down_proj", "model.layers.9.mlp.experts.98.down_proj", "model.layers.9.mlp.experts.99.down_proj", "model.layers.9.mlp.experts.100.down_proj", "model.layers.9.mlp.experts.101.down_proj", "model.layers.9.mlp.experts.102.down_proj", "model.layers.9.mlp.experts.103.down_proj", "model.layers.9.mlp.experts.104.down_proj", "model.layers.9.mlp.experts.105.down_proj", "model.layers.9.mlp.experts.106.down_proj", "model.layers.9.mlp.experts.107.down_proj", "model.layers.9.mlp.experts.108.down_proj", "model.layers.9.mlp.experts.109.down_proj", "model.layers.9.mlp.experts.110.down_proj", "model.layers.9.mlp.experts.111.down_proj", "model.layers.9.mlp.experts.112.down_proj", "model.layers.9.mlp.experts.113.down_proj", "model.layers.9.mlp.experts.114.down_proj", "model.layers.9.mlp.experts.115.down_proj", "model.layers.9.mlp.experts.116.down_proj", "model.layers.9.mlp.experts.117.down_proj", "model.layers.9.mlp.experts.118.down_proj", "model.layers.9.mlp.experts.119.down_proj", "model.layers.9.mlp.experts.120.down_proj", "model.layers.9.mlp.experts.121.down_proj", "model.layers.9.mlp.experts.122.down_proj", "model.layers.9.mlp.experts.123.down_proj", "model.layers.9.mlp.experts.124.down_proj", "model.layers.9.mlp.experts.125.down_proj", "model.layers.9.mlp.experts.126.down_proj", "model.layers.9.mlp.experts.127.down_proj", "model.layers.9.mlp.experts.128.down_proj", "model.layers.9.mlp.experts.129.down_proj", "model.layers.9.mlp.experts.130.down_proj", "model.layers.9.mlp.experts.131.down_proj", "model.layers.9.mlp.experts.132.down_proj", "model.layers.9.mlp.experts.133.down_proj", "model.layers.9.mlp.experts.134.down_proj", "model.layers.9.mlp.experts.135.down_proj", "model.layers.9.mlp.experts.136.down_proj", "model.layers.9.mlp.experts.137.down_proj", "model.layers.9.mlp.experts.138.down_proj", "model.layers.9.mlp.experts.139.down_proj", "model.layers.9.mlp.experts.140.down_proj", "model.layers.9.mlp.experts.141.down_proj", "model.layers.9.mlp.experts.142.down_proj", "model.layers.9.mlp.experts.143.down_proj", "model.layers.9.mlp.experts.144.down_proj", "model.layers.9.mlp.experts.145.down_proj", "model.layers.9.mlp.experts.146.down_proj", "model.layers.9.mlp.experts.147.down_proj", "model.layers.9.mlp.experts.148.down_proj", "model.layers.9.mlp.experts.149.down_proj", "model.layers.9.mlp.experts.150.down_proj", "model.layers.9.mlp.experts.151.down_proj", "model.layers.9.mlp.experts.152.down_proj", "model.layers.9.mlp.experts.153.down_proj", "model.layers.9.mlp.experts.154.down_proj", "model.layers.9.mlp.experts.155.down_proj", "model.layers.9.mlp.experts.156.down_proj", "model.layers.9.mlp.experts.157.down_proj", "model.layers.9.mlp.experts.158.down_proj", "model.layers.9.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002492424100637436, "dbits": 1258291200 }, { "dkld": -0.00031824707984924594, "dbits": 2516582400 }, { "dkld": -0.0004683179780840957, "dbits": 3774873600 }, { "dkld": -0.00036191903054715036, "dbits": 6291456000 } ] }, { "idx": 57, "layers": [ "model.layers.10.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002988244406879059, "dbits": 62914560 }, { "dkld": -0.00026354789733887274, "dbits": 125829120 }, { "dkld": -0.00017045317217707912, "dbits": 188743680 }, { "dkld": -0.00020934231579304574, "dbits": 314572800 } ] }, { "idx": 58, "layers": [ "model.layers.10.self_attn.k_proj", "model.layers.10.self_attn.v_proj" ], "candidates": [ { "dkld": -5.701296031475345e-05, "dbits": 10485760 }, { "dkld": 0.00030420888215303143, "dbits": 20971520 }, { "dkld": 8.737184107303342e-05, "dbits": 31457280 }, { "dkld": 8.425451815127771e-05, "dbits": 52428800 } ] }, { "idx": 59, "layers": [ "model.layers.10.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00033383760601281287, "dbits": 62914560 }, { "dkld": 3.0215829610824585e-05, "dbits": 125829120 }, { "dkld": 7.451642304658057e-05, "dbits": 188743680 }, { "dkld": 0.0001817207783460506, "dbits": 314572800 } ] }, { "idx": 60, "layers": [ "model.layers.10.mlp.shared_experts.gate_proj", "model.layers.10.mlp.shared_experts.up_proj", "model.layers.10.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -6.0782488435509596e-05, "dbits": 23592960 }, { "dkld": 8.529033511876782e-05, "dbits": 47185920 }, { "dkld": -4.52053733170088e-05, "dbits": 70778880 }, { "dkld": -6.789201870560646e-05, "dbits": 117964800 } ] }, { "idx": 61, "layers": [ "model.layers.10.mlp.experts.0.gate_proj", "model.layers.10.mlp.experts.1.gate_proj", "model.layers.10.mlp.experts.2.gate_proj", "model.layers.10.mlp.experts.3.gate_proj", "model.layers.10.mlp.experts.4.gate_proj", "model.layers.10.mlp.experts.5.gate_proj", "model.layers.10.mlp.experts.6.gate_proj", "model.layers.10.mlp.experts.7.gate_proj", "model.layers.10.mlp.experts.8.gate_proj", "model.layers.10.mlp.experts.9.gate_proj", "model.layers.10.mlp.experts.10.gate_proj", "model.layers.10.mlp.experts.11.gate_proj", "model.layers.10.mlp.experts.12.gate_proj", "model.layers.10.mlp.experts.13.gate_proj", "model.layers.10.mlp.experts.14.gate_proj", "model.layers.10.mlp.experts.15.gate_proj", "model.layers.10.mlp.experts.16.gate_proj", "model.layers.10.mlp.experts.17.gate_proj", "model.layers.10.mlp.experts.18.gate_proj", "model.layers.10.mlp.experts.19.gate_proj", "model.layers.10.mlp.experts.20.gate_proj", "model.layers.10.mlp.experts.21.gate_proj", "model.layers.10.mlp.experts.22.gate_proj", "model.layers.10.mlp.experts.23.gate_proj", "model.layers.10.mlp.experts.24.gate_proj", "model.layers.10.mlp.experts.25.gate_proj", "model.layers.10.mlp.experts.26.gate_proj", "model.layers.10.mlp.experts.27.gate_proj", "model.layers.10.mlp.experts.28.gate_proj", "model.layers.10.mlp.experts.29.gate_proj", "model.layers.10.mlp.experts.30.gate_proj", "model.layers.10.mlp.experts.31.gate_proj", "model.layers.10.mlp.experts.32.gate_proj", "model.layers.10.mlp.experts.33.gate_proj", "model.layers.10.mlp.experts.34.gate_proj", "model.layers.10.mlp.experts.35.gate_proj", "model.layers.10.mlp.experts.36.gate_proj", "model.layers.10.mlp.experts.37.gate_proj", "model.layers.10.mlp.experts.38.gate_proj", "model.layers.10.mlp.experts.39.gate_proj", "model.layers.10.mlp.experts.40.gate_proj", "model.layers.10.mlp.experts.41.gate_proj", "model.layers.10.mlp.experts.42.gate_proj", "model.layers.10.mlp.experts.43.gate_proj", "model.layers.10.mlp.experts.44.gate_proj", "model.layers.10.mlp.experts.45.gate_proj", "model.layers.10.mlp.experts.46.gate_proj", "model.layers.10.mlp.experts.47.gate_proj", "model.layers.10.mlp.experts.48.gate_proj", "model.layers.10.mlp.experts.49.gate_proj", "model.layers.10.mlp.experts.50.gate_proj", "model.layers.10.mlp.experts.51.gate_proj", "model.layers.10.mlp.experts.52.gate_proj", "model.layers.10.mlp.experts.53.gate_proj", "model.layers.10.mlp.experts.54.gate_proj", "model.layers.10.mlp.experts.55.gate_proj", "model.layers.10.mlp.experts.56.gate_proj", "model.layers.10.mlp.experts.57.gate_proj", "model.layers.10.mlp.experts.58.gate_proj", "model.layers.10.mlp.experts.59.gate_proj", "model.layers.10.mlp.experts.60.gate_proj", "model.layers.10.mlp.experts.61.gate_proj", "model.layers.10.mlp.experts.62.gate_proj", "model.layers.10.mlp.experts.63.gate_proj", "model.layers.10.mlp.experts.64.gate_proj", "model.layers.10.mlp.experts.65.gate_proj", "model.layers.10.mlp.experts.66.gate_proj", "model.layers.10.mlp.experts.67.gate_proj", "model.layers.10.mlp.experts.68.gate_proj", "model.layers.10.mlp.experts.69.gate_proj", "model.layers.10.mlp.experts.70.gate_proj", "model.layers.10.mlp.experts.71.gate_proj", "model.layers.10.mlp.experts.72.gate_proj", "model.layers.10.mlp.experts.73.gate_proj", "model.layers.10.mlp.experts.74.gate_proj", "model.layers.10.mlp.experts.75.gate_proj", "model.layers.10.mlp.experts.76.gate_proj", "model.layers.10.mlp.experts.77.gate_proj", "model.layers.10.mlp.experts.78.gate_proj", "model.layers.10.mlp.experts.79.gate_proj", "model.layers.10.mlp.experts.80.gate_proj", "model.layers.10.mlp.experts.81.gate_proj", "model.layers.10.mlp.experts.82.gate_proj", "model.layers.10.mlp.experts.83.gate_proj", "model.layers.10.mlp.experts.84.gate_proj", "model.layers.10.mlp.experts.85.gate_proj", "model.layers.10.mlp.experts.86.gate_proj", "model.layers.10.mlp.experts.87.gate_proj", "model.layers.10.mlp.experts.88.gate_proj", "model.layers.10.mlp.experts.89.gate_proj", "model.layers.10.mlp.experts.90.gate_proj", "model.layers.10.mlp.experts.91.gate_proj", "model.layers.10.mlp.experts.92.gate_proj", "model.layers.10.mlp.experts.93.gate_proj", "model.layers.10.mlp.experts.94.gate_proj", "model.layers.10.mlp.experts.95.gate_proj", "model.layers.10.mlp.experts.96.gate_proj", "model.layers.10.mlp.experts.97.gate_proj", "model.layers.10.mlp.experts.98.gate_proj", "model.layers.10.mlp.experts.99.gate_proj", "model.layers.10.mlp.experts.100.gate_proj", "model.layers.10.mlp.experts.101.gate_proj", "model.layers.10.mlp.experts.102.gate_proj", "model.layers.10.mlp.experts.103.gate_proj", "model.layers.10.mlp.experts.104.gate_proj", "model.layers.10.mlp.experts.105.gate_proj", "model.layers.10.mlp.experts.106.gate_proj", "model.layers.10.mlp.experts.107.gate_proj", "model.layers.10.mlp.experts.108.gate_proj", "model.layers.10.mlp.experts.109.gate_proj", "model.layers.10.mlp.experts.110.gate_proj", "model.layers.10.mlp.experts.111.gate_proj", "model.layers.10.mlp.experts.112.gate_proj", "model.layers.10.mlp.experts.113.gate_proj", "model.layers.10.mlp.experts.114.gate_proj", "model.layers.10.mlp.experts.115.gate_proj", "model.layers.10.mlp.experts.116.gate_proj", "model.layers.10.mlp.experts.117.gate_proj", "model.layers.10.mlp.experts.118.gate_proj", "model.layers.10.mlp.experts.119.gate_proj", "model.layers.10.mlp.experts.120.gate_proj", "model.layers.10.mlp.experts.121.gate_proj", "model.layers.10.mlp.experts.122.gate_proj", "model.layers.10.mlp.experts.123.gate_proj", "model.layers.10.mlp.experts.124.gate_proj", "model.layers.10.mlp.experts.125.gate_proj", "model.layers.10.mlp.experts.126.gate_proj", "model.layers.10.mlp.experts.127.gate_proj", "model.layers.10.mlp.experts.128.gate_proj", "model.layers.10.mlp.experts.129.gate_proj", "model.layers.10.mlp.experts.130.gate_proj", "model.layers.10.mlp.experts.131.gate_proj", "model.layers.10.mlp.experts.132.gate_proj", "model.layers.10.mlp.experts.133.gate_proj", "model.layers.10.mlp.experts.134.gate_proj", "model.layers.10.mlp.experts.135.gate_proj", "model.layers.10.mlp.experts.136.gate_proj", "model.layers.10.mlp.experts.137.gate_proj", "model.layers.10.mlp.experts.138.gate_proj", "model.layers.10.mlp.experts.139.gate_proj", "model.layers.10.mlp.experts.140.gate_proj", "model.layers.10.mlp.experts.141.gate_proj", "model.layers.10.mlp.experts.142.gate_proj", "model.layers.10.mlp.experts.143.gate_proj", "model.layers.10.mlp.experts.144.gate_proj", "model.layers.10.mlp.experts.145.gate_proj", "model.layers.10.mlp.experts.146.gate_proj", "model.layers.10.mlp.experts.147.gate_proj", "model.layers.10.mlp.experts.148.gate_proj", "model.layers.10.mlp.experts.149.gate_proj", "model.layers.10.mlp.experts.150.gate_proj", "model.layers.10.mlp.experts.151.gate_proj", "model.layers.10.mlp.experts.152.gate_proj", "model.layers.10.mlp.experts.153.gate_proj", "model.layers.10.mlp.experts.154.gate_proj", "model.layers.10.mlp.experts.155.gate_proj", "model.layers.10.mlp.experts.156.gate_proj", "model.layers.10.mlp.experts.157.gate_proj", "model.layers.10.mlp.experts.158.gate_proj", "model.layers.10.mlp.experts.159.gate_proj", "model.layers.10.mlp.experts.0.up_proj", "model.layers.10.mlp.experts.1.up_proj", "model.layers.10.mlp.experts.2.up_proj", "model.layers.10.mlp.experts.3.up_proj", "model.layers.10.mlp.experts.4.up_proj", "model.layers.10.mlp.experts.5.up_proj", "model.layers.10.mlp.experts.6.up_proj", "model.layers.10.mlp.experts.7.up_proj", "model.layers.10.mlp.experts.8.up_proj", "model.layers.10.mlp.experts.9.up_proj", "model.layers.10.mlp.experts.10.up_proj", "model.layers.10.mlp.experts.11.up_proj", "model.layers.10.mlp.experts.12.up_proj", "model.layers.10.mlp.experts.13.up_proj", "model.layers.10.mlp.experts.14.up_proj", "model.layers.10.mlp.experts.15.up_proj", "model.layers.10.mlp.experts.16.up_proj", "model.layers.10.mlp.experts.17.up_proj", "model.layers.10.mlp.experts.18.up_proj", "model.layers.10.mlp.experts.19.up_proj", "model.layers.10.mlp.experts.20.up_proj", "model.layers.10.mlp.experts.21.up_proj", "model.layers.10.mlp.experts.22.up_proj", "model.layers.10.mlp.experts.23.up_proj", "model.layers.10.mlp.experts.24.up_proj", "model.layers.10.mlp.experts.25.up_proj", "model.layers.10.mlp.experts.26.up_proj", "model.layers.10.mlp.experts.27.up_proj", "model.layers.10.mlp.experts.28.up_proj", "model.layers.10.mlp.experts.29.up_proj", "model.layers.10.mlp.experts.30.up_proj", "model.layers.10.mlp.experts.31.up_proj", "model.layers.10.mlp.experts.32.up_proj", "model.layers.10.mlp.experts.33.up_proj", "model.layers.10.mlp.experts.34.up_proj", "model.layers.10.mlp.experts.35.up_proj", "model.layers.10.mlp.experts.36.up_proj", "model.layers.10.mlp.experts.37.up_proj", "model.layers.10.mlp.experts.38.up_proj", "model.layers.10.mlp.experts.39.up_proj", "model.layers.10.mlp.experts.40.up_proj", "model.layers.10.mlp.experts.41.up_proj", "model.layers.10.mlp.experts.42.up_proj", "model.layers.10.mlp.experts.43.up_proj", "model.layers.10.mlp.experts.44.up_proj", "model.layers.10.mlp.experts.45.up_proj", "model.layers.10.mlp.experts.46.up_proj", "model.layers.10.mlp.experts.47.up_proj", "model.layers.10.mlp.experts.48.up_proj", "model.layers.10.mlp.experts.49.up_proj", "model.layers.10.mlp.experts.50.up_proj", "model.layers.10.mlp.experts.51.up_proj", "model.layers.10.mlp.experts.52.up_proj", "model.layers.10.mlp.experts.53.up_proj", "model.layers.10.mlp.experts.54.up_proj", "model.layers.10.mlp.experts.55.up_proj", "model.layers.10.mlp.experts.56.up_proj", "model.layers.10.mlp.experts.57.up_proj", "model.layers.10.mlp.experts.58.up_proj", "model.layers.10.mlp.experts.59.up_proj", "model.layers.10.mlp.experts.60.up_proj", "model.layers.10.mlp.experts.61.up_proj", "model.layers.10.mlp.experts.62.up_proj", "model.layers.10.mlp.experts.63.up_proj", "model.layers.10.mlp.experts.64.up_proj", "model.layers.10.mlp.experts.65.up_proj", "model.layers.10.mlp.experts.66.up_proj", "model.layers.10.mlp.experts.67.up_proj", "model.layers.10.mlp.experts.68.up_proj", "model.layers.10.mlp.experts.69.up_proj", "model.layers.10.mlp.experts.70.up_proj", "model.layers.10.mlp.experts.71.up_proj", "model.layers.10.mlp.experts.72.up_proj", "model.layers.10.mlp.experts.73.up_proj", "model.layers.10.mlp.experts.74.up_proj", "model.layers.10.mlp.experts.75.up_proj", "model.layers.10.mlp.experts.76.up_proj", "model.layers.10.mlp.experts.77.up_proj", "model.layers.10.mlp.experts.78.up_proj", "model.layers.10.mlp.experts.79.up_proj", "model.layers.10.mlp.experts.80.up_proj", "model.layers.10.mlp.experts.81.up_proj", "model.layers.10.mlp.experts.82.up_proj", "model.layers.10.mlp.experts.83.up_proj", "model.layers.10.mlp.experts.84.up_proj", "model.layers.10.mlp.experts.85.up_proj", "model.layers.10.mlp.experts.86.up_proj", "model.layers.10.mlp.experts.87.up_proj", "model.layers.10.mlp.experts.88.up_proj", "model.layers.10.mlp.experts.89.up_proj", "model.layers.10.mlp.experts.90.up_proj", "model.layers.10.mlp.experts.91.up_proj", "model.layers.10.mlp.experts.92.up_proj", "model.layers.10.mlp.experts.93.up_proj", "model.layers.10.mlp.experts.94.up_proj", "model.layers.10.mlp.experts.95.up_proj", "model.layers.10.mlp.experts.96.up_proj", "model.layers.10.mlp.experts.97.up_proj", "model.layers.10.mlp.experts.98.up_proj", "model.layers.10.mlp.experts.99.up_proj", "model.layers.10.mlp.experts.100.up_proj", "model.layers.10.mlp.experts.101.up_proj", "model.layers.10.mlp.experts.102.up_proj", "model.layers.10.mlp.experts.103.up_proj", "model.layers.10.mlp.experts.104.up_proj", "model.layers.10.mlp.experts.105.up_proj", "model.layers.10.mlp.experts.106.up_proj", "model.layers.10.mlp.experts.107.up_proj", "model.layers.10.mlp.experts.108.up_proj", "model.layers.10.mlp.experts.109.up_proj", "model.layers.10.mlp.experts.110.up_proj", "model.layers.10.mlp.experts.111.up_proj", "model.layers.10.mlp.experts.112.up_proj", "model.layers.10.mlp.experts.113.up_proj", "model.layers.10.mlp.experts.114.up_proj", "model.layers.10.mlp.experts.115.up_proj", "model.layers.10.mlp.experts.116.up_proj", "model.layers.10.mlp.experts.117.up_proj", "model.layers.10.mlp.experts.118.up_proj", "model.layers.10.mlp.experts.119.up_proj", "model.layers.10.mlp.experts.120.up_proj", "model.layers.10.mlp.experts.121.up_proj", "model.layers.10.mlp.experts.122.up_proj", "model.layers.10.mlp.experts.123.up_proj", "model.layers.10.mlp.experts.124.up_proj", "model.layers.10.mlp.experts.125.up_proj", "model.layers.10.mlp.experts.126.up_proj", "model.layers.10.mlp.experts.127.up_proj", "model.layers.10.mlp.experts.128.up_proj", "model.layers.10.mlp.experts.129.up_proj", "model.layers.10.mlp.experts.130.up_proj", "model.layers.10.mlp.experts.131.up_proj", "model.layers.10.mlp.experts.132.up_proj", "model.layers.10.mlp.experts.133.up_proj", "model.layers.10.mlp.experts.134.up_proj", "model.layers.10.mlp.experts.135.up_proj", "model.layers.10.mlp.experts.136.up_proj", "model.layers.10.mlp.experts.137.up_proj", "model.layers.10.mlp.experts.138.up_proj", "model.layers.10.mlp.experts.139.up_proj", "model.layers.10.mlp.experts.140.up_proj", "model.layers.10.mlp.experts.141.up_proj", "model.layers.10.mlp.experts.142.up_proj", "model.layers.10.mlp.experts.143.up_proj", "model.layers.10.mlp.experts.144.up_proj", "model.layers.10.mlp.experts.145.up_proj", "model.layers.10.mlp.experts.146.up_proj", "model.layers.10.mlp.experts.147.up_proj", "model.layers.10.mlp.experts.148.up_proj", "model.layers.10.mlp.experts.149.up_proj", "model.layers.10.mlp.experts.150.up_proj", "model.layers.10.mlp.experts.151.up_proj", "model.layers.10.mlp.experts.152.up_proj", "model.layers.10.mlp.experts.153.up_proj", "model.layers.10.mlp.experts.154.up_proj", "model.layers.10.mlp.experts.155.up_proj", "model.layers.10.mlp.experts.156.up_proj", "model.layers.10.mlp.experts.157.up_proj", "model.layers.10.mlp.experts.158.up_proj", "model.layers.10.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0001965534873306779, "dbits": 2516582400 }, { "dkld": -0.00021028779447079815, "dbits": 5033164800 }, { "dkld": -0.00020470218732954182, "dbits": 7549747200 }, { "dkld": -0.0002858545631170273, "dbits": 12582912000 } ] }, { "idx": 62, "layers": [ "model.layers.10.mlp.experts.0.down_proj", "model.layers.10.mlp.experts.1.down_proj", "model.layers.10.mlp.experts.2.down_proj", "model.layers.10.mlp.experts.3.down_proj", "model.layers.10.mlp.experts.4.down_proj", "model.layers.10.mlp.experts.5.down_proj", "model.layers.10.mlp.experts.6.down_proj", "model.layers.10.mlp.experts.7.down_proj", "model.layers.10.mlp.experts.8.down_proj", "model.layers.10.mlp.experts.9.down_proj", "model.layers.10.mlp.experts.10.down_proj", "model.layers.10.mlp.experts.11.down_proj", "model.layers.10.mlp.experts.12.down_proj", "model.layers.10.mlp.experts.13.down_proj", "model.layers.10.mlp.experts.14.down_proj", "model.layers.10.mlp.experts.15.down_proj", "model.layers.10.mlp.experts.16.down_proj", "model.layers.10.mlp.experts.17.down_proj", "model.layers.10.mlp.experts.18.down_proj", "model.layers.10.mlp.experts.19.down_proj", "model.layers.10.mlp.experts.20.down_proj", "model.layers.10.mlp.experts.21.down_proj", "model.layers.10.mlp.experts.22.down_proj", "model.layers.10.mlp.experts.23.down_proj", "model.layers.10.mlp.experts.24.down_proj", "model.layers.10.mlp.experts.25.down_proj", "model.layers.10.mlp.experts.26.down_proj", "model.layers.10.mlp.experts.27.down_proj", "model.layers.10.mlp.experts.28.down_proj", "model.layers.10.mlp.experts.29.down_proj", "model.layers.10.mlp.experts.30.down_proj", "model.layers.10.mlp.experts.31.down_proj", "model.layers.10.mlp.experts.32.down_proj", "model.layers.10.mlp.experts.33.down_proj", "model.layers.10.mlp.experts.34.down_proj", "model.layers.10.mlp.experts.35.down_proj", "model.layers.10.mlp.experts.36.down_proj", "model.layers.10.mlp.experts.37.down_proj", "model.layers.10.mlp.experts.38.down_proj", "model.layers.10.mlp.experts.39.down_proj", "model.layers.10.mlp.experts.40.down_proj", "model.layers.10.mlp.experts.41.down_proj", "model.layers.10.mlp.experts.42.down_proj", "model.layers.10.mlp.experts.43.down_proj", "model.layers.10.mlp.experts.44.down_proj", "model.layers.10.mlp.experts.45.down_proj", "model.layers.10.mlp.experts.46.down_proj", "model.layers.10.mlp.experts.47.down_proj", "model.layers.10.mlp.experts.48.down_proj", "model.layers.10.mlp.experts.49.down_proj", "model.layers.10.mlp.experts.50.down_proj", "model.layers.10.mlp.experts.51.down_proj", "model.layers.10.mlp.experts.52.down_proj", "model.layers.10.mlp.experts.53.down_proj", "model.layers.10.mlp.experts.54.down_proj", "model.layers.10.mlp.experts.55.down_proj", "model.layers.10.mlp.experts.56.down_proj", "model.layers.10.mlp.experts.57.down_proj", "model.layers.10.mlp.experts.58.down_proj", "model.layers.10.mlp.experts.59.down_proj", "model.layers.10.mlp.experts.60.down_proj", "model.layers.10.mlp.experts.61.down_proj", "model.layers.10.mlp.experts.62.down_proj", "model.layers.10.mlp.experts.63.down_proj", "model.layers.10.mlp.experts.64.down_proj", "model.layers.10.mlp.experts.65.down_proj", "model.layers.10.mlp.experts.66.down_proj", "model.layers.10.mlp.experts.67.down_proj", "model.layers.10.mlp.experts.68.down_proj", "model.layers.10.mlp.experts.69.down_proj", "model.layers.10.mlp.experts.70.down_proj", "model.layers.10.mlp.experts.71.down_proj", "model.layers.10.mlp.experts.72.down_proj", "model.layers.10.mlp.experts.73.down_proj", "model.layers.10.mlp.experts.74.down_proj", "model.layers.10.mlp.experts.75.down_proj", "model.layers.10.mlp.experts.76.down_proj", "model.layers.10.mlp.experts.77.down_proj", "model.layers.10.mlp.experts.78.down_proj", "model.layers.10.mlp.experts.79.down_proj", "model.layers.10.mlp.experts.80.down_proj", "model.layers.10.mlp.experts.81.down_proj", "model.layers.10.mlp.experts.82.down_proj", "model.layers.10.mlp.experts.83.down_proj", "model.layers.10.mlp.experts.84.down_proj", "model.layers.10.mlp.experts.85.down_proj", "model.layers.10.mlp.experts.86.down_proj", "model.layers.10.mlp.experts.87.down_proj", "model.layers.10.mlp.experts.88.down_proj", "model.layers.10.mlp.experts.89.down_proj", "model.layers.10.mlp.experts.90.down_proj", "model.layers.10.mlp.experts.91.down_proj", "model.layers.10.mlp.experts.92.down_proj", "model.layers.10.mlp.experts.93.down_proj", "model.layers.10.mlp.experts.94.down_proj", "model.layers.10.mlp.experts.95.down_proj", "model.layers.10.mlp.experts.96.down_proj", "model.layers.10.mlp.experts.97.down_proj", "model.layers.10.mlp.experts.98.down_proj", "model.layers.10.mlp.experts.99.down_proj", "model.layers.10.mlp.experts.100.down_proj", "model.layers.10.mlp.experts.101.down_proj", "model.layers.10.mlp.experts.102.down_proj", "model.layers.10.mlp.experts.103.down_proj", "model.layers.10.mlp.experts.104.down_proj", "model.layers.10.mlp.experts.105.down_proj", "model.layers.10.mlp.experts.106.down_proj", "model.layers.10.mlp.experts.107.down_proj", "model.layers.10.mlp.experts.108.down_proj", "model.layers.10.mlp.experts.109.down_proj", "model.layers.10.mlp.experts.110.down_proj", "model.layers.10.mlp.experts.111.down_proj", "model.layers.10.mlp.experts.112.down_proj", "model.layers.10.mlp.experts.113.down_proj", "model.layers.10.mlp.experts.114.down_proj", "model.layers.10.mlp.experts.115.down_proj", "model.layers.10.mlp.experts.116.down_proj", "model.layers.10.mlp.experts.117.down_proj", "model.layers.10.mlp.experts.118.down_proj", "model.layers.10.mlp.experts.119.down_proj", "model.layers.10.mlp.experts.120.down_proj", "model.layers.10.mlp.experts.121.down_proj", "model.layers.10.mlp.experts.122.down_proj", "model.layers.10.mlp.experts.123.down_proj", "model.layers.10.mlp.experts.124.down_proj", "model.layers.10.mlp.experts.125.down_proj", "model.layers.10.mlp.experts.126.down_proj", "model.layers.10.mlp.experts.127.down_proj", "model.layers.10.mlp.experts.128.down_proj", "model.layers.10.mlp.experts.129.down_proj", "model.layers.10.mlp.experts.130.down_proj", "model.layers.10.mlp.experts.131.down_proj", "model.layers.10.mlp.experts.132.down_proj", "model.layers.10.mlp.experts.133.down_proj", "model.layers.10.mlp.experts.134.down_proj", "model.layers.10.mlp.experts.135.down_proj", "model.layers.10.mlp.experts.136.down_proj", "model.layers.10.mlp.experts.137.down_proj", "model.layers.10.mlp.experts.138.down_proj", "model.layers.10.mlp.experts.139.down_proj", "model.layers.10.mlp.experts.140.down_proj", "model.layers.10.mlp.experts.141.down_proj", "model.layers.10.mlp.experts.142.down_proj", "model.layers.10.mlp.experts.143.down_proj", "model.layers.10.mlp.experts.144.down_proj", "model.layers.10.mlp.experts.145.down_proj", "model.layers.10.mlp.experts.146.down_proj", "model.layers.10.mlp.experts.147.down_proj", "model.layers.10.mlp.experts.148.down_proj", "model.layers.10.mlp.experts.149.down_proj", "model.layers.10.mlp.experts.150.down_proj", "model.layers.10.mlp.experts.151.down_proj", "model.layers.10.mlp.experts.152.down_proj", "model.layers.10.mlp.experts.153.down_proj", "model.layers.10.mlp.experts.154.down_proj", "model.layers.10.mlp.experts.155.down_proj", "model.layers.10.mlp.experts.156.down_proj", "model.layers.10.mlp.experts.157.down_proj", "model.layers.10.mlp.experts.158.down_proj", "model.layers.10.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0001820480450987899, "dbits": 1258291200 }, { "dkld": -0.00037111295387148857, "dbits": 2516582400 }, { "dkld": -0.0002446291968226516, "dbits": 3774873600 }, { "dkld": -0.0002351229079067818, "dbits": 6291456000 } ] }, { "idx": 63, "layers": [ "model.layers.11.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00010074432939290168, "dbits": 62914560 }, { "dkld": -0.0001024128869175911, "dbits": 125829120 }, { "dkld": 5.1767285913217886e-05, "dbits": 188743680 }, { "dkld": 3.4578517079345006e-05, "dbits": 314572800 } ] }, { "idx": 64, "layers": [ "model.layers.11.self_attn.k_proj", "model.layers.11.self_attn.v_proj" ], "candidates": [ { "dkld": 0.001218495517969126, "dbits": 10485760 }, { "dkld": 0.0012320306152105304, "dbits": 20971520 }, { "dkld": 0.0011404495686292565, "dbits": 31457280 }, { "dkld": 0.0009464699774980462, "dbits": 52428800 } ] }, { "idx": 65, "layers": [ "model.layers.11.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010876026004552897, "dbits": 62914560 }, { "dkld": -0.001741608045995241, "dbits": 125829120 }, { "dkld": -0.001661501452326783, "dbits": 188743680 }, { "dkld": -0.001530532352626332, "dbits": 314572800 } ] }, { "idx": 66, "layers": [ "model.layers.11.mlp.shared_experts.gate_proj", "model.layers.11.mlp.shared_experts.up_proj", "model.layers.11.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00045606531202793676, "dbits": 23592960 }, { "dkld": -0.0004361893981695286, "dbits": 47185920 }, { "dkld": -0.00031036036089063246, "dbits": 70778880 }, { "dkld": -0.00025790026411415257, "dbits": 117964800 } ] }, { "idx": 67, "layers": [ "model.layers.11.mlp.experts.0.gate_proj", "model.layers.11.mlp.experts.1.gate_proj", "model.layers.11.mlp.experts.2.gate_proj", "model.layers.11.mlp.experts.3.gate_proj", "model.layers.11.mlp.experts.4.gate_proj", "model.layers.11.mlp.experts.5.gate_proj", "model.layers.11.mlp.experts.6.gate_proj", "model.layers.11.mlp.experts.7.gate_proj", "model.layers.11.mlp.experts.8.gate_proj", "model.layers.11.mlp.experts.9.gate_proj", "model.layers.11.mlp.experts.10.gate_proj", "model.layers.11.mlp.experts.11.gate_proj", "model.layers.11.mlp.experts.12.gate_proj", "model.layers.11.mlp.experts.13.gate_proj", "model.layers.11.mlp.experts.14.gate_proj", "model.layers.11.mlp.experts.15.gate_proj", "model.layers.11.mlp.experts.16.gate_proj", "model.layers.11.mlp.experts.17.gate_proj", "model.layers.11.mlp.experts.18.gate_proj", "model.layers.11.mlp.experts.19.gate_proj", "model.layers.11.mlp.experts.20.gate_proj", "model.layers.11.mlp.experts.21.gate_proj", "model.layers.11.mlp.experts.22.gate_proj", "model.layers.11.mlp.experts.23.gate_proj", "model.layers.11.mlp.experts.24.gate_proj", "model.layers.11.mlp.experts.25.gate_proj", "model.layers.11.mlp.experts.26.gate_proj", "model.layers.11.mlp.experts.27.gate_proj", "model.layers.11.mlp.experts.28.gate_proj", "model.layers.11.mlp.experts.29.gate_proj", "model.layers.11.mlp.experts.30.gate_proj", "model.layers.11.mlp.experts.31.gate_proj", "model.layers.11.mlp.experts.32.gate_proj", "model.layers.11.mlp.experts.33.gate_proj", "model.layers.11.mlp.experts.34.gate_proj", "model.layers.11.mlp.experts.35.gate_proj", "model.layers.11.mlp.experts.36.gate_proj", "model.layers.11.mlp.experts.37.gate_proj", "model.layers.11.mlp.experts.38.gate_proj", "model.layers.11.mlp.experts.39.gate_proj", "model.layers.11.mlp.experts.40.gate_proj", "model.layers.11.mlp.experts.41.gate_proj", "model.layers.11.mlp.experts.42.gate_proj", "model.layers.11.mlp.experts.43.gate_proj", "model.layers.11.mlp.experts.44.gate_proj", "model.layers.11.mlp.experts.45.gate_proj", "model.layers.11.mlp.experts.46.gate_proj", "model.layers.11.mlp.experts.47.gate_proj", "model.layers.11.mlp.experts.48.gate_proj", "model.layers.11.mlp.experts.49.gate_proj", "model.layers.11.mlp.experts.50.gate_proj", "model.layers.11.mlp.experts.51.gate_proj", "model.layers.11.mlp.experts.52.gate_proj", "model.layers.11.mlp.experts.53.gate_proj", "model.layers.11.mlp.experts.54.gate_proj", "model.layers.11.mlp.experts.55.gate_proj", "model.layers.11.mlp.experts.56.gate_proj", "model.layers.11.mlp.experts.57.gate_proj", "model.layers.11.mlp.experts.58.gate_proj", "model.layers.11.mlp.experts.59.gate_proj", "model.layers.11.mlp.experts.60.gate_proj", "model.layers.11.mlp.experts.61.gate_proj", "model.layers.11.mlp.experts.62.gate_proj", "model.layers.11.mlp.experts.63.gate_proj", "model.layers.11.mlp.experts.64.gate_proj", "model.layers.11.mlp.experts.65.gate_proj", "model.layers.11.mlp.experts.66.gate_proj", "model.layers.11.mlp.experts.67.gate_proj", "model.layers.11.mlp.experts.68.gate_proj", "model.layers.11.mlp.experts.69.gate_proj", "model.layers.11.mlp.experts.70.gate_proj", "model.layers.11.mlp.experts.71.gate_proj", "model.layers.11.mlp.experts.72.gate_proj", "model.layers.11.mlp.experts.73.gate_proj", "model.layers.11.mlp.experts.74.gate_proj", "model.layers.11.mlp.experts.75.gate_proj", "model.layers.11.mlp.experts.76.gate_proj", "model.layers.11.mlp.experts.77.gate_proj", "model.layers.11.mlp.experts.78.gate_proj", "model.layers.11.mlp.experts.79.gate_proj", "model.layers.11.mlp.experts.80.gate_proj", "model.layers.11.mlp.experts.81.gate_proj", "model.layers.11.mlp.experts.82.gate_proj", "model.layers.11.mlp.experts.83.gate_proj", "model.layers.11.mlp.experts.84.gate_proj", "model.layers.11.mlp.experts.85.gate_proj", "model.layers.11.mlp.experts.86.gate_proj", "model.layers.11.mlp.experts.87.gate_proj", "model.layers.11.mlp.experts.88.gate_proj", "model.layers.11.mlp.experts.89.gate_proj", "model.layers.11.mlp.experts.90.gate_proj", "model.layers.11.mlp.experts.91.gate_proj", "model.layers.11.mlp.experts.92.gate_proj", "model.layers.11.mlp.experts.93.gate_proj", "model.layers.11.mlp.experts.94.gate_proj", "model.layers.11.mlp.experts.95.gate_proj", "model.layers.11.mlp.experts.96.gate_proj", "model.layers.11.mlp.experts.97.gate_proj", "model.layers.11.mlp.experts.98.gate_proj", "model.layers.11.mlp.experts.99.gate_proj", "model.layers.11.mlp.experts.100.gate_proj", "model.layers.11.mlp.experts.101.gate_proj", "model.layers.11.mlp.experts.102.gate_proj", "model.layers.11.mlp.experts.103.gate_proj", "model.layers.11.mlp.experts.104.gate_proj", "model.layers.11.mlp.experts.105.gate_proj", "model.layers.11.mlp.experts.106.gate_proj", "model.layers.11.mlp.experts.107.gate_proj", "model.layers.11.mlp.experts.108.gate_proj", "model.layers.11.mlp.experts.109.gate_proj", "model.layers.11.mlp.experts.110.gate_proj", "model.layers.11.mlp.experts.111.gate_proj", "model.layers.11.mlp.experts.112.gate_proj", "model.layers.11.mlp.experts.113.gate_proj", "model.layers.11.mlp.experts.114.gate_proj", "model.layers.11.mlp.experts.115.gate_proj", "model.layers.11.mlp.experts.116.gate_proj", "model.layers.11.mlp.experts.117.gate_proj", "model.layers.11.mlp.experts.118.gate_proj", "model.layers.11.mlp.experts.119.gate_proj", "model.layers.11.mlp.experts.120.gate_proj", "model.layers.11.mlp.experts.121.gate_proj", "model.layers.11.mlp.experts.122.gate_proj", "model.layers.11.mlp.experts.123.gate_proj", "model.layers.11.mlp.experts.124.gate_proj", "model.layers.11.mlp.experts.125.gate_proj", "model.layers.11.mlp.experts.126.gate_proj", "model.layers.11.mlp.experts.127.gate_proj", "model.layers.11.mlp.experts.128.gate_proj", "model.layers.11.mlp.experts.129.gate_proj", "model.layers.11.mlp.experts.130.gate_proj", "model.layers.11.mlp.experts.131.gate_proj", "model.layers.11.mlp.experts.132.gate_proj", "model.layers.11.mlp.experts.133.gate_proj", "model.layers.11.mlp.experts.134.gate_proj", "model.layers.11.mlp.experts.135.gate_proj", "model.layers.11.mlp.experts.136.gate_proj", "model.layers.11.mlp.experts.137.gate_proj", "model.layers.11.mlp.experts.138.gate_proj", "model.layers.11.mlp.experts.139.gate_proj", "model.layers.11.mlp.experts.140.gate_proj", "model.layers.11.mlp.experts.141.gate_proj", "model.layers.11.mlp.experts.142.gate_proj", "model.layers.11.mlp.experts.143.gate_proj", "model.layers.11.mlp.experts.144.gate_proj", "model.layers.11.mlp.experts.145.gate_proj", "model.layers.11.mlp.experts.146.gate_proj", "model.layers.11.mlp.experts.147.gate_proj", "model.layers.11.mlp.experts.148.gate_proj", "model.layers.11.mlp.experts.149.gate_proj", "model.layers.11.mlp.experts.150.gate_proj", "model.layers.11.mlp.experts.151.gate_proj", "model.layers.11.mlp.experts.152.gate_proj", "model.layers.11.mlp.experts.153.gate_proj", "model.layers.11.mlp.experts.154.gate_proj", "model.layers.11.mlp.experts.155.gate_proj", "model.layers.11.mlp.experts.156.gate_proj", "model.layers.11.mlp.experts.157.gate_proj", "model.layers.11.mlp.experts.158.gate_proj", "model.layers.11.mlp.experts.159.gate_proj", "model.layers.11.mlp.experts.0.up_proj", "model.layers.11.mlp.experts.1.up_proj", "model.layers.11.mlp.experts.2.up_proj", "model.layers.11.mlp.experts.3.up_proj", "model.layers.11.mlp.experts.4.up_proj", "model.layers.11.mlp.experts.5.up_proj", "model.layers.11.mlp.experts.6.up_proj", "model.layers.11.mlp.experts.7.up_proj", "model.layers.11.mlp.experts.8.up_proj", "model.layers.11.mlp.experts.9.up_proj", "model.layers.11.mlp.experts.10.up_proj", "model.layers.11.mlp.experts.11.up_proj", "model.layers.11.mlp.experts.12.up_proj", "model.layers.11.mlp.experts.13.up_proj", "model.layers.11.mlp.experts.14.up_proj", "model.layers.11.mlp.experts.15.up_proj", "model.layers.11.mlp.experts.16.up_proj", "model.layers.11.mlp.experts.17.up_proj", "model.layers.11.mlp.experts.18.up_proj", "model.layers.11.mlp.experts.19.up_proj", "model.layers.11.mlp.experts.20.up_proj", "model.layers.11.mlp.experts.21.up_proj", "model.layers.11.mlp.experts.22.up_proj", "model.layers.11.mlp.experts.23.up_proj", "model.layers.11.mlp.experts.24.up_proj", "model.layers.11.mlp.experts.25.up_proj", "model.layers.11.mlp.experts.26.up_proj", "model.layers.11.mlp.experts.27.up_proj", "model.layers.11.mlp.experts.28.up_proj", "model.layers.11.mlp.experts.29.up_proj", "model.layers.11.mlp.experts.30.up_proj", "model.layers.11.mlp.experts.31.up_proj", "model.layers.11.mlp.experts.32.up_proj", "model.layers.11.mlp.experts.33.up_proj", "model.layers.11.mlp.experts.34.up_proj", "model.layers.11.mlp.experts.35.up_proj", "model.layers.11.mlp.experts.36.up_proj", "model.layers.11.mlp.experts.37.up_proj", "model.layers.11.mlp.experts.38.up_proj", "model.layers.11.mlp.experts.39.up_proj", "model.layers.11.mlp.experts.40.up_proj", "model.layers.11.mlp.experts.41.up_proj", "model.layers.11.mlp.experts.42.up_proj", "model.layers.11.mlp.experts.43.up_proj", "model.layers.11.mlp.experts.44.up_proj", "model.layers.11.mlp.experts.45.up_proj", "model.layers.11.mlp.experts.46.up_proj", "model.layers.11.mlp.experts.47.up_proj", "model.layers.11.mlp.experts.48.up_proj", "model.layers.11.mlp.experts.49.up_proj", "model.layers.11.mlp.experts.50.up_proj", "model.layers.11.mlp.experts.51.up_proj", "model.layers.11.mlp.experts.52.up_proj", "model.layers.11.mlp.experts.53.up_proj", "model.layers.11.mlp.experts.54.up_proj", "model.layers.11.mlp.experts.55.up_proj", "model.layers.11.mlp.experts.56.up_proj", "model.layers.11.mlp.experts.57.up_proj", "model.layers.11.mlp.experts.58.up_proj", "model.layers.11.mlp.experts.59.up_proj", "model.layers.11.mlp.experts.60.up_proj", "model.layers.11.mlp.experts.61.up_proj", "model.layers.11.mlp.experts.62.up_proj", "model.layers.11.mlp.experts.63.up_proj", "model.layers.11.mlp.experts.64.up_proj", "model.layers.11.mlp.experts.65.up_proj", "model.layers.11.mlp.experts.66.up_proj", "model.layers.11.mlp.experts.67.up_proj", "model.layers.11.mlp.experts.68.up_proj", "model.layers.11.mlp.experts.69.up_proj", "model.layers.11.mlp.experts.70.up_proj", "model.layers.11.mlp.experts.71.up_proj", "model.layers.11.mlp.experts.72.up_proj", "model.layers.11.mlp.experts.73.up_proj", "model.layers.11.mlp.experts.74.up_proj", "model.layers.11.mlp.experts.75.up_proj", "model.layers.11.mlp.experts.76.up_proj", "model.layers.11.mlp.experts.77.up_proj", "model.layers.11.mlp.experts.78.up_proj", "model.layers.11.mlp.experts.79.up_proj", "model.layers.11.mlp.experts.80.up_proj", "model.layers.11.mlp.experts.81.up_proj", "model.layers.11.mlp.experts.82.up_proj", "model.layers.11.mlp.experts.83.up_proj", "model.layers.11.mlp.experts.84.up_proj", "model.layers.11.mlp.experts.85.up_proj", "model.layers.11.mlp.experts.86.up_proj", "model.layers.11.mlp.experts.87.up_proj", "model.layers.11.mlp.experts.88.up_proj", "model.layers.11.mlp.experts.89.up_proj", "model.layers.11.mlp.experts.90.up_proj", "model.layers.11.mlp.experts.91.up_proj", "model.layers.11.mlp.experts.92.up_proj", "model.layers.11.mlp.experts.93.up_proj", "model.layers.11.mlp.experts.94.up_proj", "model.layers.11.mlp.experts.95.up_proj", "model.layers.11.mlp.experts.96.up_proj", "model.layers.11.mlp.experts.97.up_proj", "model.layers.11.mlp.experts.98.up_proj", "model.layers.11.mlp.experts.99.up_proj", "model.layers.11.mlp.experts.100.up_proj", "model.layers.11.mlp.experts.101.up_proj", "model.layers.11.mlp.experts.102.up_proj", "model.layers.11.mlp.experts.103.up_proj", "model.layers.11.mlp.experts.104.up_proj", "model.layers.11.mlp.experts.105.up_proj", "model.layers.11.mlp.experts.106.up_proj", "model.layers.11.mlp.experts.107.up_proj", "model.layers.11.mlp.experts.108.up_proj", "model.layers.11.mlp.experts.109.up_proj", "model.layers.11.mlp.experts.110.up_proj", "model.layers.11.mlp.experts.111.up_proj", "model.layers.11.mlp.experts.112.up_proj", "model.layers.11.mlp.experts.113.up_proj", "model.layers.11.mlp.experts.114.up_proj", "model.layers.11.mlp.experts.115.up_proj", "model.layers.11.mlp.experts.116.up_proj", "model.layers.11.mlp.experts.117.up_proj", "model.layers.11.mlp.experts.118.up_proj", "model.layers.11.mlp.experts.119.up_proj", "model.layers.11.mlp.experts.120.up_proj", "model.layers.11.mlp.experts.121.up_proj", "model.layers.11.mlp.experts.122.up_proj", "model.layers.11.mlp.experts.123.up_proj", "model.layers.11.mlp.experts.124.up_proj", "model.layers.11.mlp.experts.125.up_proj", "model.layers.11.mlp.experts.126.up_proj", "model.layers.11.mlp.experts.127.up_proj", "model.layers.11.mlp.experts.128.up_proj", "model.layers.11.mlp.experts.129.up_proj", "model.layers.11.mlp.experts.130.up_proj", "model.layers.11.mlp.experts.131.up_proj", "model.layers.11.mlp.experts.132.up_proj", "model.layers.11.mlp.experts.133.up_proj", "model.layers.11.mlp.experts.134.up_proj", "model.layers.11.mlp.experts.135.up_proj", "model.layers.11.mlp.experts.136.up_proj", "model.layers.11.mlp.experts.137.up_proj", "model.layers.11.mlp.experts.138.up_proj", "model.layers.11.mlp.experts.139.up_proj", "model.layers.11.mlp.experts.140.up_proj", "model.layers.11.mlp.experts.141.up_proj", "model.layers.11.mlp.experts.142.up_proj", "model.layers.11.mlp.experts.143.up_proj", "model.layers.11.mlp.experts.144.up_proj", "model.layers.11.mlp.experts.145.up_proj", "model.layers.11.mlp.experts.146.up_proj", "model.layers.11.mlp.experts.147.up_proj", "model.layers.11.mlp.experts.148.up_proj", "model.layers.11.mlp.experts.149.up_proj", "model.layers.11.mlp.experts.150.up_proj", "model.layers.11.mlp.experts.151.up_proj", "model.layers.11.mlp.experts.152.up_proj", "model.layers.11.mlp.experts.153.up_proj", "model.layers.11.mlp.experts.154.up_proj", "model.layers.11.mlp.experts.155.up_proj", "model.layers.11.mlp.experts.156.up_proj", "model.layers.11.mlp.experts.157.up_proj", "model.layers.11.mlp.experts.158.up_proj", "model.layers.11.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0004231601022183923, "dbits": 2516582400 }, { "dkld": -0.0005358958616852871, "dbits": 5033164800 }, { "dkld": -0.0006879685446620071, "dbits": 7549747200 }, { "dkld": -0.0006855024956166772, "dbits": 12582912000 } ] }, { "idx": 68, "layers": [ "model.layers.11.mlp.experts.0.down_proj", "model.layers.11.mlp.experts.1.down_proj", "model.layers.11.mlp.experts.2.down_proj", "model.layers.11.mlp.experts.3.down_proj", "model.layers.11.mlp.experts.4.down_proj", "model.layers.11.mlp.experts.5.down_proj", "model.layers.11.mlp.experts.6.down_proj", "model.layers.11.mlp.experts.7.down_proj", "model.layers.11.mlp.experts.8.down_proj", "model.layers.11.mlp.experts.9.down_proj", "model.layers.11.mlp.experts.10.down_proj", "model.layers.11.mlp.experts.11.down_proj", "model.layers.11.mlp.experts.12.down_proj", "model.layers.11.mlp.experts.13.down_proj", "model.layers.11.mlp.experts.14.down_proj", "model.layers.11.mlp.experts.15.down_proj", "model.layers.11.mlp.experts.16.down_proj", "model.layers.11.mlp.experts.17.down_proj", "model.layers.11.mlp.experts.18.down_proj", "model.layers.11.mlp.experts.19.down_proj", "model.layers.11.mlp.experts.20.down_proj", "model.layers.11.mlp.experts.21.down_proj", "model.layers.11.mlp.experts.22.down_proj", "model.layers.11.mlp.experts.23.down_proj", "model.layers.11.mlp.experts.24.down_proj", "model.layers.11.mlp.experts.25.down_proj", "model.layers.11.mlp.experts.26.down_proj", "model.layers.11.mlp.experts.27.down_proj", "model.layers.11.mlp.experts.28.down_proj", "model.layers.11.mlp.experts.29.down_proj", "model.layers.11.mlp.experts.30.down_proj", "model.layers.11.mlp.experts.31.down_proj", "model.layers.11.mlp.experts.32.down_proj", "model.layers.11.mlp.experts.33.down_proj", "model.layers.11.mlp.experts.34.down_proj", "model.layers.11.mlp.experts.35.down_proj", "model.layers.11.mlp.experts.36.down_proj", "model.layers.11.mlp.experts.37.down_proj", "model.layers.11.mlp.experts.38.down_proj", "model.layers.11.mlp.experts.39.down_proj", "model.layers.11.mlp.experts.40.down_proj", "model.layers.11.mlp.experts.41.down_proj", "model.layers.11.mlp.experts.42.down_proj", "model.layers.11.mlp.experts.43.down_proj", "model.layers.11.mlp.experts.44.down_proj", "model.layers.11.mlp.experts.45.down_proj", "model.layers.11.mlp.experts.46.down_proj", "model.layers.11.mlp.experts.47.down_proj", "model.layers.11.mlp.experts.48.down_proj", "model.layers.11.mlp.experts.49.down_proj", "model.layers.11.mlp.experts.50.down_proj", "model.layers.11.mlp.experts.51.down_proj", "model.layers.11.mlp.experts.52.down_proj", "model.layers.11.mlp.experts.53.down_proj", "model.layers.11.mlp.experts.54.down_proj", "model.layers.11.mlp.experts.55.down_proj", "model.layers.11.mlp.experts.56.down_proj", "model.layers.11.mlp.experts.57.down_proj", "model.layers.11.mlp.experts.58.down_proj", "model.layers.11.mlp.experts.59.down_proj", "model.layers.11.mlp.experts.60.down_proj", "model.layers.11.mlp.experts.61.down_proj", "model.layers.11.mlp.experts.62.down_proj", "model.layers.11.mlp.experts.63.down_proj", "model.layers.11.mlp.experts.64.down_proj", "model.layers.11.mlp.experts.65.down_proj", "model.layers.11.mlp.experts.66.down_proj", "model.layers.11.mlp.experts.67.down_proj", "model.layers.11.mlp.experts.68.down_proj", "model.layers.11.mlp.experts.69.down_proj", "model.layers.11.mlp.experts.70.down_proj", "model.layers.11.mlp.experts.71.down_proj", "model.layers.11.mlp.experts.72.down_proj", "model.layers.11.mlp.experts.73.down_proj", "model.layers.11.mlp.experts.74.down_proj", "model.layers.11.mlp.experts.75.down_proj", "model.layers.11.mlp.experts.76.down_proj", "model.layers.11.mlp.experts.77.down_proj", "model.layers.11.mlp.experts.78.down_proj", "model.layers.11.mlp.experts.79.down_proj", "model.layers.11.mlp.experts.80.down_proj", "model.layers.11.mlp.experts.81.down_proj", "model.layers.11.mlp.experts.82.down_proj", "model.layers.11.mlp.experts.83.down_proj", "model.layers.11.mlp.experts.84.down_proj", "model.layers.11.mlp.experts.85.down_proj", "model.layers.11.mlp.experts.86.down_proj", "model.layers.11.mlp.experts.87.down_proj", "model.layers.11.mlp.experts.88.down_proj", "model.layers.11.mlp.experts.89.down_proj", "model.layers.11.mlp.experts.90.down_proj", "model.layers.11.mlp.experts.91.down_proj", "model.layers.11.mlp.experts.92.down_proj", "model.layers.11.mlp.experts.93.down_proj", "model.layers.11.mlp.experts.94.down_proj", "model.layers.11.mlp.experts.95.down_proj", "model.layers.11.mlp.experts.96.down_proj", "model.layers.11.mlp.experts.97.down_proj", "model.layers.11.mlp.experts.98.down_proj", "model.layers.11.mlp.experts.99.down_proj", "model.layers.11.mlp.experts.100.down_proj", "model.layers.11.mlp.experts.101.down_proj", "model.layers.11.mlp.experts.102.down_proj", "model.layers.11.mlp.experts.103.down_proj", "model.layers.11.mlp.experts.104.down_proj", "model.layers.11.mlp.experts.105.down_proj", "model.layers.11.mlp.experts.106.down_proj", "model.layers.11.mlp.experts.107.down_proj", "model.layers.11.mlp.experts.108.down_proj", "model.layers.11.mlp.experts.109.down_proj", "model.layers.11.mlp.experts.110.down_proj", "model.layers.11.mlp.experts.111.down_proj", "model.layers.11.mlp.experts.112.down_proj", "model.layers.11.mlp.experts.113.down_proj", "model.layers.11.mlp.experts.114.down_proj", "model.layers.11.mlp.experts.115.down_proj", "model.layers.11.mlp.experts.116.down_proj", "model.layers.11.mlp.experts.117.down_proj", "model.layers.11.mlp.experts.118.down_proj", "model.layers.11.mlp.experts.119.down_proj", "model.layers.11.mlp.experts.120.down_proj", "model.layers.11.mlp.experts.121.down_proj", "model.layers.11.mlp.experts.122.down_proj", "model.layers.11.mlp.experts.123.down_proj", "model.layers.11.mlp.experts.124.down_proj", "model.layers.11.mlp.experts.125.down_proj", "model.layers.11.mlp.experts.126.down_proj", "model.layers.11.mlp.experts.127.down_proj", "model.layers.11.mlp.experts.128.down_proj", "model.layers.11.mlp.experts.129.down_proj", "model.layers.11.mlp.experts.130.down_proj", "model.layers.11.mlp.experts.131.down_proj", "model.layers.11.mlp.experts.132.down_proj", "model.layers.11.mlp.experts.133.down_proj", "model.layers.11.mlp.experts.134.down_proj", "model.layers.11.mlp.experts.135.down_proj", "model.layers.11.mlp.experts.136.down_proj", "model.layers.11.mlp.experts.137.down_proj", "model.layers.11.mlp.experts.138.down_proj", "model.layers.11.mlp.experts.139.down_proj", "model.layers.11.mlp.experts.140.down_proj", "model.layers.11.mlp.experts.141.down_proj", "model.layers.11.mlp.experts.142.down_proj", "model.layers.11.mlp.experts.143.down_proj", "model.layers.11.mlp.experts.144.down_proj", "model.layers.11.mlp.experts.145.down_proj", "model.layers.11.mlp.experts.146.down_proj", "model.layers.11.mlp.experts.147.down_proj", "model.layers.11.mlp.experts.148.down_proj", "model.layers.11.mlp.experts.149.down_proj", "model.layers.11.mlp.experts.150.down_proj", "model.layers.11.mlp.experts.151.down_proj", "model.layers.11.mlp.experts.152.down_proj", "model.layers.11.mlp.experts.153.down_proj", "model.layers.11.mlp.experts.154.down_proj", "model.layers.11.mlp.experts.155.down_proj", "model.layers.11.mlp.experts.156.down_proj", "model.layers.11.mlp.experts.157.down_proj", "model.layers.11.mlp.experts.158.down_proj", "model.layers.11.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0005672518163919449, "dbits": 1258291200 }, { "dkld": -0.0005827613174915341, "dbits": 2516582400 }, { "dkld": -0.0006414241157472189, "dbits": 3774873600 }, { "dkld": -0.0005702663213014714, "dbits": 6291456000 } ] }, { "idx": 69, "layers": [ "model.layers.12.self_attn.q_proj" ], "candidates": [ { "dkld": 1.2642331421369724e-05, "dbits": 62914560 }, { "dkld": 1.2811273336410522e-05, "dbits": 125829120 }, { "dkld": 4.6665221452704686e-05, "dbits": 188743680 }, { "dkld": 1.5820190310475435e-05, "dbits": 314572800 } ] }, { "idx": 70, "layers": [ "model.layers.12.self_attn.k_proj", "model.layers.12.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0001492667943239212, "dbits": 10485760 }, { "dkld": 0.00026277918368577957, "dbits": 20971520 }, { "dkld": 0.0002904882654547608, "dbits": 31457280 }, { "dkld": 0.00014486508443950497, "dbits": 52428800 } ] }, { "idx": 71, "layers": [ "model.layers.12.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010460145771503532, "dbits": 62914560 }, { "dkld": -0.0007683807983994567, "dbits": 125829120 }, { "dkld": -0.00081679318100214, "dbits": 188743680 }, { "dkld": -0.0007722014561295509, "dbits": 314572800 } ] }, { "idx": 72, "layers": [ "model.layers.12.mlp.shared_experts.gate_proj", "model.layers.12.mlp.shared_experts.up_proj", "model.layers.12.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0004189413040876361, "dbits": 23592960 }, { "dkld": 0.00010449141263961514, "dbits": 47185920 }, { "dkld": 6.536841392516812e-05, "dbits": 70778880 }, { "dkld": -6.022145971656956e-05, "dbits": 117964800 } ] }, { "idx": 73, "layers": [ "model.layers.12.mlp.experts.0.gate_proj", "model.layers.12.mlp.experts.1.gate_proj", "model.layers.12.mlp.experts.2.gate_proj", "model.layers.12.mlp.experts.3.gate_proj", "model.layers.12.mlp.experts.4.gate_proj", "model.layers.12.mlp.experts.5.gate_proj", "model.layers.12.mlp.experts.6.gate_proj", "model.layers.12.mlp.experts.7.gate_proj", "model.layers.12.mlp.experts.8.gate_proj", "model.layers.12.mlp.experts.9.gate_proj", "model.layers.12.mlp.experts.10.gate_proj", "model.layers.12.mlp.experts.11.gate_proj", "model.layers.12.mlp.experts.12.gate_proj", "model.layers.12.mlp.experts.13.gate_proj", "model.layers.12.mlp.experts.14.gate_proj", "model.layers.12.mlp.experts.15.gate_proj", "model.layers.12.mlp.experts.16.gate_proj", "model.layers.12.mlp.experts.17.gate_proj", "model.layers.12.mlp.experts.18.gate_proj", "model.layers.12.mlp.experts.19.gate_proj", "model.layers.12.mlp.experts.20.gate_proj", "model.layers.12.mlp.experts.21.gate_proj", "model.layers.12.mlp.experts.22.gate_proj", "model.layers.12.mlp.experts.23.gate_proj", "model.layers.12.mlp.experts.24.gate_proj", "model.layers.12.mlp.experts.25.gate_proj", "model.layers.12.mlp.experts.26.gate_proj", "model.layers.12.mlp.experts.27.gate_proj", "model.layers.12.mlp.experts.28.gate_proj", "model.layers.12.mlp.experts.29.gate_proj", "model.layers.12.mlp.experts.30.gate_proj", "model.layers.12.mlp.experts.31.gate_proj", "model.layers.12.mlp.experts.32.gate_proj", "model.layers.12.mlp.experts.33.gate_proj", "model.layers.12.mlp.experts.34.gate_proj", "model.layers.12.mlp.experts.35.gate_proj", "model.layers.12.mlp.experts.36.gate_proj", "model.layers.12.mlp.experts.37.gate_proj", "model.layers.12.mlp.experts.38.gate_proj", "model.layers.12.mlp.experts.39.gate_proj", "model.layers.12.mlp.experts.40.gate_proj", "model.layers.12.mlp.experts.41.gate_proj", "model.layers.12.mlp.experts.42.gate_proj", "model.layers.12.mlp.experts.43.gate_proj", "model.layers.12.mlp.experts.44.gate_proj", "model.layers.12.mlp.experts.45.gate_proj", "model.layers.12.mlp.experts.46.gate_proj", "model.layers.12.mlp.experts.47.gate_proj", "model.layers.12.mlp.experts.48.gate_proj", "model.layers.12.mlp.experts.49.gate_proj", "model.layers.12.mlp.experts.50.gate_proj", "model.layers.12.mlp.experts.51.gate_proj", "model.layers.12.mlp.experts.52.gate_proj", "model.layers.12.mlp.experts.53.gate_proj", "model.layers.12.mlp.experts.54.gate_proj", "model.layers.12.mlp.experts.55.gate_proj", "model.layers.12.mlp.experts.56.gate_proj", "model.layers.12.mlp.experts.57.gate_proj", "model.layers.12.mlp.experts.58.gate_proj", "model.layers.12.mlp.experts.59.gate_proj", "model.layers.12.mlp.experts.60.gate_proj", "model.layers.12.mlp.experts.61.gate_proj", "model.layers.12.mlp.experts.62.gate_proj", "model.layers.12.mlp.experts.63.gate_proj", "model.layers.12.mlp.experts.64.gate_proj", "model.layers.12.mlp.experts.65.gate_proj", "model.layers.12.mlp.experts.66.gate_proj", "model.layers.12.mlp.experts.67.gate_proj", "model.layers.12.mlp.experts.68.gate_proj", "model.layers.12.mlp.experts.69.gate_proj", "model.layers.12.mlp.experts.70.gate_proj", "model.layers.12.mlp.experts.71.gate_proj", "model.layers.12.mlp.experts.72.gate_proj", "model.layers.12.mlp.experts.73.gate_proj", "model.layers.12.mlp.experts.74.gate_proj", "model.layers.12.mlp.experts.75.gate_proj", "model.layers.12.mlp.experts.76.gate_proj", "model.layers.12.mlp.experts.77.gate_proj", "model.layers.12.mlp.experts.78.gate_proj", "model.layers.12.mlp.experts.79.gate_proj", "model.layers.12.mlp.experts.80.gate_proj", "model.layers.12.mlp.experts.81.gate_proj", "model.layers.12.mlp.experts.82.gate_proj", "model.layers.12.mlp.experts.83.gate_proj", "model.layers.12.mlp.experts.84.gate_proj", "model.layers.12.mlp.experts.85.gate_proj", "model.layers.12.mlp.experts.86.gate_proj", "model.layers.12.mlp.experts.87.gate_proj", "model.layers.12.mlp.experts.88.gate_proj", "model.layers.12.mlp.experts.89.gate_proj", "model.layers.12.mlp.experts.90.gate_proj", "model.layers.12.mlp.experts.91.gate_proj", "model.layers.12.mlp.experts.92.gate_proj", "model.layers.12.mlp.experts.93.gate_proj", "model.layers.12.mlp.experts.94.gate_proj", "model.layers.12.mlp.experts.95.gate_proj", "model.layers.12.mlp.experts.96.gate_proj", "model.layers.12.mlp.experts.97.gate_proj", "model.layers.12.mlp.experts.98.gate_proj", "model.layers.12.mlp.experts.99.gate_proj", "model.layers.12.mlp.experts.100.gate_proj", "model.layers.12.mlp.experts.101.gate_proj", "model.layers.12.mlp.experts.102.gate_proj", "model.layers.12.mlp.experts.103.gate_proj", "model.layers.12.mlp.experts.104.gate_proj", "model.layers.12.mlp.experts.105.gate_proj", "model.layers.12.mlp.experts.106.gate_proj", "model.layers.12.mlp.experts.107.gate_proj", "model.layers.12.mlp.experts.108.gate_proj", "model.layers.12.mlp.experts.109.gate_proj", "model.layers.12.mlp.experts.110.gate_proj", "model.layers.12.mlp.experts.111.gate_proj", "model.layers.12.mlp.experts.112.gate_proj", "model.layers.12.mlp.experts.113.gate_proj", "model.layers.12.mlp.experts.114.gate_proj", "model.layers.12.mlp.experts.115.gate_proj", "model.layers.12.mlp.experts.116.gate_proj", "model.layers.12.mlp.experts.117.gate_proj", "model.layers.12.mlp.experts.118.gate_proj", "model.layers.12.mlp.experts.119.gate_proj", "model.layers.12.mlp.experts.120.gate_proj", "model.layers.12.mlp.experts.121.gate_proj", "model.layers.12.mlp.experts.122.gate_proj", "model.layers.12.mlp.experts.123.gate_proj", "model.layers.12.mlp.experts.124.gate_proj", "model.layers.12.mlp.experts.125.gate_proj", "model.layers.12.mlp.experts.126.gate_proj", "model.layers.12.mlp.experts.127.gate_proj", "model.layers.12.mlp.experts.128.gate_proj", "model.layers.12.mlp.experts.129.gate_proj", "model.layers.12.mlp.experts.130.gate_proj", "model.layers.12.mlp.experts.131.gate_proj", "model.layers.12.mlp.experts.132.gate_proj", "model.layers.12.mlp.experts.133.gate_proj", "model.layers.12.mlp.experts.134.gate_proj", "model.layers.12.mlp.experts.135.gate_proj", "model.layers.12.mlp.experts.136.gate_proj", "model.layers.12.mlp.experts.137.gate_proj", "model.layers.12.mlp.experts.138.gate_proj", "model.layers.12.mlp.experts.139.gate_proj", "model.layers.12.mlp.experts.140.gate_proj", "model.layers.12.mlp.experts.141.gate_proj", "model.layers.12.mlp.experts.142.gate_proj", "model.layers.12.mlp.experts.143.gate_proj", "model.layers.12.mlp.experts.144.gate_proj", "model.layers.12.mlp.experts.145.gate_proj", "model.layers.12.mlp.experts.146.gate_proj", "model.layers.12.mlp.experts.147.gate_proj", "model.layers.12.mlp.experts.148.gate_proj", "model.layers.12.mlp.experts.149.gate_proj", "model.layers.12.mlp.experts.150.gate_proj", "model.layers.12.mlp.experts.151.gate_proj", "model.layers.12.mlp.experts.152.gate_proj", "model.layers.12.mlp.experts.153.gate_proj", "model.layers.12.mlp.experts.154.gate_proj", "model.layers.12.mlp.experts.155.gate_proj", "model.layers.12.mlp.experts.156.gate_proj", "model.layers.12.mlp.experts.157.gate_proj", "model.layers.12.mlp.experts.158.gate_proj", "model.layers.12.mlp.experts.159.gate_proj", "model.layers.12.mlp.experts.0.up_proj", "model.layers.12.mlp.experts.1.up_proj", "model.layers.12.mlp.experts.2.up_proj", "model.layers.12.mlp.experts.3.up_proj", "model.layers.12.mlp.experts.4.up_proj", "model.layers.12.mlp.experts.5.up_proj", "model.layers.12.mlp.experts.6.up_proj", "model.layers.12.mlp.experts.7.up_proj", "model.layers.12.mlp.experts.8.up_proj", "model.layers.12.mlp.experts.9.up_proj", "model.layers.12.mlp.experts.10.up_proj", "model.layers.12.mlp.experts.11.up_proj", "model.layers.12.mlp.experts.12.up_proj", "model.layers.12.mlp.experts.13.up_proj", "model.layers.12.mlp.experts.14.up_proj", "model.layers.12.mlp.experts.15.up_proj", "model.layers.12.mlp.experts.16.up_proj", "model.layers.12.mlp.experts.17.up_proj", "model.layers.12.mlp.experts.18.up_proj", "model.layers.12.mlp.experts.19.up_proj", "model.layers.12.mlp.experts.20.up_proj", "model.layers.12.mlp.experts.21.up_proj", "model.layers.12.mlp.experts.22.up_proj", "model.layers.12.mlp.experts.23.up_proj", "model.layers.12.mlp.experts.24.up_proj", "model.layers.12.mlp.experts.25.up_proj", "model.layers.12.mlp.experts.26.up_proj", "model.layers.12.mlp.experts.27.up_proj", "model.layers.12.mlp.experts.28.up_proj", "model.layers.12.mlp.experts.29.up_proj", "model.layers.12.mlp.experts.30.up_proj", "model.layers.12.mlp.experts.31.up_proj", "model.layers.12.mlp.experts.32.up_proj", "model.layers.12.mlp.experts.33.up_proj", "model.layers.12.mlp.experts.34.up_proj", "model.layers.12.mlp.experts.35.up_proj", "model.layers.12.mlp.experts.36.up_proj", "model.layers.12.mlp.experts.37.up_proj", "model.layers.12.mlp.experts.38.up_proj", "model.layers.12.mlp.experts.39.up_proj", "model.layers.12.mlp.experts.40.up_proj", "model.layers.12.mlp.experts.41.up_proj", "model.layers.12.mlp.experts.42.up_proj", "model.layers.12.mlp.experts.43.up_proj", "model.layers.12.mlp.experts.44.up_proj", "model.layers.12.mlp.experts.45.up_proj", "model.layers.12.mlp.experts.46.up_proj", "model.layers.12.mlp.experts.47.up_proj", "model.layers.12.mlp.experts.48.up_proj", "model.layers.12.mlp.experts.49.up_proj", "model.layers.12.mlp.experts.50.up_proj", "model.layers.12.mlp.experts.51.up_proj", "model.layers.12.mlp.experts.52.up_proj", "model.layers.12.mlp.experts.53.up_proj", "model.layers.12.mlp.experts.54.up_proj", "model.layers.12.mlp.experts.55.up_proj", "model.layers.12.mlp.experts.56.up_proj", "model.layers.12.mlp.experts.57.up_proj", "model.layers.12.mlp.experts.58.up_proj", "model.layers.12.mlp.experts.59.up_proj", "model.layers.12.mlp.experts.60.up_proj", "model.layers.12.mlp.experts.61.up_proj", "model.layers.12.mlp.experts.62.up_proj", "model.layers.12.mlp.experts.63.up_proj", "model.layers.12.mlp.experts.64.up_proj", "model.layers.12.mlp.experts.65.up_proj", "model.layers.12.mlp.experts.66.up_proj", "model.layers.12.mlp.experts.67.up_proj", "model.layers.12.mlp.experts.68.up_proj", "model.layers.12.mlp.experts.69.up_proj", "model.layers.12.mlp.experts.70.up_proj", "model.layers.12.mlp.experts.71.up_proj", "model.layers.12.mlp.experts.72.up_proj", "model.layers.12.mlp.experts.73.up_proj", "model.layers.12.mlp.experts.74.up_proj", "model.layers.12.mlp.experts.75.up_proj", "model.layers.12.mlp.experts.76.up_proj", "model.layers.12.mlp.experts.77.up_proj", "model.layers.12.mlp.experts.78.up_proj", "model.layers.12.mlp.experts.79.up_proj", "model.layers.12.mlp.experts.80.up_proj", "model.layers.12.mlp.experts.81.up_proj", "model.layers.12.mlp.experts.82.up_proj", "model.layers.12.mlp.experts.83.up_proj", "model.layers.12.mlp.experts.84.up_proj", "model.layers.12.mlp.experts.85.up_proj", "model.layers.12.mlp.experts.86.up_proj", "model.layers.12.mlp.experts.87.up_proj", "model.layers.12.mlp.experts.88.up_proj", "model.layers.12.mlp.experts.89.up_proj", "model.layers.12.mlp.experts.90.up_proj", "model.layers.12.mlp.experts.91.up_proj", "model.layers.12.mlp.experts.92.up_proj", "model.layers.12.mlp.experts.93.up_proj", "model.layers.12.mlp.experts.94.up_proj", "model.layers.12.mlp.experts.95.up_proj", "model.layers.12.mlp.experts.96.up_proj", "model.layers.12.mlp.experts.97.up_proj", "model.layers.12.mlp.experts.98.up_proj", "model.layers.12.mlp.experts.99.up_proj", "model.layers.12.mlp.experts.100.up_proj", "model.layers.12.mlp.experts.101.up_proj", "model.layers.12.mlp.experts.102.up_proj", "model.layers.12.mlp.experts.103.up_proj", "model.layers.12.mlp.experts.104.up_proj", "model.layers.12.mlp.experts.105.up_proj", "model.layers.12.mlp.experts.106.up_proj", "model.layers.12.mlp.experts.107.up_proj", "model.layers.12.mlp.experts.108.up_proj", "model.layers.12.mlp.experts.109.up_proj", "model.layers.12.mlp.experts.110.up_proj", "model.layers.12.mlp.experts.111.up_proj", "model.layers.12.mlp.experts.112.up_proj", "model.layers.12.mlp.experts.113.up_proj", "model.layers.12.mlp.experts.114.up_proj", "model.layers.12.mlp.experts.115.up_proj", "model.layers.12.mlp.experts.116.up_proj", "model.layers.12.mlp.experts.117.up_proj", "model.layers.12.mlp.experts.118.up_proj", "model.layers.12.mlp.experts.119.up_proj", "model.layers.12.mlp.experts.120.up_proj", "model.layers.12.mlp.experts.121.up_proj", "model.layers.12.mlp.experts.122.up_proj", "model.layers.12.mlp.experts.123.up_proj", "model.layers.12.mlp.experts.124.up_proj", "model.layers.12.mlp.experts.125.up_proj", "model.layers.12.mlp.experts.126.up_proj", "model.layers.12.mlp.experts.127.up_proj", "model.layers.12.mlp.experts.128.up_proj", "model.layers.12.mlp.experts.129.up_proj", "model.layers.12.mlp.experts.130.up_proj", "model.layers.12.mlp.experts.131.up_proj", "model.layers.12.mlp.experts.132.up_proj", "model.layers.12.mlp.experts.133.up_proj", "model.layers.12.mlp.experts.134.up_proj", "model.layers.12.mlp.experts.135.up_proj", "model.layers.12.mlp.experts.136.up_proj", "model.layers.12.mlp.experts.137.up_proj", "model.layers.12.mlp.experts.138.up_proj", "model.layers.12.mlp.experts.139.up_proj", "model.layers.12.mlp.experts.140.up_proj", "model.layers.12.mlp.experts.141.up_proj", "model.layers.12.mlp.experts.142.up_proj", "model.layers.12.mlp.experts.143.up_proj", "model.layers.12.mlp.experts.144.up_proj", "model.layers.12.mlp.experts.145.up_proj", "model.layers.12.mlp.experts.146.up_proj", "model.layers.12.mlp.experts.147.up_proj", "model.layers.12.mlp.experts.148.up_proj", "model.layers.12.mlp.experts.149.up_proj", "model.layers.12.mlp.experts.150.up_proj", "model.layers.12.mlp.experts.151.up_proj", "model.layers.12.mlp.experts.152.up_proj", "model.layers.12.mlp.experts.153.up_proj", "model.layers.12.mlp.experts.154.up_proj", "model.layers.12.mlp.experts.155.up_proj", "model.layers.12.mlp.experts.156.up_proj", "model.layers.12.mlp.experts.157.up_proj", "model.layers.12.mlp.experts.158.up_proj", "model.layers.12.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0007062182761728847, "dbits": 2516582400 }, { "dkld": -0.0008367450907826451, "dbits": 5033164800 }, { "dkld": -0.000830700621008873, "dbits": 7549747200 }, { "dkld": -0.000787041801959279, "dbits": 12582912000 } ] }, { "idx": 74, "layers": [ "model.layers.12.mlp.experts.0.down_proj", "model.layers.12.mlp.experts.1.down_proj", "model.layers.12.mlp.experts.2.down_proj", "model.layers.12.mlp.experts.3.down_proj", "model.layers.12.mlp.experts.4.down_proj", "model.layers.12.mlp.experts.5.down_proj", "model.layers.12.mlp.experts.6.down_proj", "model.layers.12.mlp.experts.7.down_proj", "model.layers.12.mlp.experts.8.down_proj", "model.layers.12.mlp.experts.9.down_proj", "model.layers.12.mlp.experts.10.down_proj", "model.layers.12.mlp.experts.11.down_proj", "model.layers.12.mlp.experts.12.down_proj", "model.layers.12.mlp.experts.13.down_proj", "model.layers.12.mlp.experts.14.down_proj", "model.layers.12.mlp.experts.15.down_proj", "model.layers.12.mlp.experts.16.down_proj", "model.layers.12.mlp.experts.17.down_proj", "model.layers.12.mlp.experts.18.down_proj", "model.layers.12.mlp.experts.19.down_proj", "model.layers.12.mlp.experts.20.down_proj", "model.layers.12.mlp.experts.21.down_proj", "model.layers.12.mlp.experts.22.down_proj", "model.layers.12.mlp.experts.23.down_proj", "model.layers.12.mlp.experts.24.down_proj", "model.layers.12.mlp.experts.25.down_proj", "model.layers.12.mlp.experts.26.down_proj", "model.layers.12.mlp.experts.27.down_proj", "model.layers.12.mlp.experts.28.down_proj", "model.layers.12.mlp.experts.29.down_proj", "model.layers.12.mlp.experts.30.down_proj", "model.layers.12.mlp.experts.31.down_proj", "model.layers.12.mlp.experts.32.down_proj", "model.layers.12.mlp.experts.33.down_proj", "model.layers.12.mlp.experts.34.down_proj", "model.layers.12.mlp.experts.35.down_proj", "model.layers.12.mlp.experts.36.down_proj", "model.layers.12.mlp.experts.37.down_proj", "model.layers.12.mlp.experts.38.down_proj", "model.layers.12.mlp.experts.39.down_proj", "model.layers.12.mlp.experts.40.down_proj", "model.layers.12.mlp.experts.41.down_proj", "model.layers.12.mlp.experts.42.down_proj", "model.layers.12.mlp.experts.43.down_proj", "model.layers.12.mlp.experts.44.down_proj", "model.layers.12.mlp.experts.45.down_proj", "model.layers.12.mlp.experts.46.down_proj", "model.layers.12.mlp.experts.47.down_proj", "model.layers.12.mlp.experts.48.down_proj", "model.layers.12.mlp.experts.49.down_proj", "model.layers.12.mlp.experts.50.down_proj", "model.layers.12.mlp.experts.51.down_proj", "model.layers.12.mlp.experts.52.down_proj", "model.layers.12.mlp.experts.53.down_proj", "model.layers.12.mlp.experts.54.down_proj", "model.layers.12.mlp.experts.55.down_proj", "model.layers.12.mlp.experts.56.down_proj", "model.layers.12.mlp.experts.57.down_proj", "model.layers.12.mlp.experts.58.down_proj", "model.layers.12.mlp.experts.59.down_proj", "model.layers.12.mlp.experts.60.down_proj", "model.layers.12.mlp.experts.61.down_proj", "model.layers.12.mlp.experts.62.down_proj", "model.layers.12.mlp.experts.63.down_proj", "model.layers.12.mlp.experts.64.down_proj", "model.layers.12.mlp.experts.65.down_proj", "model.layers.12.mlp.experts.66.down_proj", "model.layers.12.mlp.experts.67.down_proj", "model.layers.12.mlp.experts.68.down_proj", "model.layers.12.mlp.experts.69.down_proj", "model.layers.12.mlp.experts.70.down_proj", "model.layers.12.mlp.experts.71.down_proj", "model.layers.12.mlp.experts.72.down_proj", "model.layers.12.mlp.experts.73.down_proj", "model.layers.12.mlp.experts.74.down_proj", "model.layers.12.mlp.experts.75.down_proj", "model.layers.12.mlp.experts.76.down_proj", "model.layers.12.mlp.experts.77.down_proj", "model.layers.12.mlp.experts.78.down_proj", "model.layers.12.mlp.experts.79.down_proj", "model.layers.12.mlp.experts.80.down_proj", "model.layers.12.mlp.experts.81.down_proj", "model.layers.12.mlp.experts.82.down_proj", "model.layers.12.mlp.experts.83.down_proj", "model.layers.12.mlp.experts.84.down_proj", "model.layers.12.mlp.experts.85.down_proj", "model.layers.12.mlp.experts.86.down_proj", "model.layers.12.mlp.experts.87.down_proj", "model.layers.12.mlp.experts.88.down_proj", "model.layers.12.mlp.experts.89.down_proj", "model.layers.12.mlp.experts.90.down_proj", "model.layers.12.mlp.experts.91.down_proj", "model.layers.12.mlp.experts.92.down_proj", "model.layers.12.mlp.experts.93.down_proj", "model.layers.12.mlp.experts.94.down_proj", "model.layers.12.mlp.experts.95.down_proj", "model.layers.12.mlp.experts.96.down_proj", "model.layers.12.mlp.experts.97.down_proj", "model.layers.12.mlp.experts.98.down_proj", "model.layers.12.mlp.experts.99.down_proj", "model.layers.12.mlp.experts.100.down_proj", "model.layers.12.mlp.experts.101.down_proj", "model.layers.12.mlp.experts.102.down_proj", "model.layers.12.mlp.experts.103.down_proj", "model.layers.12.mlp.experts.104.down_proj", "model.layers.12.mlp.experts.105.down_proj", "model.layers.12.mlp.experts.106.down_proj", "model.layers.12.mlp.experts.107.down_proj", "model.layers.12.mlp.experts.108.down_proj", "model.layers.12.mlp.experts.109.down_proj", "model.layers.12.mlp.experts.110.down_proj", "model.layers.12.mlp.experts.111.down_proj", "model.layers.12.mlp.experts.112.down_proj", "model.layers.12.mlp.experts.113.down_proj", "model.layers.12.mlp.experts.114.down_proj", "model.layers.12.mlp.experts.115.down_proj", "model.layers.12.mlp.experts.116.down_proj", "model.layers.12.mlp.experts.117.down_proj", "model.layers.12.mlp.experts.118.down_proj", "model.layers.12.mlp.experts.119.down_proj", "model.layers.12.mlp.experts.120.down_proj", "model.layers.12.mlp.experts.121.down_proj", "model.layers.12.mlp.experts.122.down_proj", "model.layers.12.mlp.experts.123.down_proj", "model.layers.12.mlp.experts.124.down_proj", "model.layers.12.mlp.experts.125.down_proj", "model.layers.12.mlp.experts.126.down_proj", "model.layers.12.mlp.experts.127.down_proj", "model.layers.12.mlp.experts.128.down_proj", "model.layers.12.mlp.experts.129.down_proj", "model.layers.12.mlp.experts.130.down_proj", "model.layers.12.mlp.experts.131.down_proj", "model.layers.12.mlp.experts.132.down_proj", "model.layers.12.mlp.experts.133.down_proj", "model.layers.12.mlp.experts.134.down_proj", "model.layers.12.mlp.experts.135.down_proj", "model.layers.12.mlp.experts.136.down_proj", "model.layers.12.mlp.experts.137.down_proj", "model.layers.12.mlp.experts.138.down_proj", "model.layers.12.mlp.experts.139.down_proj", "model.layers.12.mlp.experts.140.down_proj", "model.layers.12.mlp.experts.141.down_proj", "model.layers.12.mlp.experts.142.down_proj", "model.layers.12.mlp.experts.143.down_proj", "model.layers.12.mlp.experts.144.down_proj", "model.layers.12.mlp.experts.145.down_proj", "model.layers.12.mlp.experts.146.down_proj", "model.layers.12.mlp.experts.147.down_proj", "model.layers.12.mlp.experts.148.down_proj", "model.layers.12.mlp.experts.149.down_proj", "model.layers.12.mlp.experts.150.down_proj", "model.layers.12.mlp.experts.151.down_proj", "model.layers.12.mlp.experts.152.down_proj", "model.layers.12.mlp.experts.153.down_proj", "model.layers.12.mlp.experts.154.down_proj", "model.layers.12.mlp.experts.155.down_proj", "model.layers.12.mlp.experts.156.down_proj", "model.layers.12.mlp.experts.157.down_proj", "model.layers.12.mlp.experts.158.down_proj", "model.layers.12.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0007188213989138686, "dbits": 1258291200 }, { "dkld": -0.0007659700699150646, "dbits": 2516582400 }, { "dkld": -0.0008581776171922711, "dbits": 3774873600 }, { "dkld": -0.0008765665814280538, "dbits": 6291456000 } ] }, { "idx": 75, "layers": [ "model.layers.13.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00022556465119123736, "dbits": 62914560 }, { "dkld": -0.00032635349780321676, "dbits": 125829120 }, { "dkld": -0.0004628953523933971, "dbits": 188743680 }, { "dkld": -0.0005825281143188588, "dbits": 314572800 } ] }, { "idx": 76, "layers": [ "model.layers.13.self_attn.k_proj", "model.layers.13.self_attn.v_proj" ], "candidates": [ { "dkld": -0.001560320146381855, "dbits": 10485760 }, { "dkld": -0.0012620065361261396, "dbits": 20971520 }, { "dkld": -0.0011569494381546974, "dbits": 31457280 }, { "dkld": -0.0011057201772928238, "dbits": 52428800 } ] }, { "idx": 77, "layers": [ "model.layers.13.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0008718227967619979, "dbits": 62914560 }, { "dkld": -0.0007964862510562026, "dbits": 125829120 }, { "dkld": -0.0008538449183106422, "dbits": 188743680 }, { "dkld": -0.0008410945534706171, "dbits": 314572800 } ] }, { "idx": 78, "layers": [ "model.layers.13.mlp.shared_experts.gate_proj", "model.layers.13.mlp.shared_experts.up_proj", "model.layers.13.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005421032197773512, "dbits": 23592960 }, { "dkld": 0.0003183164633810409, "dbits": 47185920 }, { "dkld": 0.0002850243821740095, "dbits": 70778880 }, { "dkld": 0.00022390466183423718, "dbits": 117964800 } ] }, { "idx": 79, "layers": [ "model.layers.13.mlp.experts.0.gate_proj", "model.layers.13.mlp.experts.1.gate_proj", "model.layers.13.mlp.experts.2.gate_proj", "model.layers.13.mlp.experts.3.gate_proj", "model.layers.13.mlp.experts.4.gate_proj", "model.layers.13.mlp.experts.5.gate_proj", "model.layers.13.mlp.experts.6.gate_proj", "model.layers.13.mlp.experts.7.gate_proj", "model.layers.13.mlp.experts.8.gate_proj", "model.layers.13.mlp.experts.9.gate_proj", "model.layers.13.mlp.experts.10.gate_proj", "model.layers.13.mlp.experts.11.gate_proj", "model.layers.13.mlp.experts.12.gate_proj", "model.layers.13.mlp.experts.13.gate_proj", "model.layers.13.mlp.experts.14.gate_proj", "model.layers.13.mlp.experts.15.gate_proj", "model.layers.13.mlp.experts.16.gate_proj", "model.layers.13.mlp.experts.17.gate_proj", "model.layers.13.mlp.experts.18.gate_proj", "model.layers.13.mlp.experts.19.gate_proj", "model.layers.13.mlp.experts.20.gate_proj", "model.layers.13.mlp.experts.21.gate_proj", "model.layers.13.mlp.experts.22.gate_proj", "model.layers.13.mlp.experts.23.gate_proj", "model.layers.13.mlp.experts.24.gate_proj", "model.layers.13.mlp.experts.25.gate_proj", "model.layers.13.mlp.experts.26.gate_proj", "model.layers.13.mlp.experts.27.gate_proj", "model.layers.13.mlp.experts.28.gate_proj", "model.layers.13.mlp.experts.29.gate_proj", "model.layers.13.mlp.experts.30.gate_proj", "model.layers.13.mlp.experts.31.gate_proj", "model.layers.13.mlp.experts.32.gate_proj", "model.layers.13.mlp.experts.33.gate_proj", "model.layers.13.mlp.experts.34.gate_proj", "model.layers.13.mlp.experts.35.gate_proj", "model.layers.13.mlp.experts.36.gate_proj", "model.layers.13.mlp.experts.37.gate_proj", "model.layers.13.mlp.experts.38.gate_proj", "model.layers.13.mlp.experts.39.gate_proj", "model.layers.13.mlp.experts.40.gate_proj", "model.layers.13.mlp.experts.41.gate_proj", "model.layers.13.mlp.experts.42.gate_proj", "model.layers.13.mlp.experts.43.gate_proj", "model.layers.13.mlp.experts.44.gate_proj", "model.layers.13.mlp.experts.45.gate_proj", "model.layers.13.mlp.experts.46.gate_proj", "model.layers.13.mlp.experts.47.gate_proj", "model.layers.13.mlp.experts.48.gate_proj", "model.layers.13.mlp.experts.49.gate_proj", "model.layers.13.mlp.experts.50.gate_proj", "model.layers.13.mlp.experts.51.gate_proj", "model.layers.13.mlp.experts.52.gate_proj", "model.layers.13.mlp.experts.53.gate_proj", "model.layers.13.mlp.experts.54.gate_proj", "model.layers.13.mlp.experts.55.gate_proj", "model.layers.13.mlp.experts.56.gate_proj", "model.layers.13.mlp.experts.57.gate_proj", "model.layers.13.mlp.experts.58.gate_proj", "model.layers.13.mlp.experts.59.gate_proj", "model.layers.13.mlp.experts.60.gate_proj", "model.layers.13.mlp.experts.61.gate_proj", "model.layers.13.mlp.experts.62.gate_proj", "model.layers.13.mlp.experts.63.gate_proj", "model.layers.13.mlp.experts.64.gate_proj", "model.layers.13.mlp.experts.65.gate_proj", "model.layers.13.mlp.experts.66.gate_proj", "model.layers.13.mlp.experts.67.gate_proj", "model.layers.13.mlp.experts.68.gate_proj", "model.layers.13.mlp.experts.69.gate_proj", "model.layers.13.mlp.experts.70.gate_proj", "model.layers.13.mlp.experts.71.gate_proj", "model.layers.13.mlp.experts.72.gate_proj", "model.layers.13.mlp.experts.73.gate_proj", "model.layers.13.mlp.experts.74.gate_proj", "model.layers.13.mlp.experts.75.gate_proj", "model.layers.13.mlp.experts.76.gate_proj", "model.layers.13.mlp.experts.77.gate_proj", "model.layers.13.mlp.experts.78.gate_proj", "model.layers.13.mlp.experts.79.gate_proj", "model.layers.13.mlp.experts.80.gate_proj", "model.layers.13.mlp.experts.81.gate_proj", "model.layers.13.mlp.experts.82.gate_proj", "model.layers.13.mlp.experts.83.gate_proj", "model.layers.13.mlp.experts.84.gate_proj", "model.layers.13.mlp.experts.85.gate_proj", "model.layers.13.mlp.experts.86.gate_proj", "model.layers.13.mlp.experts.87.gate_proj", "model.layers.13.mlp.experts.88.gate_proj", "model.layers.13.mlp.experts.89.gate_proj", "model.layers.13.mlp.experts.90.gate_proj", "model.layers.13.mlp.experts.91.gate_proj", "model.layers.13.mlp.experts.92.gate_proj", "model.layers.13.mlp.experts.93.gate_proj", "model.layers.13.mlp.experts.94.gate_proj", "model.layers.13.mlp.experts.95.gate_proj", "model.layers.13.mlp.experts.96.gate_proj", "model.layers.13.mlp.experts.97.gate_proj", "model.layers.13.mlp.experts.98.gate_proj", "model.layers.13.mlp.experts.99.gate_proj", "model.layers.13.mlp.experts.100.gate_proj", "model.layers.13.mlp.experts.101.gate_proj", "model.layers.13.mlp.experts.102.gate_proj", "model.layers.13.mlp.experts.103.gate_proj", "model.layers.13.mlp.experts.104.gate_proj", "model.layers.13.mlp.experts.105.gate_proj", "model.layers.13.mlp.experts.106.gate_proj", "model.layers.13.mlp.experts.107.gate_proj", "model.layers.13.mlp.experts.108.gate_proj", "model.layers.13.mlp.experts.109.gate_proj", "model.layers.13.mlp.experts.110.gate_proj", "model.layers.13.mlp.experts.111.gate_proj", "model.layers.13.mlp.experts.112.gate_proj", "model.layers.13.mlp.experts.113.gate_proj", "model.layers.13.mlp.experts.114.gate_proj", "model.layers.13.mlp.experts.115.gate_proj", "model.layers.13.mlp.experts.116.gate_proj", "model.layers.13.mlp.experts.117.gate_proj", "model.layers.13.mlp.experts.118.gate_proj", "model.layers.13.mlp.experts.119.gate_proj", "model.layers.13.mlp.experts.120.gate_proj", "model.layers.13.mlp.experts.121.gate_proj", "model.layers.13.mlp.experts.122.gate_proj", "model.layers.13.mlp.experts.123.gate_proj", "model.layers.13.mlp.experts.124.gate_proj", "model.layers.13.mlp.experts.125.gate_proj", "model.layers.13.mlp.experts.126.gate_proj", "model.layers.13.mlp.experts.127.gate_proj", "model.layers.13.mlp.experts.128.gate_proj", "model.layers.13.mlp.experts.129.gate_proj", "model.layers.13.mlp.experts.130.gate_proj", "model.layers.13.mlp.experts.131.gate_proj", "model.layers.13.mlp.experts.132.gate_proj", "model.layers.13.mlp.experts.133.gate_proj", "model.layers.13.mlp.experts.134.gate_proj", "model.layers.13.mlp.experts.135.gate_proj", "model.layers.13.mlp.experts.136.gate_proj", "model.layers.13.mlp.experts.137.gate_proj", "model.layers.13.mlp.experts.138.gate_proj", "model.layers.13.mlp.experts.139.gate_proj", "model.layers.13.mlp.experts.140.gate_proj", "model.layers.13.mlp.experts.141.gate_proj", "model.layers.13.mlp.experts.142.gate_proj", "model.layers.13.mlp.experts.143.gate_proj", "model.layers.13.mlp.experts.144.gate_proj", "model.layers.13.mlp.experts.145.gate_proj", "model.layers.13.mlp.experts.146.gate_proj", "model.layers.13.mlp.experts.147.gate_proj", "model.layers.13.mlp.experts.148.gate_proj", "model.layers.13.mlp.experts.149.gate_proj", "model.layers.13.mlp.experts.150.gate_proj", "model.layers.13.mlp.experts.151.gate_proj", "model.layers.13.mlp.experts.152.gate_proj", "model.layers.13.mlp.experts.153.gate_proj", "model.layers.13.mlp.experts.154.gate_proj", "model.layers.13.mlp.experts.155.gate_proj", "model.layers.13.mlp.experts.156.gate_proj", "model.layers.13.mlp.experts.157.gate_proj", "model.layers.13.mlp.experts.158.gate_proj", "model.layers.13.mlp.experts.159.gate_proj", "model.layers.13.mlp.experts.0.up_proj", "model.layers.13.mlp.experts.1.up_proj", "model.layers.13.mlp.experts.2.up_proj", "model.layers.13.mlp.experts.3.up_proj", "model.layers.13.mlp.experts.4.up_proj", "model.layers.13.mlp.experts.5.up_proj", "model.layers.13.mlp.experts.6.up_proj", "model.layers.13.mlp.experts.7.up_proj", "model.layers.13.mlp.experts.8.up_proj", "model.layers.13.mlp.experts.9.up_proj", "model.layers.13.mlp.experts.10.up_proj", "model.layers.13.mlp.experts.11.up_proj", "model.layers.13.mlp.experts.12.up_proj", "model.layers.13.mlp.experts.13.up_proj", "model.layers.13.mlp.experts.14.up_proj", "model.layers.13.mlp.experts.15.up_proj", "model.layers.13.mlp.experts.16.up_proj", "model.layers.13.mlp.experts.17.up_proj", "model.layers.13.mlp.experts.18.up_proj", "model.layers.13.mlp.experts.19.up_proj", "model.layers.13.mlp.experts.20.up_proj", "model.layers.13.mlp.experts.21.up_proj", "model.layers.13.mlp.experts.22.up_proj", "model.layers.13.mlp.experts.23.up_proj", "model.layers.13.mlp.experts.24.up_proj", "model.layers.13.mlp.experts.25.up_proj", "model.layers.13.mlp.experts.26.up_proj", "model.layers.13.mlp.experts.27.up_proj", "model.layers.13.mlp.experts.28.up_proj", "model.layers.13.mlp.experts.29.up_proj", "model.layers.13.mlp.experts.30.up_proj", "model.layers.13.mlp.experts.31.up_proj", "model.layers.13.mlp.experts.32.up_proj", "model.layers.13.mlp.experts.33.up_proj", "model.layers.13.mlp.experts.34.up_proj", "model.layers.13.mlp.experts.35.up_proj", "model.layers.13.mlp.experts.36.up_proj", "model.layers.13.mlp.experts.37.up_proj", "model.layers.13.mlp.experts.38.up_proj", "model.layers.13.mlp.experts.39.up_proj", "model.layers.13.mlp.experts.40.up_proj", "model.layers.13.mlp.experts.41.up_proj", "model.layers.13.mlp.experts.42.up_proj", "model.layers.13.mlp.experts.43.up_proj", "model.layers.13.mlp.experts.44.up_proj", "model.layers.13.mlp.experts.45.up_proj", "model.layers.13.mlp.experts.46.up_proj", "model.layers.13.mlp.experts.47.up_proj", "model.layers.13.mlp.experts.48.up_proj", "model.layers.13.mlp.experts.49.up_proj", "model.layers.13.mlp.experts.50.up_proj", "model.layers.13.mlp.experts.51.up_proj", "model.layers.13.mlp.experts.52.up_proj", "model.layers.13.mlp.experts.53.up_proj", "model.layers.13.mlp.experts.54.up_proj", "model.layers.13.mlp.experts.55.up_proj", "model.layers.13.mlp.experts.56.up_proj", "model.layers.13.mlp.experts.57.up_proj", "model.layers.13.mlp.experts.58.up_proj", "model.layers.13.mlp.experts.59.up_proj", "model.layers.13.mlp.experts.60.up_proj", "model.layers.13.mlp.experts.61.up_proj", "model.layers.13.mlp.experts.62.up_proj", "model.layers.13.mlp.experts.63.up_proj", "model.layers.13.mlp.experts.64.up_proj", "model.layers.13.mlp.experts.65.up_proj", "model.layers.13.mlp.experts.66.up_proj", "model.layers.13.mlp.experts.67.up_proj", "model.layers.13.mlp.experts.68.up_proj", "model.layers.13.mlp.experts.69.up_proj", "model.layers.13.mlp.experts.70.up_proj", "model.layers.13.mlp.experts.71.up_proj", "model.layers.13.mlp.experts.72.up_proj", "model.layers.13.mlp.experts.73.up_proj", "model.layers.13.mlp.experts.74.up_proj", "model.layers.13.mlp.experts.75.up_proj", "model.layers.13.mlp.experts.76.up_proj", "model.layers.13.mlp.experts.77.up_proj", "model.layers.13.mlp.experts.78.up_proj", "model.layers.13.mlp.experts.79.up_proj", "model.layers.13.mlp.experts.80.up_proj", "model.layers.13.mlp.experts.81.up_proj", "model.layers.13.mlp.experts.82.up_proj", "model.layers.13.mlp.experts.83.up_proj", "model.layers.13.mlp.experts.84.up_proj", "model.layers.13.mlp.experts.85.up_proj", "model.layers.13.mlp.experts.86.up_proj", "model.layers.13.mlp.experts.87.up_proj", "model.layers.13.mlp.experts.88.up_proj", "model.layers.13.mlp.experts.89.up_proj", "model.layers.13.mlp.experts.90.up_proj", "model.layers.13.mlp.experts.91.up_proj", "model.layers.13.mlp.experts.92.up_proj", "model.layers.13.mlp.experts.93.up_proj", "model.layers.13.mlp.experts.94.up_proj", "model.layers.13.mlp.experts.95.up_proj", "model.layers.13.mlp.experts.96.up_proj", "model.layers.13.mlp.experts.97.up_proj", "model.layers.13.mlp.experts.98.up_proj", "model.layers.13.mlp.experts.99.up_proj", "model.layers.13.mlp.experts.100.up_proj", "model.layers.13.mlp.experts.101.up_proj", "model.layers.13.mlp.experts.102.up_proj", "model.layers.13.mlp.experts.103.up_proj", "model.layers.13.mlp.experts.104.up_proj", "model.layers.13.mlp.experts.105.up_proj", "model.layers.13.mlp.experts.106.up_proj", "model.layers.13.mlp.experts.107.up_proj", "model.layers.13.mlp.experts.108.up_proj", "model.layers.13.mlp.experts.109.up_proj", "model.layers.13.mlp.experts.110.up_proj", "model.layers.13.mlp.experts.111.up_proj", "model.layers.13.mlp.experts.112.up_proj", "model.layers.13.mlp.experts.113.up_proj", "model.layers.13.mlp.experts.114.up_proj", "model.layers.13.mlp.experts.115.up_proj", "model.layers.13.mlp.experts.116.up_proj", "model.layers.13.mlp.experts.117.up_proj", "model.layers.13.mlp.experts.118.up_proj", "model.layers.13.mlp.experts.119.up_proj", "model.layers.13.mlp.experts.120.up_proj", "model.layers.13.mlp.experts.121.up_proj", "model.layers.13.mlp.experts.122.up_proj", "model.layers.13.mlp.experts.123.up_proj", "model.layers.13.mlp.experts.124.up_proj", "model.layers.13.mlp.experts.125.up_proj", "model.layers.13.mlp.experts.126.up_proj", "model.layers.13.mlp.experts.127.up_proj", "model.layers.13.mlp.experts.128.up_proj", "model.layers.13.mlp.experts.129.up_proj", "model.layers.13.mlp.experts.130.up_proj", "model.layers.13.mlp.experts.131.up_proj", "model.layers.13.mlp.experts.132.up_proj", "model.layers.13.mlp.experts.133.up_proj", "model.layers.13.mlp.experts.134.up_proj", "model.layers.13.mlp.experts.135.up_proj", "model.layers.13.mlp.experts.136.up_proj", "model.layers.13.mlp.experts.137.up_proj", "model.layers.13.mlp.experts.138.up_proj", "model.layers.13.mlp.experts.139.up_proj", "model.layers.13.mlp.experts.140.up_proj", "model.layers.13.mlp.experts.141.up_proj", "model.layers.13.mlp.experts.142.up_proj", "model.layers.13.mlp.experts.143.up_proj", "model.layers.13.mlp.experts.144.up_proj", "model.layers.13.mlp.experts.145.up_proj", "model.layers.13.mlp.experts.146.up_proj", "model.layers.13.mlp.experts.147.up_proj", "model.layers.13.mlp.experts.148.up_proj", "model.layers.13.mlp.experts.149.up_proj", "model.layers.13.mlp.experts.150.up_proj", "model.layers.13.mlp.experts.151.up_proj", "model.layers.13.mlp.experts.152.up_proj", "model.layers.13.mlp.experts.153.up_proj", "model.layers.13.mlp.experts.154.up_proj", "model.layers.13.mlp.experts.155.up_proj", "model.layers.13.mlp.experts.156.up_proj", "model.layers.13.mlp.experts.157.up_proj", "model.layers.13.mlp.experts.158.up_proj", "model.layers.13.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0003146473318338394, "dbits": 2516582400 }, { "dkld": -0.0005077000707388035, "dbits": 5033164800 }, { "dkld": -0.0006446502171456869, "dbits": 7549747200 }, { "dkld": -0.0006204998120665633, "dbits": 12582912000 } ] }, { "idx": 80, "layers": [ "model.layers.13.mlp.experts.0.down_proj", "model.layers.13.mlp.experts.1.down_proj", "model.layers.13.mlp.experts.2.down_proj", "model.layers.13.mlp.experts.3.down_proj", "model.layers.13.mlp.experts.4.down_proj", "model.layers.13.mlp.experts.5.down_proj", "model.layers.13.mlp.experts.6.down_proj", "model.layers.13.mlp.experts.7.down_proj", "model.layers.13.mlp.experts.8.down_proj", "model.layers.13.mlp.experts.9.down_proj", "model.layers.13.mlp.experts.10.down_proj", "model.layers.13.mlp.experts.11.down_proj", "model.layers.13.mlp.experts.12.down_proj", "model.layers.13.mlp.experts.13.down_proj", "model.layers.13.mlp.experts.14.down_proj", "model.layers.13.mlp.experts.15.down_proj", "model.layers.13.mlp.experts.16.down_proj", "model.layers.13.mlp.experts.17.down_proj", "model.layers.13.mlp.experts.18.down_proj", "model.layers.13.mlp.experts.19.down_proj", "model.layers.13.mlp.experts.20.down_proj", "model.layers.13.mlp.experts.21.down_proj", "model.layers.13.mlp.experts.22.down_proj", "model.layers.13.mlp.experts.23.down_proj", "model.layers.13.mlp.experts.24.down_proj", "model.layers.13.mlp.experts.25.down_proj", "model.layers.13.mlp.experts.26.down_proj", "model.layers.13.mlp.experts.27.down_proj", "model.layers.13.mlp.experts.28.down_proj", "model.layers.13.mlp.experts.29.down_proj", "model.layers.13.mlp.experts.30.down_proj", "model.layers.13.mlp.experts.31.down_proj", "model.layers.13.mlp.experts.32.down_proj", "model.layers.13.mlp.experts.33.down_proj", "model.layers.13.mlp.experts.34.down_proj", "model.layers.13.mlp.experts.35.down_proj", "model.layers.13.mlp.experts.36.down_proj", "model.layers.13.mlp.experts.37.down_proj", "model.layers.13.mlp.experts.38.down_proj", "model.layers.13.mlp.experts.39.down_proj", "model.layers.13.mlp.experts.40.down_proj", "model.layers.13.mlp.experts.41.down_proj", "model.layers.13.mlp.experts.42.down_proj", "model.layers.13.mlp.experts.43.down_proj", "model.layers.13.mlp.experts.44.down_proj", "model.layers.13.mlp.experts.45.down_proj", "model.layers.13.mlp.experts.46.down_proj", "model.layers.13.mlp.experts.47.down_proj", "model.layers.13.mlp.experts.48.down_proj", "model.layers.13.mlp.experts.49.down_proj", "model.layers.13.mlp.experts.50.down_proj", "model.layers.13.mlp.experts.51.down_proj", "model.layers.13.mlp.experts.52.down_proj", "model.layers.13.mlp.experts.53.down_proj", "model.layers.13.mlp.experts.54.down_proj", "model.layers.13.mlp.experts.55.down_proj", "model.layers.13.mlp.experts.56.down_proj", "model.layers.13.mlp.experts.57.down_proj", "model.layers.13.mlp.experts.58.down_proj", "model.layers.13.mlp.experts.59.down_proj", "model.layers.13.mlp.experts.60.down_proj", "model.layers.13.mlp.experts.61.down_proj", "model.layers.13.mlp.experts.62.down_proj", "model.layers.13.mlp.experts.63.down_proj", "model.layers.13.mlp.experts.64.down_proj", "model.layers.13.mlp.experts.65.down_proj", "model.layers.13.mlp.experts.66.down_proj", "model.layers.13.mlp.experts.67.down_proj", "model.layers.13.mlp.experts.68.down_proj", "model.layers.13.mlp.experts.69.down_proj", "model.layers.13.mlp.experts.70.down_proj", "model.layers.13.mlp.experts.71.down_proj", "model.layers.13.mlp.experts.72.down_proj", "model.layers.13.mlp.experts.73.down_proj", "model.layers.13.mlp.experts.74.down_proj", "model.layers.13.mlp.experts.75.down_proj", "model.layers.13.mlp.experts.76.down_proj", "model.layers.13.mlp.experts.77.down_proj", "model.layers.13.mlp.experts.78.down_proj", "model.layers.13.mlp.experts.79.down_proj", "model.layers.13.mlp.experts.80.down_proj", "model.layers.13.mlp.experts.81.down_proj", "model.layers.13.mlp.experts.82.down_proj", "model.layers.13.mlp.experts.83.down_proj", "model.layers.13.mlp.experts.84.down_proj", "model.layers.13.mlp.experts.85.down_proj", "model.layers.13.mlp.experts.86.down_proj", "model.layers.13.mlp.experts.87.down_proj", "model.layers.13.mlp.experts.88.down_proj", "model.layers.13.mlp.experts.89.down_proj", "model.layers.13.mlp.experts.90.down_proj", "model.layers.13.mlp.experts.91.down_proj", "model.layers.13.mlp.experts.92.down_proj", "model.layers.13.mlp.experts.93.down_proj", "model.layers.13.mlp.experts.94.down_proj", "model.layers.13.mlp.experts.95.down_proj", "model.layers.13.mlp.experts.96.down_proj", "model.layers.13.mlp.experts.97.down_proj", "model.layers.13.mlp.experts.98.down_proj", "model.layers.13.mlp.experts.99.down_proj", "model.layers.13.mlp.experts.100.down_proj", "model.layers.13.mlp.experts.101.down_proj", "model.layers.13.mlp.experts.102.down_proj", "model.layers.13.mlp.experts.103.down_proj", "model.layers.13.mlp.experts.104.down_proj", "model.layers.13.mlp.experts.105.down_proj", "model.layers.13.mlp.experts.106.down_proj", "model.layers.13.mlp.experts.107.down_proj", "model.layers.13.mlp.experts.108.down_proj", "model.layers.13.mlp.experts.109.down_proj", "model.layers.13.mlp.experts.110.down_proj", "model.layers.13.mlp.experts.111.down_proj", "model.layers.13.mlp.experts.112.down_proj", "model.layers.13.mlp.experts.113.down_proj", "model.layers.13.mlp.experts.114.down_proj", "model.layers.13.mlp.experts.115.down_proj", "model.layers.13.mlp.experts.116.down_proj", "model.layers.13.mlp.experts.117.down_proj", "model.layers.13.mlp.experts.118.down_proj", "model.layers.13.mlp.experts.119.down_proj", "model.layers.13.mlp.experts.120.down_proj", "model.layers.13.mlp.experts.121.down_proj", "model.layers.13.mlp.experts.122.down_proj", "model.layers.13.mlp.experts.123.down_proj", "model.layers.13.mlp.experts.124.down_proj", "model.layers.13.mlp.experts.125.down_proj", "model.layers.13.mlp.experts.126.down_proj", "model.layers.13.mlp.experts.127.down_proj", "model.layers.13.mlp.experts.128.down_proj", "model.layers.13.mlp.experts.129.down_proj", "model.layers.13.mlp.experts.130.down_proj", "model.layers.13.mlp.experts.131.down_proj", "model.layers.13.mlp.experts.132.down_proj", "model.layers.13.mlp.experts.133.down_proj", "model.layers.13.mlp.experts.134.down_proj", "model.layers.13.mlp.experts.135.down_proj", "model.layers.13.mlp.experts.136.down_proj", "model.layers.13.mlp.experts.137.down_proj", "model.layers.13.mlp.experts.138.down_proj", "model.layers.13.mlp.experts.139.down_proj", "model.layers.13.mlp.experts.140.down_proj", "model.layers.13.mlp.experts.141.down_proj", "model.layers.13.mlp.experts.142.down_proj", "model.layers.13.mlp.experts.143.down_proj", "model.layers.13.mlp.experts.144.down_proj", "model.layers.13.mlp.experts.145.down_proj", "model.layers.13.mlp.experts.146.down_proj", "model.layers.13.mlp.experts.147.down_proj", "model.layers.13.mlp.experts.148.down_proj", "model.layers.13.mlp.experts.149.down_proj", "model.layers.13.mlp.experts.150.down_proj", "model.layers.13.mlp.experts.151.down_proj", "model.layers.13.mlp.experts.152.down_proj", "model.layers.13.mlp.experts.153.down_proj", "model.layers.13.mlp.experts.154.down_proj", "model.layers.13.mlp.experts.155.down_proj", "model.layers.13.mlp.experts.156.down_proj", "model.layers.13.mlp.experts.157.down_proj", "model.layers.13.mlp.experts.158.down_proj", "model.layers.13.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003412343561649406, "dbits": 1258291200 }, { "dkld": -0.0006413660012185574, "dbits": 2516582400 }, { "dkld": -0.0006615659222006798, "dbits": 3774873600 }, { "dkld": -0.0004897413775324932, "dbits": 6291456000 } ] }, { "idx": 81, "layers": [ "model.layers.14.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00017539057880640307, "dbits": 62914560 }, { "dkld": -4.0668621659287196e-05, "dbits": 125829120 }, { "dkld": -0.00011349916458130993, "dbits": 188743680 }, { "dkld": -0.0002624135464429911, "dbits": 314572800 } ] }, { "idx": 82, "layers": [ "model.layers.14.self_attn.k_proj", "model.layers.14.self_attn.v_proj" ], "candidates": [ { "dkld": -0.000247276574373248, "dbits": 10485760 }, { "dkld": 0.00013315007090568265, "dbits": 20971520 }, { "dkld": 2.8476864099499788e-05, "dbits": 31457280 }, { "dkld": 0.0001638818532228442, "dbits": 52428800 } ] }, { "idx": 83, "layers": [ "model.layers.14.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00036756712943315784, "dbits": 62914560 }, { "dkld": -0.0008674360811710413, "dbits": 125829120 }, { "dkld": -0.000769883394241333, "dbits": 188743680 }, { "dkld": -0.0008361891843378655, "dbits": 314572800 } ] }, { "idx": 84, "layers": [ "model.layers.14.mlp.shared_experts.gate_proj", "model.layers.14.mlp.shared_experts.up_proj", "model.layers.14.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 3.894222900270583e-05, "dbits": 23592960 }, { "dkld": 6.427448242901646e-05, "dbits": 47185920 }, { "dkld": -6.703957915306369e-05, "dbits": 70778880 }, { "dkld": -0.00019377181306481917, "dbits": 117964800 } ] }, { "idx": 85, "layers": [ "model.layers.14.mlp.experts.0.gate_proj", "model.layers.14.mlp.experts.1.gate_proj", "model.layers.14.mlp.experts.2.gate_proj", "model.layers.14.mlp.experts.3.gate_proj", "model.layers.14.mlp.experts.4.gate_proj", "model.layers.14.mlp.experts.5.gate_proj", "model.layers.14.mlp.experts.6.gate_proj", "model.layers.14.mlp.experts.7.gate_proj", "model.layers.14.mlp.experts.8.gate_proj", "model.layers.14.mlp.experts.9.gate_proj", "model.layers.14.mlp.experts.10.gate_proj", "model.layers.14.mlp.experts.11.gate_proj", "model.layers.14.mlp.experts.12.gate_proj", "model.layers.14.mlp.experts.13.gate_proj", "model.layers.14.mlp.experts.14.gate_proj", "model.layers.14.mlp.experts.15.gate_proj", "model.layers.14.mlp.experts.16.gate_proj", "model.layers.14.mlp.experts.17.gate_proj", "model.layers.14.mlp.experts.18.gate_proj", "model.layers.14.mlp.experts.19.gate_proj", "model.layers.14.mlp.experts.20.gate_proj", "model.layers.14.mlp.experts.21.gate_proj", "model.layers.14.mlp.experts.22.gate_proj", "model.layers.14.mlp.experts.23.gate_proj", "model.layers.14.mlp.experts.24.gate_proj", "model.layers.14.mlp.experts.25.gate_proj", "model.layers.14.mlp.experts.26.gate_proj", "model.layers.14.mlp.experts.27.gate_proj", "model.layers.14.mlp.experts.28.gate_proj", "model.layers.14.mlp.experts.29.gate_proj", "model.layers.14.mlp.experts.30.gate_proj", "model.layers.14.mlp.experts.31.gate_proj", "model.layers.14.mlp.experts.32.gate_proj", "model.layers.14.mlp.experts.33.gate_proj", "model.layers.14.mlp.experts.34.gate_proj", "model.layers.14.mlp.experts.35.gate_proj", "model.layers.14.mlp.experts.36.gate_proj", "model.layers.14.mlp.experts.37.gate_proj", "model.layers.14.mlp.experts.38.gate_proj", "model.layers.14.mlp.experts.39.gate_proj", "model.layers.14.mlp.experts.40.gate_proj", "model.layers.14.mlp.experts.41.gate_proj", "model.layers.14.mlp.experts.42.gate_proj", "model.layers.14.mlp.experts.43.gate_proj", "model.layers.14.mlp.experts.44.gate_proj", "model.layers.14.mlp.experts.45.gate_proj", "model.layers.14.mlp.experts.46.gate_proj", "model.layers.14.mlp.experts.47.gate_proj", "model.layers.14.mlp.experts.48.gate_proj", "model.layers.14.mlp.experts.49.gate_proj", "model.layers.14.mlp.experts.50.gate_proj", "model.layers.14.mlp.experts.51.gate_proj", "model.layers.14.mlp.experts.52.gate_proj", "model.layers.14.mlp.experts.53.gate_proj", "model.layers.14.mlp.experts.54.gate_proj", "model.layers.14.mlp.experts.55.gate_proj", "model.layers.14.mlp.experts.56.gate_proj", "model.layers.14.mlp.experts.57.gate_proj", "model.layers.14.mlp.experts.58.gate_proj", "model.layers.14.mlp.experts.59.gate_proj", "model.layers.14.mlp.experts.60.gate_proj", "model.layers.14.mlp.experts.61.gate_proj", "model.layers.14.mlp.experts.62.gate_proj", "model.layers.14.mlp.experts.63.gate_proj", "model.layers.14.mlp.experts.64.gate_proj", "model.layers.14.mlp.experts.65.gate_proj", "model.layers.14.mlp.experts.66.gate_proj", "model.layers.14.mlp.experts.67.gate_proj", "model.layers.14.mlp.experts.68.gate_proj", "model.layers.14.mlp.experts.69.gate_proj", "model.layers.14.mlp.experts.70.gate_proj", "model.layers.14.mlp.experts.71.gate_proj", "model.layers.14.mlp.experts.72.gate_proj", "model.layers.14.mlp.experts.73.gate_proj", "model.layers.14.mlp.experts.74.gate_proj", "model.layers.14.mlp.experts.75.gate_proj", "model.layers.14.mlp.experts.76.gate_proj", "model.layers.14.mlp.experts.77.gate_proj", "model.layers.14.mlp.experts.78.gate_proj", "model.layers.14.mlp.experts.79.gate_proj", "model.layers.14.mlp.experts.80.gate_proj", "model.layers.14.mlp.experts.81.gate_proj", "model.layers.14.mlp.experts.82.gate_proj", "model.layers.14.mlp.experts.83.gate_proj", "model.layers.14.mlp.experts.84.gate_proj", "model.layers.14.mlp.experts.85.gate_proj", "model.layers.14.mlp.experts.86.gate_proj", "model.layers.14.mlp.experts.87.gate_proj", "model.layers.14.mlp.experts.88.gate_proj", "model.layers.14.mlp.experts.89.gate_proj", "model.layers.14.mlp.experts.90.gate_proj", "model.layers.14.mlp.experts.91.gate_proj", "model.layers.14.mlp.experts.92.gate_proj", "model.layers.14.mlp.experts.93.gate_proj", "model.layers.14.mlp.experts.94.gate_proj", "model.layers.14.mlp.experts.95.gate_proj", "model.layers.14.mlp.experts.96.gate_proj", "model.layers.14.mlp.experts.97.gate_proj", "model.layers.14.mlp.experts.98.gate_proj", "model.layers.14.mlp.experts.99.gate_proj", "model.layers.14.mlp.experts.100.gate_proj", "model.layers.14.mlp.experts.101.gate_proj", "model.layers.14.mlp.experts.102.gate_proj", "model.layers.14.mlp.experts.103.gate_proj", "model.layers.14.mlp.experts.104.gate_proj", "model.layers.14.mlp.experts.105.gate_proj", "model.layers.14.mlp.experts.106.gate_proj", "model.layers.14.mlp.experts.107.gate_proj", "model.layers.14.mlp.experts.108.gate_proj", "model.layers.14.mlp.experts.109.gate_proj", "model.layers.14.mlp.experts.110.gate_proj", "model.layers.14.mlp.experts.111.gate_proj", "model.layers.14.mlp.experts.112.gate_proj", "model.layers.14.mlp.experts.113.gate_proj", "model.layers.14.mlp.experts.114.gate_proj", "model.layers.14.mlp.experts.115.gate_proj", "model.layers.14.mlp.experts.116.gate_proj", "model.layers.14.mlp.experts.117.gate_proj", "model.layers.14.mlp.experts.118.gate_proj", "model.layers.14.mlp.experts.119.gate_proj", "model.layers.14.mlp.experts.120.gate_proj", "model.layers.14.mlp.experts.121.gate_proj", "model.layers.14.mlp.experts.122.gate_proj", "model.layers.14.mlp.experts.123.gate_proj", "model.layers.14.mlp.experts.124.gate_proj", "model.layers.14.mlp.experts.125.gate_proj", "model.layers.14.mlp.experts.126.gate_proj", "model.layers.14.mlp.experts.127.gate_proj", "model.layers.14.mlp.experts.128.gate_proj", "model.layers.14.mlp.experts.129.gate_proj", "model.layers.14.mlp.experts.130.gate_proj", "model.layers.14.mlp.experts.131.gate_proj", "model.layers.14.mlp.experts.132.gate_proj", "model.layers.14.mlp.experts.133.gate_proj", "model.layers.14.mlp.experts.134.gate_proj", "model.layers.14.mlp.experts.135.gate_proj", "model.layers.14.mlp.experts.136.gate_proj", "model.layers.14.mlp.experts.137.gate_proj", "model.layers.14.mlp.experts.138.gate_proj", "model.layers.14.mlp.experts.139.gate_proj", "model.layers.14.mlp.experts.140.gate_proj", "model.layers.14.mlp.experts.141.gate_proj", "model.layers.14.mlp.experts.142.gate_proj", "model.layers.14.mlp.experts.143.gate_proj", "model.layers.14.mlp.experts.144.gate_proj", "model.layers.14.mlp.experts.145.gate_proj", "model.layers.14.mlp.experts.146.gate_proj", "model.layers.14.mlp.experts.147.gate_proj", "model.layers.14.mlp.experts.148.gate_proj", "model.layers.14.mlp.experts.149.gate_proj", "model.layers.14.mlp.experts.150.gate_proj", "model.layers.14.mlp.experts.151.gate_proj", "model.layers.14.mlp.experts.152.gate_proj", "model.layers.14.mlp.experts.153.gate_proj", "model.layers.14.mlp.experts.154.gate_proj", "model.layers.14.mlp.experts.155.gate_proj", "model.layers.14.mlp.experts.156.gate_proj", "model.layers.14.mlp.experts.157.gate_proj", "model.layers.14.mlp.experts.158.gate_proj", "model.layers.14.mlp.experts.159.gate_proj", "model.layers.14.mlp.experts.0.up_proj", "model.layers.14.mlp.experts.1.up_proj", "model.layers.14.mlp.experts.2.up_proj", "model.layers.14.mlp.experts.3.up_proj", "model.layers.14.mlp.experts.4.up_proj", "model.layers.14.mlp.experts.5.up_proj", "model.layers.14.mlp.experts.6.up_proj", "model.layers.14.mlp.experts.7.up_proj", "model.layers.14.mlp.experts.8.up_proj", "model.layers.14.mlp.experts.9.up_proj", "model.layers.14.mlp.experts.10.up_proj", "model.layers.14.mlp.experts.11.up_proj", "model.layers.14.mlp.experts.12.up_proj", "model.layers.14.mlp.experts.13.up_proj", "model.layers.14.mlp.experts.14.up_proj", "model.layers.14.mlp.experts.15.up_proj", "model.layers.14.mlp.experts.16.up_proj", "model.layers.14.mlp.experts.17.up_proj", "model.layers.14.mlp.experts.18.up_proj", "model.layers.14.mlp.experts.19.up_proj", "model.layers.14.mlp.experts.20.up_proj", "model.layers.14.mlp.experts.21.up_proj", "model.layers.14.mlp.experts.22.up_proj", "model.layers.14.mlp.experts.23.up_proj", "model.layers.14.mlp.experts.24.up_proj", "model.layers.14.mlp.experts.25.up_proj", "model.layers.14.mlp.experts.26.up_proj", "model.layers.14.mlp.experts.27.up_proj", "model.layers.14.mlp.experts.28.up_proj", "model.layers.14.mlp.experts.29.up_proj", "model.layers.14.mlp.experts.30.up_proj", "model.layers.14.mlp.experts.31.up_proj", "model.layers.14.mlp.experts.32.up_proj", "model.layers.14.mlp.experts.33.up_proj", "model.layers.14.mlp.experts.34.up_proj", "model.layers.14.mlp.experts.35.up_proj", "model.layers.14.mlp.experts.36.up_proj", "model.layers.14.mlp.experts.37.up_proj", "model.layers.14.mlp.experts.38.up_proj", "model.layers.14.mlp.experts.39.up_proj", "model.layers.14.mlp.experts.40.up_proj", "model.layers.14.mlp.experts.41.up_proj", "model.layers.14.mlp.experts.42.up_proj", "model.layers.14.mlp.experts.43.up_proj", "model.layers.14.mlp.experts.44.up_proj", "model.layers.14.mlp.experts.45.up_proj", "model.layers.14.mlp.experts.46.up_proj", "model.layers.14.mlp.experts.47.up_proj", "model.layers.14.mlp.experts.48.up_proj", "model.layers.14.mlp.experts.49.up_proj", "model.layers.14.mlp.experts.50.up_proj", "model.layers.14.mlp.experts.51.up_proj", "model.layers.14.mlp.experts.52.up_proj", "model.layers.14.mlp.experts.53.up_proj", "model.layers.14.mlp.experts.54.up_proj", "model.layers.14.mlp.experts.55.up_proj", "model.layers.14.mlp.experts.56.up_proj", "model.layers.14.mlp.experts.57.up_proj", "model.layers.14.mlp.experts.58.up_proj", "model.layers.14.mlp.experts.59.up_proj", "model.layers.14.mlp.experts.60.up_proj", "model.layers.14.mlp.experts.61.up_proj", "model.layers.14.mlp.experts.62.up_proj", "model.layers.14.mlp.experts.63.up_proj", "model.layers.14.mlp.experts.64.up_proj", "model.layers.14.mlp.experts.65.up_proj", "model.layers.14.mlp.experts.66.up_proj", "model.layers.14.mlp.experts.67.up_proj", "model.layers.14.mlp.experts.68.up_proj", "model.layers.14.mlp.experts.69.up_proj", "model.layers.14.mlp.experts.70.up_proj", "model.layers.14.mlp.experts.71.up_proj", "model.layers.14.mlp.experts.72.up_proj", "model.layers.14.mlp.experts.73.up_proj", "model.layers.14.mlp.experts.74.up_proj", "model.layers.14.mlp.experts.75.up_proj", "model.layers.14.mlp.experts.76.up_proj", "model.layers.14.mlp.experts.77.up_proj", "model.layers.14.mlp.experts.78.up_proj", "model.layers.14.mlp.experts.79.up_proj", "model.layers.14.mlp.experts.80.up_proj", "model.layers.14.mlp.experts.81.up_proj", "model.layers.14.mlp.experts.82.up_proj", "model.layers.14.mlp.experts.83.up_proj", "model.layers.14.mlp.experts.84.up_proj", "model.layers.14.mlp.experts.85.up_proj", "model.layers.14.mlp.experts.86.up_proj", "model.layers.14.mlp.experts.87.up_proj", "model.layers.14.mlp.experts.88.up_proj", "model.layers.14.mlp.experts.89.up_proj", "model.layers.14.mlp.experts.90.up_proj", "model.layers.14.mlp.experts.91.up_proj", "model.layers.14.mlp.experts.92.up_proj", "model.layers.14.mlp.experts.93.up_proj", "model.layers.14.mlp.experts.94.up_proj", "model.layers.14.mlp.experts.95.up_proj", "model.layers.14.mlp.experts.96.up_proj", "model.layers.14.mlp.experts.97.up_proj", "model.layers.14.mlp.experts.98.up_proj", "model.layers.14.mlp.experts.99.up_proj", "model.layers.14.mlp.experts.100.up_proj", "model.layers.14.mlp.experts.101.up_proj", "model.layers.14.mlp.experts.102.up_proj", "model.layers.14.mlp.experts.103.up_proj", "model.layers.14.mlp.experts.104.up_proj", "model.layers.14.mlp.experts.105.up_proj", "model.layers.14.mlp.experts.106.up_proj", "model.layers.14.mlp.experts.107.up_proj", "model.layers.14.mlp.experts.108.up_proj", "model.layers.14.mlp.experts.109.up_proj", "model.layers.14.mlp.experts.110.up_proj", "model.layers.14.mlp.experts.111.up_proj", "model.layers.14.mlp.experts.112.up_proj", "model.layers.14.mlp.experts.113.up_proj", "model.layers.14.mlp.experts.114.up_proj", "model.layers.14.mlp.experts.115.up_proj", "model.layers.14.mlp.experts.116.up_proj", "model.layers.14.mlp.experts.117.up_proj", "model.layers.14.mlp.experts.118.up_proj", "model.layers.14.mlp.experts.119.up_proj", "model.layers.14.mlp.experts.120.up_proj", "model.layers.14.mlp.experts.121.up_proj", "model.layers.14.mlp.experts.122.up_proj", "model.layers.14.mlp.experts.123.up_proj", "model.layers.14.mlp.experts.124.up_proj", "model.layers.14.mlp.experts.125.up_proj", "model.layers.14.mlp.experts.126.up_proj", "model.layers.14.mlp.experts.127.up_proj", "model.layers.14.mlp.experts.128.up_proj", "model.layers.14.mlp.experts.129.up_proj", "model.layers.14.mlp.experts.130.up_proj", "model.layers.14.mlp.experts.131.up_proj", "model.layers.14.mlp.experts.132.up_proj", "model.layers.14.mlp.experts.133.up_proj", "model.layers.14.mlp.experts.134.up_proj", "model.layers.14.mlp.experts.135.up_proj", "model.layers.14.mlp.experts.136.up_proj", "model.layers.14.mlp.experts.137.up_proj", "model.layers.14.mlp.experts.138.up_proj", "model.layers.14.mlp.experts.139.up_proj", "model.layers.14.mlp.experts.140.up_proj", "model.layers.14.mlp.experts.141.up_proj", "model.layers.14.mlp.experts.142.up_proj", "model.layers.14.mlp.experts.143.up_proj", "model.layers.14.mlp.experts.144.up_proj", "model.layers.14.mlp.experts.145.up_proj", "model.layers.14.mlp.experts.146.up_proj", "model.layers.14.mlp.experts.147.up_proj", "model.layers.14.mlp.experts.148.up_proj", "model.layers.14.mlp.experts.149.up_proj", "model.layers.14.mlp.experts.150.up_proj", "model.layers.14.mlp.experts.151.up_proj", "model.layers.14.mlp.experts.152.up_proj", "model.layers.14.mlp.experts.153.up_proj", "model.layers.14.mlp.experts.154.up_proj", "model.layers.14.mlp.experts.155.up_proj", "model.layers.14.mlp.experts.156.up_proj", "model.layers.14.mlp.experts.157.up_proj", "model.layers.14.mlp.experts.158.up_proj", "model.layers.14.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00070933047682048, "dbits": 2516582400 }, { "dkld": -0.0009640314616262968, "dbits": 5033164800 }, { "dkld": -0.0008902505971491448, "dbits": 7549747200 }, { "dkld": -0.0009000447578728227, "dbits": 12582912000 } ] }, { "idx": 86, "layers": [ "model.layers.14.mlp.experts.0.down_proj", "model.layers.14.mlp.experts.1.down_proj", "model.layers.14.mlp.experts.2.down_proj", "model.layers.14.mlp.experts.3.down_proj", "model.layers.14.mlp.experts.4.down_proj", "model.layers.14.mlp.experts.5.down_proj", "model.layers.14.mlp.experts.6.down_proj", "model.layers.14.mlp.experts.7.down_proj", "model.layers.14.mlp.experts.8.down_proj", "model.layers.14.mlp.experts.9.down_proj", "model.layers.14.mlp.experts.10.down_proj", "model.layers.14.mlp.experts.11.down_proj", "model.layers.14.mlp.experts.12.down_proj", "model.layers.14.mlp.experts.13.down_proj", "model.layers.14.mlp.experts.14.down_proj", "model.layers.14.mlp.experts.15.down_proj", "model.layers.14.mlp.experts.16.down_proj", "model.layers.14.mlp.experts.17.down_proj", "model.layers.14.mlp.experts.18.down_proj", "model.layers.14.mlp.experts.19.down_proj", "model.layers.14.mlp.experts.20.down_proj", "model.layers.14.mlp.experts.21.down_proj", "model.layers.14.mlp.experts.22.down_proj", "model.layers.14.mlp.experts.23.down_proj", "model.layers.14.mlp.experts.24.down_proj", "model.layers.14.mlp.experts.25.down_proj", "model.layers.14.mlp.experts.26.down_proj", "model.layers.14.mlp.experts.27.down_proj", "model.layers.14.mlp.experts.28.down_proj", "model.layers.14.mlp.experts.29.down_proj", "model.layers.14.mlp.experts.30.down_proj", "model.layers.14.mlp.experts.31.down_proj", "model.layers.14.mlp.experts.32.down_proj", "model.layers.14.mlp.experts.33.down_proj", "model.layers.14.mlp.experts.34.down_proj", "model.layers.14.mlp.experts.35.down_proj", "model.layers.14.mlp.experts.36.down_proj", "model.layers.14.mlp.experts.37.down_proj", "model.layers.14.mlp.experts.38.down_proj", "model.layers.14.mlp.experts.39.down_proj", "model.layers.14.mlp.experts.40.down_proj", "model.layers.14.mlp.experts.41.down_proj", "model.layers.14.mlp.experts.42.down_proj", "model.layers.14.mlp.experts.43.down_proj", "model.layers.14.mlp.experts.44.down_proj", "model.layers.14.mlp.experts.45.down_proj", "model.layers.14.mlp.experts.46.down_proj", "model.layers.14.mlp.experts.47.down_proj", "model.layers.14.mlp.experts.48.down_proj", "model.layers.14.mlp.experts.49.down_proj", "model.layers.14.mlp.experts.50.down_proj", "model.layers.14.mlp.experts.51.down_proj", "model.layers.14.mlp.experts.52.down_proj", "model.layers.14.mlp.experts.53.down_proj", "model.layers.14.mlp.experts.54.down_proj", "model.layers.14.mlp.experts.55.down_proj", "model.layers.14.mlp.experts.56.down_proj", "model.layers.14.mlp.experts.57.down_proj", "model.layers.14.mlp.experts.58.down_proj", "model.layers.14.mlp.experts.59.down_proj", "model.layers.14.mlp.experts.60.down_proj", "model.layers.14.mlp.experts.61.down_proj", "model.layers.14.mlp.experts.62.down_proj", "model.layers.14.mlp.experts.63.down_proj", "model.layers.14.mlp.experts.64.down_proj", "model.layers.14.mlp.experts.65.down_proj", "model.layers.14.mlp.experts.66.down_proj", "model.layers.14.mlp.experts.67.down_proj", "model.layers.14.mlp.experts.68.down_proj", "model.layers.14.mlp.experts.69.down_proj", "model.layers.14.mlp.experts.70.down_proj", "model.layers.14.mlp.experts.71.down_proj", "model.layers.14.mlp.experts.72.down_proj", "model.layers.14.mlp.experts.73.down_proj", "model.layers.14.mlp.experts.74.down_proj", "model.layers.14.mlp.experts.75.down_proj", "model.layers.14.mlp.experts.76.down_proj", "model.layers.14.mlp.experts.77.down_proj", "model.layers.14.mlp.experts.78.down_proj", "model.layers.14.mlp.experts.79.down_proj", "model.layers.14.mlp.experts.80.down_proj", "model.layers.14.mlp.experts.81.down_proj", "model.layers.14.mlp.experts.82.down_proj", "model.layers.14.mlp.experts.83.down_proj", "model.layers.14.mlp.experts.84.down_proj", "model.layers.14.mlp.experts.85.down_proj", "model.layers.14.mlp.experts.86.down_proj", "model.layers.14.mlp.experts.87.down_proj", "model.layers.14.mlp.experts.88.down_proj", "model.layers.14.mlp.experts.89.down_proj", "model.layers.14.mlp.experts.90.down_proj", "model.layers.14.mlp.experts.91.down_proj", "model.layers.14.mlp.experts.92.down_proj", "model.layers.14.mlp.experts.93.down_proj", "model.layers.14.mlp.experts.94.down_proj", "model.layers.14.mlp.experts.95.down_proj", "model.layers.14.mlp.experts.96.down_proj", "model.layers.14.mlp.experts.97.down_proj", "model.layers.14.mlp.experts.98.down_proj", "model.layers.14.mlp.experts.99.down_proj", "model.layers.14.mlp.experts.100.down_proj", "model.layers.14.mlp.experts.101.down_proj", "model.layers.14.mlp.experts.102.down_proj", "model.layers.14.mlp.experts.103.down_proj", "model.layers.14.mlp.experts.104.down_proj", "model.layers.14.mlp.experts.105.down_proj", "model.layers.14.mlp.experts.106.down_proj", "model.layers.14.mlp.experts.107.down_proj", "model.layers.14.mlp.experts.108.down_proj", "model.layers.14.mlp.experts.109.down_proj", "model.layers.14.mlp.experts.110.down_proj", "model.layers.14.mlp.experts.111.down_proj", "model.layers.14.mlp.experts.112.down_proj", "model.layers.14.mlp.experts.113.down_proj", "model.layers.14.mlp.experts.114.down_proj", "model.layers.14.mlp.experts.115.down_proj", "model.layers.14.mlp.experts.116.down_proj", "model.layers.14.mlp.experts.117.down_proj", "model.layers.14.mlp.experts.118.down_proj", "model.layers.14.mlp.experts.119.down_proj", "model.layers.14.mlp.experts.120.down_proj", "model.layers.14.mlp.experts.121.down_proj", "model.layers.14.mlp.experts.122.down_proj", "model.layers.14.mlp.experts.123.down_proj", "model.layers.14.mlp.experts.124.down_proj", "model.layers.14.mlp.experts.125.down_proj", "model.layers.14.mlp.experts.126.down_proj", "model.layers.14.mlp.experts.127.down_proj", "model.layers.14.mlp.experts.128.down_proj", "model.layers.14.mlp.experts.129.down_proj", "model.layers.14.mlp.experts.130.down_proj", "model.layers.14.mlp.experts.131.down_proj", "model.layers.14.mlp.experts.132.down_proj", "model.layers.14.mlp.experts.133.down_proj", "model.layers.14.mlp.experts.134.down_proj", "model.layers.14.mlp.experts.135.down_proj", "model.layers.14.mlp.experts.136.down_proj", "model.layers.14.mlp.experts.137.down_proj", "model.layers.14.mlp.experts.138.down_proj", "model.layers.14.mlp.experts.139.down_proj", "model.layers.14.mlp.experts.140.down_proj", "model.layers.14.mlp.experts.141.down_proj", "model.layers.14.mlp.experts.142.down_proj", "model.layers.14.mlp.experts.143.down_proj", "model.layers.14.mlp.experts.144.down_proj", "model.layers.14.mlp.experts.145.down_proj", "model.layers.14.mlp.experts.146.down_proj", "model.layers.14.mlp.experts.147.down_proj", "model.layers.14.mlp.experts.148.down_proj", "model.layers.14.mlp.experts.149.down_proj", "model.layers.14.mlp.experts.150.down_proj", "model.layers.14.mlp.experts.151.down_proj", "model.layers.14.mlp.experts.152.down_proj", "model.layers.14.mlp.experts.153.down_proj", "model.layers.14.mlp.experts.154.down_proj", "model.layers.14.mlp.experts.155.down_proj", "model.layers.14.mlp.experts.156.down_proj", "model.layers.14.mlp.experts.157.down_proj", "model.layers.14.mlp.experts.158.down_proj", "model.layers.14.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003753071650862694, "dbits": 1258291200 }, { "dkld": -0.0006729567423462979, "dbits": 2516582400 }, { "dkld": -0.0006041129119694344, "dbits": 3774873600 }, { "dkld": -0.0007475363090634346, "dbits": 6291456000 } ] }, { "idx": 87, "layers": [ "model.layers.15.self_attn.q_proj" ], "candidates": [ { "dkld": 8.212868124246597e-05, "dbits": 62914560 }, { "dkld": 4.549678415059766e-05, "dbits": 125829120 }, { "dkld": 3.7954747676843814e-05, "dbits": 188743680 }, { "dkld": 4.887022078037262e-05, "dbits": 314572800 } ] }, { "idx": 88, "layers": [ "model.layers.15.self_attn.k_proj", "model.layers.15.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0005389675498008728, "dbits": 10485760 }, { "dkld": 0.00031313961371778887, "dbits": 20971520 }, { "dkld": 0.00036792233586310785, "dbits": 31457280 }, { "dkld": 0.0005186359398066942, "dbits": 52428800 } ] }, { "idx": 89, "layers": [ "model.layers.15.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00012528821825981418, "dbits": 62914560 }, { "dkld": -0.00032952502369881786, "dbits": 125829120 }, { "dkld": -0.0002874404191970936, "dbits": 188743680 }, { "dkld": -0.00033569792285562117, "dbits": 314572800 } ] }, { "idx": 90, "layers": [ "model.layers.15.mlp.shared_experts.gate_proj", "model.layers.15.mlp.shared_experts.up_proj", "model.layers.15.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00016638766974211294, "dbits": 23592960 }, { "dkld": 5.781780928372504e-05, "dbits": 47185920 }, { "dkld": -9.909961372614462e-05, "dbits": 70778880 }, { "dkld": -0.00013401079922914505, "dbits": 117964800 } ] }, { "idx": 91, "layers": [ "model.layers.15.mlp.experts.0.gate_proj", "model.layers.15.mlp.experts.1.gate_proj", "model.layers.15.mlp.experts.2.gate_proj", "model.layers.15.mlp.experts.3.gate_proj", "model.layers.15.mlp.experts.4.gate_proj", "model.layers.15.mlp.experts.5.gate_proj", "model.layers.15.mlp.experts.6.gate_proj", "model.layers.15.mlp.experts.7.gate_proj", "model.layers.15.mlp.experts.8.gate_proj", "model.layers.15.mlp.experts.9.gate_proj", "model.layers.15.mlp.experts.10.gate_proj", "model.layers.15.mlp.experts.11.gate_proj", "model.layers.15.mlp.experts.12.gate_proj", "model.layers.15.mlp.experts.13.gate_proj", "model.layers.15.mlp.experts.14.gate_proj", "model.layers.15.mlp.experts.15.gate_proj", "model.layers.15.mlp.experts.16.gate_proj", "model.layers.15.mlp.experts.17.gate_proj", "model.layers.15.mlp.experts.18.gate_proj", "model.layers.15.mlp.experts.19.gate_proj", "model.layers.15.mlp.experts.20.gate_proj", "model.layers.15.mlp.experts.21.gate_proj", "model.layers.15.mlp.experts.22.gate_proj", "model.layers.15.mlp.experts.23.gate_proj", "model.layers.15.mlp.experts.24.gate_proj", "model.layers.15.mlp.experts.25.gate_proj", "model.layers.15.mlp.experts.26.gate_proj", "model.layers.15.mlp.experts.27.gate_proj", "model.layers.15.mlp.experts.28.gate_proj", "model.layers.15.mlp.experts.29.gate_proj", "model.layers.15.mlp.experts.30.gate_proj", "model.layers.15.mlp.experts.31.gate_proj", "model.layers.15.mlp.experts.32.gate_proj", "model.layers.15.mlp.experts.33.gate_proj", "model.layers.15.mlp.experts.34.gate_proj", "model.layers.15.mlp.experts.35.gate_proj", "model.layers.15.mlp.experts.36.gate_proj", "model.layers.15.mlp.experts.37.gate_proj", "model.layers.15.mlp.experts.38.gate_proj", "model.layers.15.mlp.experts.39.gate_proj", "model.layers.15.mlp.experts.40.gate_proj", "model.layers.15.mlp.experts.41.gate_proj", "model.layers.15.mlp.experts.42.gate_proj", "model.layers.15.mlp.experts.43.gate_proj", "model.layers.15.mlp.experts.44.gate_proj", "model.layers.15.mlp.experts.45.gate_proj", "model.layers.15.mlp.experts.46.gate_proj", "model.layers.15.mlp.experts.47.gate_proj", "model.layers.15.mlp.experts.48.gate_proj", "model.layers.15.mlp.experts.49.gate_proj", "model.layers.15.mlp.experts.50.gate_proj", "model.layers.15.mlp.experts.51.gate_proj", "model.layers.15.mlp.experts.52.gate_proj", "model.layers.15.mlp.experts.53.gate_proj", "model.layers.15.mlp.experts.54.gate_proj", "model.layers.15.mlp.experts.55.gate_proj", "model.layers.15.mlp.experts.56.gate_proj", "model.layers.15.mlp.experts.57.gate_proj", "model.layers.15.mlp.experts.58.gate_proj", "model.layers.15.mlp.experts.59.gate_proj", "model.layers.15.mlp.experts.60.gate_proj", "model.layers.15.mlp.experts.61.gate_proj", "model.layers.15.mlp.experts.62.gate_proj", "model.layers.15.mlp.experts.63.gate_proj", "model.layers.15.mlp.experts.64.gate_proj", "model.layers.15.mlp.experts.65.gate_proj", "model.layers.15.mlp.experts.66.gate_proj", "model.layers.15.mlp.experts.67.gate_proj", "model.layers.15.mlp.experts.68.gate_proj", "model.layers.15.mlp.experts.69.gate_proj", "model.layers.15.mlp.experts.70.gate_proj", "model.layers.15.mlp.experts.71.gate_proj", "model.layers.15.mlp.experts.72.gate_proj", "model.layers.15.mlp.experts.73.gate_proj", "model.layers.15.mlp.experts.74.gate_proj", "model.layers.15.mlp.experts.75.gate_proj", "model.layers.15.mlp.experts.76.gate_proj", "model.layers.15.mlp.experts.77.gate_proj", "model.layers.15.mlp.experts.78.gate_proj", "model.layers.15.mlp.experts.79.gate_proj", "model.layers.15.mlp.experts.80.gate_proj", "model.layers.15.mlp.experts.81.gate_proj", "model.layers.15.mlp.experts.82.gate_proj", "model.layers.15.mlp.experts.83.gate_proj", "model.layers.15.mlp.experts.84.gate_proj", "model.layers.15.mlp.experts.85.gate_proj", "model.layers.15.mlp.experts.86.gate_proj", "model.layers.15.mlp.experts.87.gate_proj", "model.layers.15.mlp.experts.88.gate_proj", "model.layers.15.mlp.experts.89.gate_proj", "model.layers.15.mlp.experts.90.gate_proj", "model.layers.15.mlp.experts.91.gate_proj", "model.layers.15.mlp.experts.92.gate_proj", "model.layers.15.mlp.experts.93.gate_proj", "model.layers.15.mlp.experts.94.gate_proj", "model.layers.15.mlp.experts.95.gate_proj", "model.layers.15.mlp.experts.96.gate_proj", "model.layers.15.mlp.experts.97.gate_proj", "model.layers.15.mlp.experts.98.gate_proj", "model.layers.15.mlp.experts.99.gate_proj", "model.layers.15.mlp.experts.100.gate_proj", "model.layers.15.mlp.experts.101.gate_proj", "model.layers.15.mlp.experts.102.gate_proj", "model.layers.15.mlp.experts.103.gate_proj", "model.layers.15.mlp.experts.104.gate_proj", "model.layers.15.mlp.experts.105.gate_proj", "model.layers.15.mlp.experts.106.gate_proj", "model.layers.15.mlp.experts.107.gate_proj", "model.layers.15.mlp.experts.108.gate_proj", "model.layers.15.mlp.experts.109.gate_proj", "model.layers.15.mlp.experts.110.gate_proj", "model.layers.15.mlp.experts.111.gate_proj", "model.layers.15.mlp.experts.112.gate_proj", "model.layers.15.mlp.experts.113.gate_proj", "model.layers.15.mlp.experts.114.gate_proj", "model.layers.15.mlp.experts.115.gate_proj", "model.layers.15.mlp.experts.116.gate_proj", "model.layers.15.mlp.experts.117.gate_proj", "model.layers.15.mlp.experts.118.gate_proj", "model.layers.15.mlp.experts.119.gate_proj", "model.layers.15.mlp.experts.120.gate_proj", "model.layers.15.mlp.experts.121.gate_proj", "model.layers.15.mlp.experts.122.gate_proj", "model.layers.15.mlp.experts.123.gate_proj", "model.layers.15.mlp.experts.124.gate_proj", "model.layers.15.mlp.experts.125.gate_proj", "model.layers.15.mlp.experts.126.gate_proj", "model.layers.15.mlp.experts.127.gate_proj", "model.layers.15.mlp.experts.128.gate_proj", "model.layers.15.mlp.experts.129.gate_proj", "model.layers.15.mlp.experts.130.gate_proj", "model.layers.15.mlp.experts.131.gate_proj", "model.layers.15.mlp.experts.132.gate_proj", "model.layers.15.mlp.experts.133.gate_proj", "model.layers.15.mlp.experts.134.gate_proj", "model.layers.15.mlp.experts.135.gate_proj", "model.layers.15.mlp.experts.136.gate_proj", "model.layers.15.mlp.experts.137.gate_proj", "model.layers.15.mlp.experts.138.gate_proj", "model.layers.15.mlp.experts.139.gate_proj", "model.layers.15.mlp.experts.140.gate_proj", "model.layers.15.mlp.experts.141.gate_proj", "model.layers.15.mlp.experts.142.gate_proj", "model.layers.15.mlp.experts.143.gate_proj", "model.layers.15.mlp.experts.144.gate_proj", "model.layers.15.mlp.experts.145.gate_proj", "model.layers.15.mlp.experts.146.gate_proj", "model.layers.15.mlp.experts.147.gate_proj", "model.layers.15.mlp.experts.148.gate_proj", "model.layers.15.mlp.experts.149.gate_proj", "model.layers.15.mlp.experts.150.gate_proj", "model.layers.15.mlp.experts.151.gate_proj", "model.layers.15.mlp.experts.152.gate_proj", "model.layers.15.mlp.experts.153.gate_proj", "model.layers.15.mlp.experts.154.gate_proj", "model.layers.15.mlp.experts.155.gate_proj", "model.layers.15.mlp.experts.156.gate_proj", "model.layers.15.mlp.experts.157.gate_proj", "model.layers.15.mlp.experts.158.gate_proj", "model.layers.15.mlp.experts.159.gate_proj", "model.layers.15.mlp.experts.0.up_proj", "model.layers.15.mlp.experts.1.up_proj", "model.layers.15.mlp.experts.2.up_proj", "model.layers.15.mlp.experts.3.up_proj", "model.layers.15.mlp.experts.4.up_proj", "model.layers.15.mlp.experts.5.up_proj", "model.layers.15.mlp.experts.6.up_proj", "model.layers.15.mlp.experts.7.up_proj", "model.layers.15.mlp.experts.8.up_proj", "model.layers.15.mlp.experts.9.up_proj", "model.layers.15.mlp.experts.10.up_proj", "model.layers.15.mlp.experts.11.up_proj", "model.layers.15.mlp.experts.12.up_proj", "model.layers.15.mlp.experts.13.up_proj", "model.layers.15.mlp.experts.14.up_proj", "model.layers.15.mlp.experts.15.up_proj", "model.layers.15.mlp.experts.16.up_proj", "model.layers.15.mlp.experts.17.up_proj", "model.layers.15.mlp.experts.18.up_proj", "model.layers.15.mlp.experts.19.up_proj", "model.layers.15.mlp.experts.20.up_proj", "model.layers.15.mlp.experts.21.up_proj", "model.layers.15.mlp.experts.22.up_proj", "model.layers.15.mlp.experts.23.up_proj", "model.layers.15.mlp.experts.24.up_proj", "model.layers.15.mlp.experts.25.up_proj", "model.layers.15.mlp.experts.26.up_proj", "model.layers.15.mlp.experts.27.up_proj", "model.layers.15.mlp.experts.28.up_proj", "model.layers.15.mlp.experts.29.up_proj", "model.layers.15.mlp.experts.30.up_proj", "model.layers.15.mlp.experts.31.up_proj", "model.layers.15.mlp.experts.32.up_proj", "model.layers.15.mlp.experts.33.up_proj", "model.layers.15.mlp.experts.34.up_proj", "model.layers.15.mlp.experts.35.up_proj", "model.layers.15.mlp.experts.36.up_proj", "model.layers.15.mlp.experts.37.up_proj", "model.layers.15.mlp.experts.38.up_proj", "model.layers.15.mlp.experts.39.up_proj", "model.layers.15.mlp.experts.40.up_proj", "model.layers.15.mlp.experts.41.up_proj", "model.layers.15.mlp.experts.42.up_proj", "model.layers.15.mlp.experts.43.up_proj", "model.layers.15.mlp.experts.44.up_proj", "model.layers.15.mlp.experts.45.up_proj", "model.layers.15.mlp.experts.46.up_proj", "model.layers.15.mlp.experts.47.up_proj", "model.layers.15.mlp.experts.48.up_proj", "model.layers.15.mlp.experts.49.up_proj", "model.layers.15.mlp.experts.50.up_proj", "model.layers.15.mlp.experts.51.up_proj", "model.layers.15.mlp.experts.52.up_proj", "model.layers.15.mlp.experts.53.up_proj", "model.layers.15.mlp.experts.54.up_proj", "model.layers.15.mlp.experts.55.up_proj", "model.layers.15.mlp.experts.56.up_proj", "model.layers.15.mlp.experts.57.up_proj", "model.layers.15.mlp.experts.58.up_proj", "model.layers.15.mlp.experts.59.up_proj", "model.layers.15.mlp.experts.60.up_proj", "model.layers.15.mlp.experts.61.up_proj", "model.layers.15.mlp.experts.62.up_proj", "model.layers.15.mlp.experts.63.up_proj", "model.layers.15.mlp.experts.64.up_proj", "model.layers.15.mlp.experts.65.up_proj", "model.layers.15.mlp.experts.66.up_proj", "model.layers.15.mlp.experts.67.up_proj", "model.layers.15.mlp.experts.68.up_proj", "model.layers.15.mlp.experts.69.up_proj", "model.layers.15.mlp.experts.70.up_proj", "model.layers.15.mlp.experts.71.up_proj", "model.layers.15.mlp.experts.72.up_proj", "model.layers.15.mlp.experts.73.up_proj", "model.layers.15.mlp.experts.74.up_proj", "model.layers.15.mlp.experts.75.up_proj", "model.layers.15.mlp.experts.76.up_proj", "model.layers.15.mlp.experts.77.up_proj", "model.layers.15.mlp.experts.78.up_proj", "model.layers.15.mlp.experts.79.up_proj", "model.layers.15.mlp.experts.80.up_proj", "model.layers.15.mlp.experts.81.up_proj", "model.layers.15.mlp.experts.82.up_proj", "model.layers.15.mlp.experts.83.up_proj", "model.layers.15.mlp.experts.84.up_proj", "model.layers.15.mlp.experts.85.up_proj", "model.layers.15.mlp.experts.86.up_proj", "model.layers.15.mlp.experts.87.up_proj", "model.layers.15.mlp.experts.88.up_proj", "model.layers.15.mlp.experts.89.up_proj", "model.layers.15.mlp.experts.90.up_proj", "model.layers.15.mlp.experts.91.up_proj", "model.layers.15.mlp.experts.92.up_proj", "model.layers.15.mlp.experts.93.up_proj", "model.layers.15.mlp.experts.94.up_proj", "model.layers.15.mlp.experts.95.up_proj", "model.layers.15.mlp.experts.96.up_proj", "model.layers.15.mlp.experts.97.up_proj", "model.layers.15.mlp.experts.98.up_proj", "model.layers.15.mlp.experts.99.up_proj", "model.layers.15.mlp.experts.100.up_proj", "model.layers.15.mlp.experts.101.up_proj", "model.layers.15.mlp.experts.102.up_proj", "model.layers.15.mlp.experts.103.up_proj", "model.layers.15.mlp.experts.104.up_proj", "model.layers.15.mlp.experts.105.up_proj", "model.layers.15.mlp.experts.106.up_proj", "model.layers.15.mlp.experts.107.up_proj", "model.layers.15.mlp.experts.108.up_proj", "model.layers.15.mlp.experts.109.up_proj", "model.layers.15.mlp.experts.110.up_proj", "model.layers.15.mlp.experts.111.up_proj", "model.layers.15.mlp.experts.112.up_proj", "model.layers.15.mlp.experts.113.up_proj", "model.layers.15.mlp.experts.114.up_proj", "model.layers.15.mlp.experts.115.up_proj", "model.layers.15.mlp.experts.116.up_proj", "model.layers.15.mlp.experts.117.up_proj", "model.layers.15.mlp.experts.118.up_proj", "model.layers.15.mlp.experts.119.up_proj", "model.layers.15.mlp.experts.120.up_proj", "model.layers.15.mlp.experts.121.up_proj", "model.layers.15.mlp.experts.122.up_proj", "model.layers.15.mlp.experts.123.up_proj", "model.layers.15.mlp.experts.124.up_proj", "model.layers.15.mlp.experts.125.up_proj", "model.layers.15.mlp.experts.126.up_proj", "model.layers.15.mlp.experts.127.up_proj", "model.layers.15.mlp.experts.128.up_proj", "model.layers.15.mlp.experts.129.up_proj", "model.layers.15.mlp.experts.130.up_proj", "model.layers.15.mlp.experts.131.up_proj", "model.layers.15.mlp.experts.132.up_proj", "model.layers.15.mlp.experts.133.up_proj", "model.layers.15.mlp.experts.134.up_proj", "model.layers.15.mlp.experts.135.up_proj", "model.layers.15.mlp.experts.136.up_proj", "model.layers.15.mlp.experts.137.up_proj", "model.layers.15.mlp.experts.138.up_proj", "model.layers.15.mlp.experts.139.up_proj", "model.layers.15.mlp.experts.140.up_proj", "model.layers.15.mlp.experts.141.up_proj", "model.layers.15.mlp.experts.142.up_proj", "model.layers.15.mlp.experts.143.up_proj", "model.layers.15.mlp.experts.144.up_proj", "model.layers.15.mlp.experts.145.up_proj", "model.layers.15.mlp.experts.146.up_proj", "model.layers.15.mlp.experts.147.up_proj", "model.layers.15.mlp.experts.148.up_proj", "model.layers.15.mlp.experts.149.up_proj", "model.layers.15.mlp.experts.150.up_proj", "model.layers.15.mlp.experts.151.up_proj", "model.layers.15.mlp.experts.152.up_proj", "model.layers.15.mlp.experts.153.up_proj", "model.layers.15.mlp.experts.154.up_proj", "model.layers.15.mlp.experts.155.up_proj", "model.layers.15.mlp.experts.156.up_proj", "model.layers.15.mlp.experts.157.up_proj", "model.layers.15.mlp.experts.158.up_proj", "model.layers.15.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.001028525363653901, "dbits": 2516582400 }, { "dkld": -0.0009217391721904333, "dbits": 5033164800 }, { "dkld": -0.001096879784017804, "dbits": 7549747200 }, { "dkld": -0.001118579413741827, "dbits": 12582912000 } ] }, { "idx": 92, "layers": [ "model.layers.15.mlp.experts.0.down_proj", "model.layers.15.mlp.experts.1.down_proj", "model.layers.15.mlp.experts.2.down_proj", "model.layers.15.mlp.experts.3.down_proj", "model.layers.15.mlp.experts.4.down_proj", "model.layers.15.mlp.experts.5.down_proj", "model.layers.15.mlp.experts.6.down_proj", "model.layers.15.mlp.experts.7.down_proj", "model.layers.15.mlp.experts.8.down_proj", "model.layers.15.mlp.experts.9.down_proj", "model.layers.15.mlp.experts.10.down_proj", "model.layers.15.mlp.experts.11.down_proj", "model.layers.15.mlp.experts.12.down_proj", "model.layers.15.mlp.experts.13.down_proj", "model.layers.15.mlp.experts.14.down_proj", "model.layers.15.mlp.experts.15.down_proj", "model.layers.15.mlp.experts.16.down_proj", "model.layers.15.mlp.experts.17.down_proj", "model.layers.15.mlp.experts.18.down_proj", "model.layers.15.mlp.experts.19.down_proj", "model.layers.15.mlp.experts.20.down_proj", "model.layers.15.mlp.experts.21.down_proj", "model.layers.15.mlp.experts.22.down_proj", "model.layers.15.mlp.experts.23.down_proj", "model.layers.15.mlp.experts.24.down_proj", "model.layers.15.mlp.experts.25.down_proj", "model.layers.15.mlp.experts.26.down_proj", "model.layers.15.mlp.experts.27.down_proj", "model.layers.15.mlp.experts.28.down_proj", "model.layers.15.mlp.experts.29.down_proj", "model.layers.15.mlp.experts.30.down_proj", "model.layers.15.mlp.experts.31.down_proj", "model.layers.15.mlp.experts.32.down_proj", "model.layers.15.mlp.experts.33.down_proj", "model.layers.15.mlp.experts.34.down_proj", "model.layers.15.mlp.experts.35.down_proj", "model.layers.15.mlp.experts.36.down_proj", "model.layers.15.mlp.experts.37.down_proj", "model.layers.15.mlp.experts.38.down_proj", "model.layers.15.mlp.experts.39.down_proj", "model.layers.15.mlp.experts.40.down_proj", "model.layers.15.mlp.experts.41.down_proj", "model.layers.15.mlp.experts.42.down_proj", "model.layers.15.mlp.experts.43.down_proj", "model.layers.15.mlp.experts.44.down_proj", "model.layers.15.mlp.experts.45.down_proj", "model.layers.15.mlp.experts.46.down_proj", "model.layers.15.mlp.experts.47.down_proj", "model.layers.15.mlp.experts.48.down_proj", "model.layers.15.mlp.experts.49.down_proj", "model.layers.15.mlp.experts.50.down_proj", "model.layers.15.mlp.experts.51.down_proj", "model.layers.15.mlp.experts.52.down_proj", "model.layers.15.mlp.experts.53.down_proj", "model.layers.15.mlp.experts.54.down_proj", "model.layers.15.mlp.experts.55.down_proj", "model.layers.15.mlp.experts.56.down_proj", "model.layers.15.mlp.experts.57.down_proj", "model.layers.15.mlp.experts.58.down_proj", "model.layers.15.mlp.experts.59.down_proj", "model.layers.15.mlp.experts.60.down_proj", "model.layers.15.mlp.experts.61.down_proj", "model.layers.15.mlp.experts.62.down_proj", "model.layers.15.mlp.experts.63.down_proj", "model.layers.15.mlp.experts.64.down_proj", "model.layers.15.mlp.experts.65.down_proj", "model.layers.15.mlp.experts.66.down_proj", "model.layers.15.mlp.experts.67.down_proj", "model.layers.15.mlp.experts.68.down_proj", "model.layers.15.mlp.experts.69.down_proj", "model.layers.15.mlp.experts.70.down_proj", "model.layers.15.mlp.experts.71.down_proj", "model.layers.15.mlp.experts.72.down_proj", "model.layers.15.mlp.experts.73.down_proj", "model.layers.15.mlp.experts.74.down_proj", "model.layers.15.mlp.experts.75.down_proj", "model.layers.15.mlp.experts.76.down_proj", "model.layers.15.mlp.experts.77.down_proj", "model.layers.15.mlp.experts.78.down_proj", "model.layers.15.mlp.experts.79.down_proj", "model.layers.15.mlp.experts.80.down_proj", "model.layers.15.mlp.experts.81.down_proj", "model.layers.15.mlp.experts.82.down_proj", "model.layers.15.mlp.experts.83.down_proj", "model.layers.15.mlp.experts.84.down_proj", "model.layers.15.mlp.experts.85.down_proj", "model.layers.15.mlp.experts.86.down_proj", "model.layers.15.mlp.experts.87.down_proj", "model.layers.15.mlp.experts.88.down_proj", "model.layers.15.mlp.experts.89.down_proj", "model.layers.15.mlp.experts.90.down_proj", "model.layers.15.mlp.experts.91.down_proj", "model.layers.15.mlp.experts.92.down_proj", "model.layers.15.mlp.experts.93.down_proj", "model.layers.15.mlp.experts.94.down_proj", "model.layers.15.mlp.experts.95.down_proj", "model.layers.15.mlp.experts.96.down_proj", "model.layers.15.mlp.experts.97.down_proj", "model.layers.15.mlp.experts.98.down_proj", "model.layers.15.mlp.experts.99.down_proj", "model.layers.15.mlp.experts.100.down_proj", "model.layers.15.mlp.experts.101.down_proj", "model.layers.15.mlp.experts.102.down_proj", "model.layers.15.mlp.experts.103.down_proj", "model.layers.15.mlp.experts.104.down_proj", "model.layers.15.mlp.experts.105.down_proj", "model.layers.15.mlp.experts.106.down_proj", "model.layers.15.mlp.experts.107.down_proj", "model.layers.15.mlp.experts.108.down_proj", "model.layers.15.mlp.experts.109.down_proj", "model.layers.15.mlp.experts.110.down_proj", "model.layers.15.mlp.experts.111.down_proj", "model.layers.15.mlp.experts.112.down_proj", "model.layers.15.mlp.experts.113.down_proj", "model.layers.15.mlp.experts.114.down_proj", "model.layers.15.mlp.experts.115.down_proj", "model.layers.15.mlp.experts.116.down_proj", "model.layers.15.mlp.experts.117.down_proj", "model.layers.15.mlp.experts.118.down_proj", "model.layers.15.mlp.experts.119.down_proj", "model.layers.15.mlp.experts.120.down_proj", "model.layers.15.mlp.experts.121.down_proj", "model.layers.15.mlp.experts.122.down_proj", "model.layers.15.mlp.experts.123.down_proj", "model.layers.15.mlp.experts.124.down_proj", "model.layers.15.mlp.experts.125.down_proj", "model.layers.15.mlp.experts.126.down_proj", "model.layers.15.mlp.experts.127.down_proj", "model.layers.15.mlp.experts.128.down_proj", "model.layers.15.mlp.experts.129.down_proj", "model.layers.15.mlp.experts.130.down_proj", "model.layers.15.mlp.experts.131.down_proj", "model.layers.15.mlp.experts.132.down_proj", "model.layers.15.mlp.experts.133.down_proj", "model.layers.15.mlp.experts.134.down_proj", "model.layers.15.mlp.experts.135.down_proj", "model.layers.15.mlp.experts.136.down_proj", "model.layers.15.mlp.experts.137.down_proj", "model.layers.15.mlp.experts.138.down_proj", "model.layers.15.mlp.experts.139.down_proj", "model.layers.15.mlp.experts.140.down_proj", "model.layers.15.mlp.experts.141.down_proj", "model.layers.15.mlp.experts.142.down_proj", "model.layers.15.mlp.experts.143.down_proj", "model.layers.15.mlp.experts.144.down_proj", "model.layers.15.mlp.experts.145.down_proj", "model.layers.15.mlp.experts.146.down_proj", "model.layers.15.mlp.experts.147.down_proj", "model.layers.15.mlp.experts.148.down_proj", "model.layers.15.mlp.experts.149.down_proj", "model.layers.15.mlp.experts.150.down_proj", "model.layers.15.mlp.experts.151.down_proj", "model.layers.15.mlp.experts.152.down_proj", "model.layers.15.mlp.experts.153.down_proj", "model.layers.15.mlp.experts.154.down_proj", "model.layers.15.mlp.experts.155.down_proj", "model.layers.15.mlp.experts.156.down_proj", "model.layers.15.mlp.experts.157.down_proj", "model.layers.15.mlp.experts.158.down_proj", "model.layers.15.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0006942315958440304, "dbits": 1258291200 }, { "dkld": -0.0006604485213756589, "dbits": 2516582400 }, { "dkld": -0.0007232288829982364, "dbits": 3774873600 }, { "dkld": -0.0007396108470857227, "dbits": 6291456000 } ] }, { "idx": 93, "layers": [ "model.layers.16.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00042295334860682765, "dbits": 62914560 }, { "dkld": -1.540072262287695e-05, "dbits": 125829120 }, { "dkld": -0.00014321189373732446, "dbits": 188743680 }, { "dkld": -7.18832015991211e-05, "dbits": 314572800 } ] }, { "idx": 94, "layers": [ "model.layers.16.self_attn.k_proj", "model.layers.16.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0004703273996710666, "dbits": 10485760 }, { "dkld": 0.00031845234334468286, "dbits": 20971520 }, { "dkld": 0.0003970423713326454, "dbits": 31457280 }, { "dkld": 0.00027460679411887845, "dbits": 52428800 } ] }, { "idx": 95, "layers": [ "model.layers.16.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00011351136490703184, "dbits": 62914560 }, { "dkld": -0.00020776130259037018, "dbits": 125829120 }, { "dkld": -0.0001731723546981867, "dbits": 188743680 }, { "dkld": -0.00012824069708586294, "dbits": 314572800 } ] }, { "idx": 96, "layers": [ "model.layers.16.mlp.shared_experts.gate_proj", "model.layers.16.mlp.shared_experts.up_proj", "model.layers.16.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -3.616735339165289e-05, "dbits": 23592960 }, { "dkld": -0.0001504405401647202, "dbits": 47185920 }, { "dkld": -6.596185266971588e-05, "dbits": 70778880 }, { "dkld": 7.94221647083676e-05, "dbits": 117964800 } ] }, { "idx": 97, "layers": [ "model.layers.16.mlp.experts.0.gate_proj", "model.layers.16.mlp.experts.1.gate_proj", "model.layers.16.mlp.experts.2.gate_proj", "model.layers.16.mlp.experts.3.gate_proj", "model.layers.16.mlp.experts.4.gate_proj", "model.layers.16.mlp.experts.5.gate_proj", "model.layers.16.mlp.experts.6.gate_proj", "model.layers.16.mlp.experts.7.gate_proj", "model.layers.16.mlp.experts.8.gate_proj", "model.layers.16.mlp.experts.9.gate_proj", "model.layers.16.mlp.experts.10.gate_proj", "model.layers.16.mlp.experts.11.gate_proj", "model.layers.16.mlp.experts.12.gate_proj", "model.layers.16.mlp.experts.13.gate_proj", "model.layers.16.mlp.experts.14.gate_proj", "model.layers.16.mlp.experts.15.gate_proj", "model.layers.16.mlp.experts.16.gate_proj", "model.layers.16.mlp.experts.17.gate_proj", "model.layers.16.mlp.experts.18.gate_proj", "model.layers.16.mlp.experts.19.gate_proj", "model.layers.16.mlp.experts.20.gate_proj", "model.layers.16.mlp.experts.21.gate_proj", "model.layers.16.mlp.experts.22.gate_proj", "model.layers.16.mlp.experts.23.gate_proj", "model.layers.16.mlp.experts.24.gate_proj", "model.layers.16.mlp.experts.25.gate_proj", "model.layers.16.mlp.experts.26.gate_proj", "model.layers.16.mlp.experts.27.gate_proj", "model.layers.16.mlp.experts.28.gate_proj", "model.layers.16.mlp.experts.29.gate_proj", "model.layers.16.mlp.experts.30.gate_proj", "model.layers.16.mlp.experts.31.gate_proj", "model.layers.16.mlp.experts.32.gate_proj", "model.layers.16.mlp.experts.33.gate_proj", "model.layers.16.mlp.experts.34.gate_proj", "model.layers.16.mlp.experts.35.gate_proj", "model.layers.16.mlp.experts.36.gate_proj", "model.layers.16.mlp.experts.37.gate_proj", "model.layers.16.mlp.experts.38.gate_proj", "model.layers.16.mlp.experts.39.gate_proj", "model.layers.16.mlp.experts.40.gate_proj", "model.layers.16.mlp.experts.41.gate_proj", "model.layers.16.mlp.experts.42.gate_proj", "model.layers.16.mlp.experts.43.gate_proj", "model.layers.16.mlp.experts.44.gate_proj", "model.layers.16.mlp.experts.45.gate_proj", "model.layers.16.mlp.experts.46.gate_proj", "model.layers.16.mlp.experts.47.gate_proj", "model.layers.16.mlp.experts.48.gate_proj", "model.layers.16.mlp.experts.49.gate_proj", "model.layers.16.mlp.experts.50.gate_proj", "model.layers.16.mlp.experts.51.gate_proj", "model.layers.16.mlp.experts.52.gate_proj", "model.layers.16.mlp.experts.53.gate_proj", "model.layers.16.mlp.experts.54.gate_proj", "model.layers.16.mlp.experts.55.gate_proj", "model.layers.16.mlp.experts.56.gate_proj", "model.layers.16.mlp.experts.57.gate_proj", "model.layers.16.mlp.experts.58.gate_proj", "model.layers.16.mlp.experts.59.gate_proj", "model.layers.16.mlp.experts.60.gate_proj", "model.layers.16.mlp.experts.61.gate_proj", "model.layers.16.mlp.experts.62.gate_proj", "model.layers.16.mlp.experts.63.gate_proj", "model.layers.16.mlp.experts.64.gate_proj", "model.layers.16.mlp.experts.65.gate_proj", "model.layers.16.mlp.experts.66.gate_proj", "model.layers.16.mlp.experts.67.gate_proj", "model.layers.16.mlp.experts.68.gate_proj", "model.layers.16.mlp.experts.69.gate_proj", "model.layers.16.mlp.experts.70.gate_proj", "model.layers.16.mlp.experts.71.gate_proj", "model.layers.16.mlp.experts.72.gate_proj", "model.layers.16.mlp.experts.73.gate_proj", "model.layers.16.mlp.experts.74.gate_proj", "model.layers.16.mlp.experts.75.gate_proj", "model.layers.16.mlp.experts.76.gate_proj", "model.layers.16.mlp.experts.77.gate_proj", "model.layers.16.mlp.experts.78.gate_proj", "model.layers.16.mlp.experts.79.gate_proj", "model.layers.16.mlp.experts.80.gate_proj", "model.layers.16.mlp.experts.81.gate_proj", "model.layers.16.mlp.experts.82.gate_proj", "model.layers.16.mlp.experts.83.gate_proj", "model.layers.16.mlp.experts.84.gate_proj", "model.layers.16.mlp.experts.85.gate_proj", "model.layers.16.mlp.experts.86.gate_proj", "model.layers.16.mlp.experts.87.gate_proj", "model.layers.16.mlp.experts.88.gate_proj", "model.layers.16.mlp.experts.89.gate_proj", "model.layers.16.mlp.experts.90.gate_proj", "model.layers.16.mlp.experts.91.gate_proj", "model.layers.16.mlp.experts.92.gate_proj", "model.layers.16.mlp.experts.93.gate_proj", "model.layers.16.mlp.experts.94.gate_proj", "model.layers.16.mlp.experts.95.gate_proj", "model.layers.16.mlp.experts.96.gate_proj", "model.layers.16.mlp.experts.97.gate_proj", "model.layers.16.mlp.experts.98.gate_proj", "model.layers.16.mlp.experts.99.gate_proj", "model.layers.16.mlp.experts.100.gate_proj", "model.layers.16.mlp.experts.101.gate_proj", "model.layers.16.mlp.experts.102.gate_proj", "model.layers.16.mlp.experts.103.gate_proj", "model.layers.16.mlp.experts.104.gate_proj", "model.layers.16.mlp.experts.105.gate_proj", "model.layers.16.mlp.experts.106.gate_proj", "model.layers.16.mlp.experts.107.gate_proj", "model.layers.16.mlp.experts.108.gate_proj", "model.layers.16.mlp.experts.109.gate_proj", "model.layers.16.mlp.experts.110.gate_proj", "model.layers.16.mlp.experts.111.gate_proj", "model.layers.16.mlp.experts.112.gate_proj", "model.layers.16.mlp.experts.113.gate_proj", "model.layers.16.mlp.experts.114.gate_proj", "model.layers.16.mlp.experts.115.gate_proj", "model.layers.16.mlp.experts.116.gate_proj", "model.layers.16.mlp.experts.117.gate_proj", "model.layers.16.mlp.experts.118.gate_proj", "model.layers.16.mlp.experts.119.gate_proj", "model.layers.16.mlp.experts.120.gate_proj", "model.layers.16.mlp.experts.121.gate_proj", "model.layers.16.mlp.experts.122.gate_proj", "model.layers.16.mlp.experts.123.gate_proj", "model.layers.16.mlp.experts.124.gate_proj", "model.layers.16.mlp.experts.125.gate_proj", "model.layers.16.mlp.experts.126.gate_proj", "model.layers.16.mlp.experts.127.gate_proj", "model.layers.16.mlp.experts.128.gate_proj", "model.layers.16.mlp.experts.129.gate_proj", "model.layers.16.mlp.experts.130.gate_proj", "model.layers.16.mlp.experts.131.gate_proj", "model.layers.16.mlp.experts.132.gate_proj", "model.layers.16.mlp.experts.133.gate_proj", "model.layers.16.mlp.experts.134.gate_proj", "model.layers.16.mlp.experts.135.gate_proj", "model.layers.16.mlp.experts.136.gate_proj", "model.layers.16.mlp.experts.137.gate_proj", "model.layers.16.mlp.experts.138.gate_proj", "model.layers.16.mlp.experts.139.gate_proj", "model.layers.16.mlp.experts.140.gate_proj", "model.layers.16.mlp.experts.141.gate_proj", "model.layers.16.mlp.experts.142.gate_proj", "model.layers.16.mlp.experts.143.gate_proj", "model.layers.16.mlp.experts.144.gate_proj", "model.layers.16.mlp.experts.145.gate_proj", "model.layers.16.mlp.experts.146.gate_proj", "model.layers.16.mlp.experts.147.gate_proj", "model.layers.16.mlp.experts.148.gate_proj", "model.layers.16.mlp.experts.149.gate_proj", "model.layers.16.mlp.experts.150.gate_proj", "model.layers.16.mlp.experts.151.gate_proj", "model.layers.16.mlp.experts.152.gate_proj", "model.layers.16.mlp.experts.153.gate_proj", "model.layers.16.mlp.experts.154.gate_proj", "model.layers.16.mlp.experts.155.gate_proj", "model.layers.16.mlp.experts.156.gate_proj", "model.layers.16.mlp.experts.157.gate_proj", "model.layers.16.mlp.experts.158.gate_proj", "model.layers.16.mlp.experts.159.gate_proj", "model.layers.16.mlp.experts.0.up_proj", "model.layers.16.mlp.experts.1.up_proj", "model.layers.16.mlp.experts.2.up_proj", "model.layers.16.mlp.experts.3.up_proj", "model.layers.16.mlp.experts.4.up_proj", "model.layers.16.mlp.experts.5.up_proj", "model.layers.16.mlp.experts.6.up_proj", "model.layers.16.mlp.experts.7.up_proj", "model.layers.16.mlp.experts.8.up_proj", "model.layers.16.mlp.experts.9.up_proj", "model.layers.16.mlp.experts.10.up_proj", "model.layers.16.mlp.experts.11.up_proj", "model.layers.16.mlp.experts.12.up_proj", "model.layers.16.mlp.experts.13.up_proj", "model.layers.16.mlp.experts.14.up_proj", "model.layers.16.mlp.experts.15.up_proj", "model.layers.16.mlp.experts.16.up_proj", "model.layers.16.mlp.experts.17.up_proj", "model.layers.16.mlp.experts.18.up_proj", "model.layers.16.mlp.experts.19.up_proj", "model.layers.16.mlp.experts.20.up_proj", "model.layers.16.mlp.experts.21.up_proj", "model.layers.16.mlp.experts.22.up_proj", "model.layers.16.mlp.experts.23.up_proj", "model.layers.16.mlp.experts.24.up_proj", "model.layers.16.mlp.experts.25.up_proj", "model.layers.16.mlp.experts.26.up_proj", "model.layers.16.mlp.experts.27.up_proj", "model.layers.16.mlp.experts.28.up_proj", "model.layers.16.mlp.experts.29.up_proj", "model.layers.16.mlp.experts.30.up_proj", "model.layers.16.mlp.experts.31.up_proj", "model.layers.16.mlp.experts.32.up_proj", "model.layers.16.mlp.experts.33.up_proj", "model.layers.16.mlp.experts.34.up_proj", "model.layers.16.mlp.experts.35.up_proj", "model.layers.16.mlp.experts.36.up_proj", "model.layers.16.mlp.experts.37.up_proj", "model.layers.16.mlp.experts.38.up_proj", "model.layers.16.mlp.experts.39.up_proj", "model.layers.16.mlp.experts.40.up_proj", "model.layers.16.mlp.experts.41.up_proj", "model.layers.16.mlp.experts.42.up_proj", "model.layers.16.mlp.experts.43.up_proj", "model.layers.16.mlp.experts.44.up_proj", "model.layers.16.mlp.experts.45.up_proj", "model.layers.16.mlp.experts.46.up_proj", "model.layers.16.mlp.experts.47.up_proj", "model.layers.16.mlp.experts.48.up_proj", "model.layers.16.mlp.experts.49.up_proj", "model.layers.16.mlp.experts.50.up_proj", "model.layers.16.mlp.experts.51.up_proj", "model.layers.16.mlp.experts.52.up_proj", "model.layers.16.mlp.experts.53.up_proj", "model.layers.16.mlp.experts.54.up_proj", "model.layers.16.mlp.experts.55.up_proj", "model.layers.16.mlp.experts.56.up_proj", "model.layers.16.mlp.experts.57.up_proj", "model.layers.16.mlp.experts.58.up_proj", "model.layers.16.mlp.experts.59.up_proj", "model.layers.16.mlp.experts.60.up_proj", "model.layers.16.mlp.experts.61.up_proj", "model.layers.16.mlp.experts.62.up_proj", "model.layers.16.mlp.experts.63.up_proj", "model.layers.16.mlp.experts.64.up_proj", "model.layers.16.mlp.experts.65.up_proj", "model.layers.16.mlp.experts.66.up_proj", "model.layers.16.mlp.experts.67.up_proj", "model.layers.16.mlp.experts.68.up_proj", "model.layers.16.mlp.experts.69.up_proj", "model.layers.16.mlp.experts.70.up_proj", "model.layers.16.mlp.experts.71.up_proj", "model.layers.16.mlp.experts.72.up_proj", "model.layers.16.mlp.experts.73.up_proj", "model.layers.16.mlp.experts.74.up_proj", "model.layers.16.mlp.experts.75.up_proj", "model.layers.16.mlp.experts.76.up_proj", "model.layers.16.mlp.experts.77.up_proj", "model.layers.16.mlp.experts.78.up_proj", "model.layers.16.mlp.experts.79.up_proj", "model.layers.16.mlp.experts.80.up_proj", "model.layers.16.mlp.experts.81.up_proj", "model.layers.16.mlp.experts.82.up_proj", "model.layers.16.mlp.experts.83.up_proj", "model.layers.16.mlp.experts.84.up_proj", "model.layers.16.mlp.experts.85.up_proj", "model.layers.16.mlp.experts.86.up_proj", "model.layers.16.mlp.experts.87.up_proj", "model.layers.16.mlp.experts.88.up_proj", "model.layers.16.mlp.experts.89.up_proj", "model.layers.16.mlp.experts.90.up_proj", "model.layers.16.mlp.experts.91.up_proj", "model.layers.16.mlp.experts.92.up_proj", "model.layers.16.mlp.experts.93.up_proj", "model.layers.16.mlp.experts.94.up_proj", "model.layers.16.mlp.experts.95.up_proj", "model.layers.16.mlp.experts.96.up_proj", "model.layers.16.mlp.experts.97.up_proj", "model.layers.16.mlp.experts.98.up_proj", "model.layers.16.mlp.experts.99.up_proj", "model.layers.16.mlp.experts.100.up_proj", "model.layers.16.mlp.experts.101.up_proj", "model.layers.16.mlp.experts.102.up_proj", "model.layers.16.mlp.experts.103.up_proj", "model.layers.16.mlp.experts.104.up_proj", "model.layers.16.mlp.experts.105.up_proj", "model.layers.16.mlp.experts.106.up_proj", "model.layers.16.mlp.experts.107.up_proj", "model.layers.16.mlp.experts.108.up_proj", "model.layers.16.mlp.experts.109.up_proj", "model.layers.16.mlp.experts.110.up_proj", "model.layers.16.mlp.experts.111.up_proj", "model.layers.16.mlp.experts.112.up_proj", "model.layers.16.mlp.experts.113.up_proj", "model.layers.16.mlp.experts.114.up_proj", "model.layers.16.mlp.experts.115.up_proj", "model.layers.16.mlp.experts.116.up_proj", "model.layers.16.mlp.experts.117.up_proj", "model.layers.16.mlp.experts.118.up_proj", "model.layers.16.mlp.experts.119.up_proj", "model.layers.16.mlp.experts.120.up_proj", "model.layers.16.mlp.experts.121.up_proj", "model.layers.16.mlp.experts.122.up_proj", "model.layers.16.mlp.experts.123.up_proj", "model.layers.16.mlp.experts.124.up_proj", "model.layers.16.mlp.experts.125.up_proj", "model.layers.16.mlp.experts.126.up_proj", "model.layers.16.mlp.experts.127.up_proj", "model.layers.16.mlp.experts.128.up_proj", "model.layers.16.mlp.experts.129.up_proj", "model.layers.16.mlp.experts.130.up_proj", "model.layers.16.mlp.experts.131.up_proj", "model.layers.16.mlp.experts.132.up_proj", "model.layers.16.mlp.experts.133.up_proj", "model.layers.16.mlp.experts.134.up_proj", "model.layers.16.mlp.experts.135.up_proj", "model.layers.16.mlp.experts.136.up_proj", "model.layers.16.mlp.experts.137.up_proj", "model.layers.16.mlp.experts.138.up_proj", "model.layers.16.mlp.experts.139.up_proj", "model.layers.16.mlp.experts.140.up_proj", "model.layers.16.mlp.experts.141.up_proj", "model.layers.16.mlp.experts.142.up_proj", "model.layers.16.mlp.experts.143.up_proj", "model.layers.16.mlp.experts.144.up_proj", "model.layers.16.mlp.experts.145.up_proj", "model.layers.16.mlp.experts.146.up_proj", "model.layers.16.mlp.experts.147.up_proj", "model.layers.16.mlp.experts.148.up_proj", "model.layers.16.mlp.experts.149.up_proj", "model.layers.16.mlp.experts.150.up_proj", "model.layers.16.mlp.experts.151.up_proj", "model.layers.16.mlp.experts.152.up_proj", "model.layers.16.mlp.experts.153.up_proj", "model.layers.16.mlp.experts.154.up_proj", "model.layers.16.mlp.experts.155.up_proj", "model.layers.16.mlp.experts.156.up_proj", "model.layers.16.mlp.experts.157.up_proj", "model.layers.16.mlp.experts.158.up_proj", "model.layers.16.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000914327986538413, "dbits": 2516582400 }, { "dkld": -0.0013693409040570342, "dbits": 5033164800 }, { "dkld": -0.0014174433425068883, "dbits": 7549747200 }, { "dkld": -0.0013825222849845886, "dbits": 12582912000 } ] }, { "idx": 98, "layers": [ "model.layers.16.mlp.experts.0.down_proj", "model.layers.16.mlp.experts.1.down_proj", "model.layers.16.mlp.experts.2.down_proj", "model.layers.16.mlp.experts.3.down_proj", "model.layers.16.mlp.experts.4.down_proj", "model.layers.16.mlp.experts.5.down_proj", "model.layers.16.mlp.experts.6.down_proj", "model.layers.16.mlp.experts.7.down_proj", "model.layers.16.mlp.experts.8.down_proj", "model.layers.16.mlp.experts.9.down_proj", "model.layers.16.mlp.experts.10.down_proj", "model.layers.16.mlp.experts.11.down_proj", "model.layers.16.mlp.experts.12.down_proj", "model.layers.16.mlp.experts.13.down_proj", "model.layers.16.mlp.experts.14.down_proj", "model.layers.16.mlp.experts.15.down_proj", "model.layers.16.mlp.experts.16.down_proj", "model.layers.16.mlp.experts.17.down_proj", "model.layers.16.mlp.experts.18.down_proj", "model.layers.16.mlp.experts.19.down_proj", "model.layers.16.mlp.experts.20.down_proj", "model.layers.16.mlp.experts.21.down_proj", "model.layers.16.mlp.experts.22.down_proj", "model.layers.16.mlp.experts.23.down_proj", "model.layers.16.mlp.experts.24.down_proj", "model.layers.16.mlp.experts.25.down_proj", "model.layers.16.mlp.experts.26.down_proj", "model.layers.16.mlp.experts.27.down_proj", "model.layers.16.mlp.experts.28.down_proj", "model.layers.16.mlp.experts.29.down_proj", "model.layers.16.mlp.experts.30.down_proj", "model.layers.16.mlp.experts.31.down_proj", "model.layers.16.mlp.experts.32.down_proj", "model.layers.16.mlp.experts.33.down_proj", "model.layers.16.mlp.experts.34.down_proj", "model.layers.16.mlp.experts.35.down_proj", "model.layers.16.mlp.experts.36.down_proj", "model.layers.16.mlp.experts.37.down_proj", "model.layers.16.mlp.experts.38.down_proj", "model.layers.16.mlp.experts.39.down_proj", "model.layers.16.mlp.experts.40.down_proj", "model.layers.16.mlp.experts.41.down_proj", "model.layers.16.mlp.experts.42.down_proj", "model.layers.16.mlp.experts.43.down_proj", "model.layers.16.mlp.experts.44.down_proj", "model.layers.16.mlp.experts.45.down_proj", "model.layers.16.mlp.experts.46.down_proj", "model.layers.16.mlp.experts.47.down_proj", "model.layers.16.mlp.experts.48.down_proj", "model.layers.16.mlp.experts.49.down_proj", "model.layers.16.mlp.experts.50.down_proj", "model.layers.16.mlp.experts.51.down_proj", "model.layers.16.mlp.experts.52.down_proj", "model.layers.16.mlp.experts.53.down_proj", "model.layers.16.mlp.experts.54.down_proj", "model.layers.16.mlp.experts.55.down_proj", "model.layers.16.mlp.experts.56.down_proj", "model.layers.16.mlp.experts.57.down_proj", "model.layers.16.mlp.experts.58.down_proj", "model.layers.16.mlp.experts.59.down_proj", "model.layers.16.mlp.experts.60.down_proj", "model.layers.16.mlp.experts.61.down_proj", "model.layers.16.mlp.experts.62.down_proj", "model.layers.16.mlp.experts.63.down_proj", "model.layers.16.mlp.experts.64.down_proj", "model.layers.16.mlp.experts.65.down_proj", "model.layers.16.mlp.experts.66.down_proj", "model.layers.16.mlp.experts.67.down_proj", "model.layers.16.mlp.experts.68.down_proj", "model.layers.16.mlp.experts.69.down_proj", "model.layers.16.mlp.experts.70.down_proj", "model.layers.16.mlp.experts.71.down_proj", "model.layers.16.mlp.experts.72.down_proj", "model.layers.16.mlp.experts.73.down_proj", "model.layers.16.mlp.experts.74.down_proj", "model.layers.16.mlp.experts.75.down_proj", "model.layers.16.mlp.experts.76.down_proj", "model.layers.16.mlp.experts.77.down_proj", "model.layers.16.mlp.experts.78.down_proj", "model.layers.16.mlp.experts.79.down_proj", "model.layers.16.mlp.experts.80.down_proj", "model.layers.16.mlp.experts.81.down_proj", "model.layers.16.mlp.experts.82.down_proj", "model.layers.16.mlp.experts.83.down_proj", "model.layers.16.mlp.experts.84.down_proj", "model.layers.16.mlp.experts.85.down_proj", "model.layers.16.mlp.experts.86.down_proj", "model.layers.16.mlp.experts.87.down_proj", "model.layers.16.mlp.experts.88.down_proj", "model.layers.16.mlp.experts.89.down_proj", "model.layers.16.mlp.experts.90.down_proj", "model.layers.16.mlp.experts.91.down_proj", "model.layers.16.mlp.experts.92.down_proj", "model.layers.16.mlp.experts.93.down_proj", "model.layers.16.mlp.experts.94.down_proj", "model.layers.16.mlp.experts.95.down_proj", "model.layers.16.mlp.experts.96.down_proj", "model.layers.16.mlp.experts.97.down_proj", "model.layers.16.mlp.experts.98.down_proj", "model.layers.16.mlp.experts.99.down_proj", "model.layers.16.mlp.experts.100.down_proj", "model.layers.16.mlp.experts.101.down_proj", "model.layers.16.mlp.experts.102.down_proj", "model.layers.16.mlp.experts.103.down_proj", "model.layers.16.mlp.experts.104.down_proj", "model.layers.16.mlp.experts.105.down_proj", "model.layers.16.mlp.experts.106.down_proj", "model.layers.16.mlp.experts.107.down_proj", "model.layers.16.mlp.experts.108.down_proj", "model.layers.16.mlp.experts.109.down_proj", "model.layers.16.mlp.experts.110.down_proj", "model.layers.16.mlp.experts.111.down_proj", "model.layers.16.mlp.experts.112.down_proj", "model.layers.16.mlp.experts.113.down_proj", "model.layers.16.mlp.experts.114.down_proj", "model.layers.16.mlp.experts.115.down_proj", "model.layers.16.mlp.experts.116.down_proj", "model.layers.16.mlp.experts.117.down_proj", "model.layers.16.mlp.experts.118.down_proj", "model.layers.16.mlp.experts.119.down_proj", "model.layers.16.mlp.experts.120.down_proj", "model.layers.16.mlp.experts.121.down_proj", "model.layers.16.mlp.experts.122.down_proj", "model.layers.16.mlp.experts.123.down_proj", "model.layers.16.mlp.experts.124.down_proj", "model.layers.16.mlp.experts.125.down_proj", "model.layers.16.mlp.experts.126.down_proj", "model.layers.16.mlp.experts.127.down_proj", "model.layers.16.mlp.experts.128.down_proj", "model.layers.16.mlp.experts.129.down_proj", "model.layers.16.mlp.experts.130.down_proj", "model.layers.16.mlp.experts.131.down_proj", "model.layers.16.mlp.experts.132.down_proj", "model.layers.16.mlp.experts.133.down_proj", "model.layers.16.mlp.experts.134.down_proj", "model.layers.16.mlp.experts.135.down_proj", "model.layers.16.mlp.experts.136.down_proj", "model.layers.16.mlp.experts.137.down_proj", "model.layers.16.mlp.experts.138.down_proj", "model.layers.16.mlp.experts.139.down_proj", "model.layers.16.mlp.experts.140.down_proj", "model.layers.16.mlp.experts.141.down_proj", "model.layers.16.mlp.experts.142.down_proj", "model.layers.16.mlp.experts.143.down_proj", "model.layers.16.mlp.experts.144.down_proj", "model.layers.16.mlp.experts.145.down_proj", "model.layers.16.mlp.experts.146.down_proj", "model.layers.16.mlp.experts.147.down_proj", "model.layers.16.mlp.experts.148.down_proj", "model.layers.16.mlp.experts.149.down_proj", "model.layers.16.mlp.experts.150.down_proj", "model.layers.16.mlp.experts.151.down_proj", "model.layers.16.mlp.experts.152.down_proj", "model.layers.16.mlp.experts.153.down_proj", "model.layers.16.mlp.experts.154.down_proj", "model.layers.16.mlp.experts.155.down_proj", "model.layers.16.mlp.experts.156.down_proj", "model.layers.16.mlp.experts.157.down_proj", "model.layers.16.mlp.experts.158.down_proj", "model.layers.16.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0006464190781116513, "dbits": 1258291200 }, { "dkld": -0.0010341108776629054, "dbits": 2516582400 }, { "dkld": -0.0008694645017385538, "dbits": 3774873600 }, { "dkld": -0.0008648464456200711, "dbits": 6291456000 } ] }, { "idx": 99, "layers": [ "model.layers.17.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0001528874970972427, "dbits": 62914560 }, { "dkld": 0.00012905225157737454, "dbits": 125829120 }, { "dkld": 0.00017324257642030438, "dbits": 188743680 }, { "dkld": 7.381122559307896e-05, "dbits": 314572800 } ] }, { "idx": 100, "layers": [ "model.layers.17.self_attn.k_proj", "model.layers.17.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007525031454861247, "dbits": 10485760 }, { "dkld": -0.0007797244004905252, "dbits": 20971520 }, { "dkld": -0.0006349615752697102, "dbits": 31457280 }, { "dkld": -0.0006232423707842882, "dbits": 52428800 } ] }, { "idx": 101, "layers": [ "model.layers.17.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003640933893621079, "dbits": 62914560 }, { "dkld": -0.0003175339661538601, "dbits": 125829120 }, { "dkld": -0.00035976953804493783, "dbits": 188743680 }, { "dkld": -0.00031684050336480696, "dbits": 314572800 } ] }, { "idx": 102, "layers": [ "model.layers.17.mlp.shared_experts.gate_proj", "model.layers.17.mlp.shared_experts.up_proj", "model.layers.17.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0004763086326420307, "dbits": 23592960 }, { "dkld": 0.0003759149461984579, "dbits": 47185920 }, { "dkld": 0.00039403485134242855, "dbits": 70778880 }, { "dkld": 0.00036550480872392377, "dbits": 117964800 } ] }, { "idx": 103, "layers": [ "model.layers.17.mlp.experts.0.gate_proj", "model.layers.17.mlp.experts.1.gate_proj", "model.layers.17.mlp.experts.2.gate_proj", "model.layers.17.mlp.experts.3.gate_proj", "model.layers.17.mlp.experts.4.gate_proj", "model.layers.17.mlp.experts.5.gate_proj", "model.layers.17.mlp.experts.6.gate_proj", "model.layers.17.mlp.experts.7.gate_proj", "model.layers.17.mlp.experts.8.gate_proj", "model.layers.17.mlp.experts.9.gate_proj", "model.layers.17.mlp.experts.10.gate_proj", "model.layers.17.mlp.experts.11.gate_proj", "model.layers.17.mlp.experts.12.gate_proj", "model.layers.17.mlp.experts.13.gate_proj", "model.layers.17.mlp.experts.14.gate_proj", "model.layers.17.mlp.experts.15.gate_proj", "model.layers.17.mlp.experts.16.gate_proj", "model.layers.17.mlp.experts.17.gate_proj", "model.layers.17.mlp.experts.18.gate_proj", "model.layers.17.mlp.experts.19.gate_proj", "model.layers.17.mlp.experts.20.gate_proj", "model.layers.17.mlp.experts.21.gate_proj", "model.layers.17.mlp.experts.22.gate_proj", "model.layers.17.mlp.experts.23.gate_proj", "model.layers.17.mlp.experts.24.gate_proj", "model.layers.17.mlp.experts.25.gate_proj", "model.layers.17.mlp.experts.26.gate_proj", "model.layers.17.mlp.experts.27.gate_proj", "model.layers.17.mlp.experts.28.gate_proj", "model.layers.17.mlp.experts.29.gate_proj", "model.layers.17.mlp.experts.30.gate_proj", "model.layers.17.mlp.experts.31.gate_proj", "model.layers.17.mlp.experts.32.gate_proj", "model.layers.17.mlp.experts.33.gate_proj", "model.layers.17.mlp.experts.34.gate_proj", "model.layers.17.mlp.experts.35.gate_proj", "model.layers.17.mlp.experts.36.gate_proj", "model.layers.17.mlp.experts.37.gate_proj", "model.layers.17.mlp.experts.38.gate_proj", "model.layers.17.mlp.experts.39.gate_proj", "model.layers.17.mlp.experts.40.gate_proj", "model.layers.17.mlp.experts.41.gate_proj", "model.layers.17.mlp.experts.42.gate_proj", "model.layers.17.mlp.experts.43.gate_proj", "model.layers.17.mlp.experts.44.gate_proj", "model.layers.17.mlp.experts.45.gate_proj", "model.layers.17.mlp.experts.46.gate_proj", "model.layers.17.mlp.experts.47.gate_proj", "model.layers.17.mlp.experts.48.gate_proj", "model.layers.17.mlp.experts.49.gate_proj", "model.layers.17.mlp.experts.50.gate_proj", "model.layers.17.mlp.experts.51.gate_proj", "model.layers.17.mlp.experts.52.gate_proj", "model.layers.17.mlp.experts.53.gate_proj", "model.layers.17.mlp.experts.54.gate_proj", "model.layers.17.mlp.experts.55.gate_proj", "model.layers.17.mlp.experts.56.gate_proj", "model.layers.17.mlp.experts.57.gate_proj", "model.layers.17.mlp.experts.58.gate_proj", "model.layers.17.mlp.experts.59.gate_proj", "model.layers.17.mlp.experts.60.gate_proj", "model.layers.17.mlp.experts.61.gate_proj", "model.layers.17.mlp.experts.62.gate_proj", "model.layers.17.mlp.experts.63.gate_proj", "model.layers.17.mlp.experts.64.gate_proj", "model.layers.17.mlp.experts.65.gate_proj", "model.layers.17.mlp.experts.66.gate_proj", "model.layers.17.mlp.experts.67.gate_proj", "model.layers.17.mlp.experts.68.gate_proj", "model.layers.17.mlp.experts.69.gate_proj", "model.layers.17.mlp.experts.70.gate_proj", "model.layers.17.mlp.experts.71.gate_proj", "model.layers.17.mlp.experts.72.gate_proj", "model.layers.17.mlp.experts.73.gate_proj", "model.layers.17.mlp.experts.74.gate_proj", "model.layers.17.mlp.experts.75.gate_proj", "model.layers.17.mlp.experts.76.gate_proj", "model.layers.17.mlp.experts.77.gate_proj", "model.layers.17.mlp.experts.78.gate_proj", "model.layers.17.mlp.experts.79.gate_proj", "model.layers.17.mlp.experts.80.gate_proj", "model.layers.17.mlp.experts.81.gate_proj", "model.layers.17.mlp.experts.82.gate_proj", "model.layers.17.mlp.experts.83.gate_proj", "model.layers.17.mlp.experts.84.gate_proj", "model.layers.17.mlp.experts.85.gate_proj", "model.layers.17.mlp.experts.86.gate_proj", "model.layers.17.mlp.experts.87.gate_proj", "model.layers.17.mlp.experts.88.gate_proj", "model.layers.17.mlp.experts.89.gate_proj", "model.layers.17.mlp.experts.90.gate_proj", "model.layers.17.mlp.experts.91.gate_proj", "model.layers.17.mlp.experts.92.gate_proj", "model.layers.17.mlp.experts.93.gate_proj", "model.layers.17.mlp.experts.94.gate_proj", "model.layers.17.mlp.experts.95.gate_proj", "model.layers.17.mlp.experts.96.gate_proj", "model.layers.17.mlp.experts.97.gate_proj", "model.layers.17.mlp.experts.98.gate_proj", "model.layers.17.mlp.experts.99.gate_proj", "model.layers.17.mlp.experts.100.gate_proj", "model.layers.17.mlp.experts.101.gate_proj", "model.layers.17.mlp.experts.102.gate_proj", "model.layers.17.mlp.experts.103.gate_proj", "model.layers.17.mlp.experts.104.gate_proj", "model.layers.17.mlp.experts.105.gate_proj", "model.layers.17.mlp.experts.106.gate_proj", "model.layers.17.mlp.experts.107.gate_proj", "model.layers.17.mlp.experts.108.gate_proj", "model.layers.17.mlp.experts.109.gate_proj", "model.layers.17.mlp.experts.110.gate_proj", "model.layers.17.mlp.experts.111.gate_proj", "model.layers.17.mlp.experts.112.gate_proj", "model.layers.17.mlp.experts.113.gate_proj", "model.layers.17.mlp.experts.114.gate_proj", "model.layers.17.mlp.experts.115.gate_proj", "model.layers.17.mlp.experts.116.gate_proj", "model.layers.17.mlp.experts.117.gate_proj", "model.layers.17.mlp.experts.118.gate_proj", "model.layers.17.mlp.experts.119.gate_proj", "model.layers.17.mlp.experts.120.gate_proj", "model.layers.17.mlp.experts.121.gate_proj", "model.layers.17.mlp.experts.122.gate_proj", "model.layers.17.mlp.experts.123.gate_proj", "model.layers.17.mlp.experts.124.gate_proj", "model.layers.17.mlp.experts.125.gate_proj", "model.layers.17.mlp.experts.126.gate_proj", "model.layers.17.mlp.experts.127.gate_proj", "model.layers.17.mlp.experts.128.gate_proj", "model.layers.17.mlp.experts.129.gate_proj", "model.layers.17.mlp.experts.130.gate_proj", "model.layers.17.mlp.experts.131.gate_proj", "model.layers.17.mlp.experts.132.gate_proj", "model.layers.17.mlp.experts.133.gate_proj", "model.layers.17.mlp.experts.134.gate_proj", "model.layers.17.mlp.experts.135.gate_proj", "model.layers.17.mlp.experts.136.gate_proj", "model.layers.17.mlp.experts.137.gate_proj", "model.layers.17.mlp.experts.138.gate_proj", "model.layers.17.mlp.experts.139.gate_proj", "model.layers.17.mlp.experts.140.gate_proj", "model.layers.17.mlp.experts.141.gate_proj", "model.layers.17.mlp.experts.142.gate_proj", "model.layers.17.mlp.experts.143.gate_proj", "model.layers.17.mlp.experts.144.gate_proj", "model.layers.17.mlp.experts.145.gate_proj", "model.layers.17.mlp.experts.146.gate_proj", "model.layers.17.mlp.experts.147.gate_proj", "model.layers.17.mlp.experts.148.gate_proj", "model.layers.17.mlp.experts.149.gate_proj", "model.layers.17.mlp.experts.150.gate_proj", "model.layers.17.mlp.experts.151.gate_proj", "model.layers.17.mlp.experts.152.gate_proj", "model.layers.17.mlp.experts.153.gate_proj", "model.layers.17.mlp.experts.154.gate_proj", "model.layers.17.mlp.experts.155.gate_proj", "model.layers.17.mlp.experts.156.gate_proj", "model.layers.17.mlp.experts.157.gate_proj", "model.layers.17.mlp.experts.158.gate_proj", "model.layers.17.mlp.experts.159.gate_proj", "model.layers.17.mlp.experts.0.up_proj", "model.layers.17.mlp.experts.1.up_proj", "model.layers.17.mlp.experts.2.up_proj", "model.layers.17.mlp.experts.3.up_proj", "model.layers.17.mlp.experts.4.up_proj", "model.layers.17.mlp.experts.5.up_proj", "model.layers.17.mlp.experts.6.up_proj", "model.layers.17.mlp.experts.7.up_proj", "model.layers.17.mlp.experts.8.up_proj", "model.layers.17.mlp.experts.9.up_proj", "model.layers.17.mlp.experts.10.up_proj", "model.layers.17.mlp.experts.11.up_proj", "model.layers.17.mlp.experts.12.up_proj", "model.layers.17.mlp.experts.13.up_proj", "model.layers.17.mlp.experts.14.up_proj", "model.layers.17.mlp.experts.15.up_proj", "model.layers.17.mlp.experts.16.up_proj", "model.layers.17.mlp.experts.17.up_proj", "model.layers.17.mlp.experts.18.up_proj", "model.layers.17.mlp.experts.19.up_proj", "model.layers.17.mlp.experts.20.up_proj", "model.layers.17.mlp.experts.21.up_proj", "model.layers.17.mlp.experts.22.up_proj", "model.layers.17.mlp.experts.23.up_proj", "model.layers.17.mlp.experts.24.up_proj", "model.layers.17.mlp.experts.25.up_proj", "model.layers.17.mlp.experts.26.up_proj", "model.layers.17.mlp.experts.27.up_proj", "model.layers.17.mlp.experts.28.up_proj", "model.layers.17.mlp.experts.29.up_proj", "model.layers.17.mlp.experts.30.up_proj", "model.layers.17.mlp.experts.31.up_proj", "model.layers.17.mlp.experts.32.up_proj", "model.layers.17.mlp.experts.33.up_proj", "model.layers.17.mlp.experts.34.up_proj", "model.layers.17.mlp.experts.35.up_proj", "model.layers.17.mlp.experts.36.up_proj", "model.layers.17.mlp.experts.37.up_proj", "model.layers.17.mlp.experts.38.up_proj", "model.layers.17.mlp.experts.39.up_proj", "model.layers.17.mlp.experts.40.up_proj", "model.layers.17.mlp.experts.41.up_proj", "model.layers.17.mlp.experts.42.up_proj", "model.layers.17.mlp.experts.43.up_proj", "model.layers.17.mlp.experts.44.up_proj", "model.layers.17.mlp.experts.45.up_proj", "model.layers.17.mlp.experts.46.up_proj", "model.layers.17.mlp.experts.47.up_proj", "model.layers.17.mlp.experts.48.up_proj", "model.layers.17.mlp.experts.49.up_proj", "model.layers.17.mlp.experts.50.up_proj", "model.layers.17.mlp.experts.51.up_proj", "model.layers.17.mlp.experts.52.up_proj", "model.layers.17.mlp.experts.53.up_proj", "model.layers.17.mlp.experts.54.up_proj", "model.layers.17.mlp.experts.55.up_proj", "model.layers.17.mlp.experts.56.up_proj", "model.layers.17.mlp.experts.57.up_proj", "model.layers.17.mlp.experts.58.up_proj", "model.layers.17.mlp.experts.59.up_proj", "model.layers.17.mlp.experts.60.up_proj", "model.layers.17.mlp.experts.61.up_proj", "model.layers.17.mlp.experts.62.up_proj", "model.layers.17.mlp.experts.63.up_proj", "model.layers.17.mlp.experts.64.up_proj", "model.layers.17.mlp.experts.65.up_proj", "model.layers.17.mlp.experts.66.up_proj", "model.layers.17.mlp.experts.67.up_proj", "model.layers.17.mlp.experts.68.up_proj", "model.layers.17.mlp.experts.69.up_proj", "model.layers.17.mlp.experts.70.up_proj", "model.layers.17.mlp.experts.71.up_proj", "model.layers.17.mlp.experts.72.up_proj", "model.layers.17.mlp.experts.73.up_proj", "model.layers.17.mlp.experts.74.up_proj", "model.layers.17.mlp.experts.75.up_proj", "model.layers.17.mlp.experts.76.up_proj", "model.layers.17.mlp.experts.77.up_proj", "model.layers.17.mlp.experts.78.up_proj", "model.layers.17.mlp.experts.79.up_proj", "model.layers.17.mlp.experts.80.up_proj", "model.layers.17.mlp.experts.81.up_proj", "model.layers.17.mlp.experts.82.up_proj", "model.layers.17.mlp.experts.83.up_proj", "model.layers.17.mlp.experts.84.up_proj", "model.layers.17.mlp.experts.85.up_proj", "model.layers.17.mlp.experts.86.up_proj", "model.layers.17.mlp.experts.87.up_proj", "model.layers.17.mlp.experts.88.up_proj", "model.layers.17.mlp.experts.89.up_proj", "model.layers.17.mlp.experts.90.up_proj", "model.layers.17.mlp.experts.91.up_proj", "model.layers.17.mlp.experts.92.up_proj", "model.layers.17.mlp.experts.93.up_proj", "model.layers.17.mlp.experts.94.up_proj", "model.layers.17.mlp.experts.95.up_proj", "model.layers.17.mlp.experts.96.up_proj", "model.layers.17.mlp.experts.97.up_proj", "model.layers.17.mlp.experts.98.up_proj", "model.layers.17.mlp.experts.99.up_proj", "model.layers.17.mlp.experts.100.up_proj", "model.layers.17.mlp.experts.101.up_proj", "model.layers.17.mlp.experts.102.up_proj", "model.layers.17.mlp.experts.103.up_proj", "model.layers.17.mlp.experts.104.up_proj", "model.layers.17.mlp.experts.105.up_proj", "model.layers.17.mlp.experts.106.up_proj", "model.layers.17.mlp.experts.107.up_proj", "model.layers.17.mlp.experts.108.up_proj", "model.layers.17.mlp.experts.109.up_proj", "model.layers.17.mlp.experts.110.up_proj", "model.layers.17.mlp.experts.111.up_proj", "model.layers.17.mlp.experts.112.up_proj", "model.layers.17.mlp.experts.113.up_proj", "model.layers.17.mlp.experts.114.up_proj", "model.layers.17.mlp.experts.115.up_proj", "model.layers.17.mlp.experts.116.up_proj", "model.layers.17.mlp.experts.117.up_proj", "model.layers.17.mlp.experts.118.up_proj", "model.layers.17.mlp.experts.119.up_proj", "model.layers.17.mlp.experts.120.up_proj", "model.layers.17.mlp.experts.121.up_proj", "model.layers.17.mlp.experts.122.up_proj", "model.layers.17.mlp.experts.123.up_proj", "model.layers.17.mlp.experts.124.up_proj", "model.layers.17.mlp.experts.125.up_proj", "model.layers.17.mlp.experts.126.up_proj", "model.layers.17.mlp.experts.127.up_proj", "model.layers.17.mlp.experts.128.up_proj", "model.layers.17.mlp.experts.129.up_proj", "model.layers.17.mlp.experts.130.up_proj", "model.layers.17.mlp.experts.131.up_proj", "model.layers.17.mlp.experts.132.up_proj", "model.layers.17.mlp.experts.133.up_proj", "model.layers.17.mlp.experts.134.up_proj", "model.layers.17.mlp.experts.135.up_proj", "model.layers.17.mlp.experts.136.up_proj", "model.layers.17.mlp.experts.137.up_proj", "model.layers.17.mlp.experts.138.up_proj", "model.layers.17.mlp.experts.139.up_proj", "model.layers.17.mlp.experts.140.up_proj", "model.layers.17.mlp.experts.141.up_proj", "model.layers.17.mlp.experts.142.up_proj", "model.layers.17.mlp.experts.143.up_proj", "model.layers.17.mlp.experts.144.up_proj", "model.layers.17.mlp.experts.145.up_proj", "model.layers.17.mlp.experts.146.up_proj", "model.layers.17.mlp.experts.147.up_proj", "model.layers.17.mlp.experts.148.up_proj", "model.layers.17.mlp.experts.149.up_proj", "model.layers.17.mlp.experts.150.up_proj", "model.layers.17.mlp.experts.151.up_proj", "model.layers.17.mlp.experts.152.up_proj", "model.layers.17.mlp.experts.153.up_proj", "model.layers.17.mlp.experts.154.up_proj", "model.layers.17.mlp.experts.155.up_proj", "model.layers.17.mlp.experts.156.up_proj", "model.layers.17.mlp.experts.157.up_proj", "model.layers.17.mlp.experts.158.up_proj", "model.layers.17.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0010001182556152455, "dbits": 2516582400 }, { "dkld": -0.0009533122181892423, "dbits": 5033164800 }, { "dkld": -0.0010655464604497067, "dbits": 7549747200 }, { "dkld": -0.001146352477371701, "dbits": 12582912000 } ] }, { "idx": 104, "layers": [ "model.layers.17.mlp.experts.0.down_proj", "model.layers.17.mlp.experts.1.down_proj", "model.layers.17.mlp.experts.2.down_proj", "model.layers.17.mlp.experts.3.down_proj", "model.layers.17.mlp.experts.4.down_proj", "model.layers.17.mlp.experts.5.down_proj", "model.layers.17.mlp.experts.6.down_proj", "model.layers.17.mlp.experts.7.down_proj", "model.layers.17.mlp.experts.8.down_proj", "model.layers.17.mlp.experts.9.down_proj", "model.layers.17.mlp.experts.10.down_proj", "model.layers.17.mlp.experts.11.down_proj", "model.layers.17.mlp.experts.12.down_proj", "model.layers.17.mlp.experts.13.down_proj", "model.layers.17.mlp.experts.14.down_proj", "model.layers.17.mlp.experts.15.down_proj", "model.layers.17.mlp.experts.16.down_proj", "model.layers.17.mlp.experts.17.down_proj", "model.layers.17.mlp.experts.18.down_proj", "model.layers.17.mlp.experts.19.down_proj", "model.layers.17.mlp.experts.20.down_proj", "model.layers.17.mlp.experts.21.down_proj", "model.layers.17.mlp.experts.22.down_proj", "model.layers.17.mlp.experts.23.down_proj", "model.layers.17.mlp.experts.24.down_proj", "model.layers.17.mlp.experts.25.down_proj", "model.layers.17.mlp.experts.26.down_proj", "model.layers.17.mlp.experts.27.down_proj", "model.layers.17.mlp.experts.28.down_proj", "model.layers.17.mlp.experts.29.down_proj", "model.layers.17.mlp.experts.30.down_proj", "model.layers.17.mlp.experts.31.down_proj", "model.layers.17.mlp.experts.32.down_proj", "model.layers.17.mlp.experts.33.down_proj", "model.layers.17.mlp.experts.34.down_proj", "model.layers.17.mlp.experts.35.down_proj", "model.layers.17.mlp.experts.36.down_proj", "model.layers.17.mlp.experts.37.down_proj", "model.layers.17.mlp.experts.38.down_proj", "model.layers.17.mlp.experts.39.down_proj", "model.layers.17.mlp.experts.40.down_proj", "model.layers.17.mlp.experts.41.down_proj", "model.layers.17.mlp.experts.42.down_proj", "model.layers.17.mlp.experts.43.down_proj", "model.layers.17.mlp.experts.44.down_proj", "model.layers.17.mlp.experts.45.down_proj", "model.layers.17.mlp.experts.46.down_proj", "model.layers.17.mlp.experts.47.down_proj", "model.layers.17.mlp.experts.48.down_proj", "model.layers.17.mlp.experts.49.down_proj", "model.layers.17.mlp.experts.50.down_proj", "model.layers.17.mlp.experts.51.down_proj", "model.layers.17.mlp.experts.52.down_proj", "model.layers.17.mlp.experts.53.down_proj", "model.layers.17.mlp.experts.54.down_proj", "model.layers.17.mlp.experts.55.down_proj", "model.layers.17.mlp.experts.56.down_proj", "model.layers.17.mlp.experts.57.down_proj", "model.layers.17.mlp.experts.58.down_proj", "model.layers.17.mlp.experts.59.down_proj", "model.layers.17.mlp.experts.60.down_proj", "model.layers.17.mlp.experts.61.down_proj", "model.layers.17.mlp.experts.62.down_proj", "model.layers.17.mlp.experts.63.down_proj", "model.layers.17.mlp.experts.64.down_proj", "model.layers.17.mlp.experts.65.down_proj", "model.layers.17.mlp.experts.66.down_proj", "model.layers.17.mlp.experts.67.down_proj", "model.layers.17.mlp.experts.68.down_proj", "model.layers.17.mlp.experts.69.down_proj", "model.layers.17.mlp.experts.70.down_proj", "model.layers.17.mlp.experts.71.down_proj", "model.layers.17.mlp.experts.72.down_proj", "model.layers.17.mlp.experts.73.down_proj", "model.layers.17.mlp.experts.74.down_proj", "model.layers.17.mlp.experts.75.down_proj", "model.layers.17.mlp.experts.76.down_proj", "model.layers.17.mlp.experts.77.down_proj", "model.layers.17.mlp.experts.78.down_proj", "model.layers.17.mlp.experts.79.down_proj", "model.layers.17.mlp.experts.80.down_proj", "model.layers.17.mlp.experts.81.down_proj", "model.layers.17.mlp.experts.82.down_proj", "model.layers.17.mlp.experts.83.down_proj", "model.layers.17.mlp.experts.84.down_proj", "model.layers.17.mlp.experts.85.down_proj", "model.layers.17.mlp.experts.86.down_proj", "model.layers.17.mlp.experts.87.down_proj", "model.layers.17.mlp.experts.88.down_proj", "model.layers.17.mlp.experts.89.down_proj", "model.layers.17.mlp.experts.90.down_proj", "model.layers.17.mlp.experts.91.down_proj", "model.layers.17.mlp.experts.92.down_proj", "model.layers.17.mlp.experts.93.down_proj", "model.layers.17.mlp.experts.94.down_proj", "model.layers.17.mlp.experts.95.down_proj", "model.layers.17.mlp.experts.96.down_proj", "model.layers.17.mlp.experts.97.down_proj", "model.layers.17.mlp.experts.98.down_proj", "model.layers.17.mlp.experts.99.down_proj", "model.layers.17.mlp.experts.100.down_proj", "model.layers.17.mlp.experts.101.down_proj", "model.layers.17.mlp.experts.102.down_proj", "model.layers.17.mlp.experts.103.down_proj", "model.layers.17.mlp.experts.104.down_proj", "model.layers.17.mlp.experts.105.down_proj", "model.layers.17.mlp.experts.106.down_proj", "model.layers.17.mlp.experts.107.down_proj", "model.layers.17.mlp.experts.108.down_proj", "model.layers.17.mlp.experts.109.down_proj", "model.layers.17.mlp.experts.110.down_proj", "model.layers.17.mlp.experts.111.down_proj", "model.layers.17.mlp.experts.112.down_proj", "model.layers.17.mlp.experts.113.down_proj", "model.layers.17.mlp.experts.114.down_proj", "model.layers.17.mlp.experts.115.down_proj", "model.layers.17.mlp.experts.116.down_proj", "model.layers.17.mlp.experts.117.down_proj", "model.layers.17.mlp.experts.118.down_proj", "model.layers.17.mlp.experts.119.down_proj", "model.layers.17.mlp.experts.120.down_proj", "model.layers.17.mlp.experts.121.down_proj", "model.layers.17.mlp.experts.122.down_proj", "model.layers.17.mlp.experts.123.down_proj", "model.layers.17.mlp.experts.124.down_proj", "model.layers.17.mlp.experts.125.down_proj", "model.layers.17.mlp.experts.126.down_proj", "model.layers.17.mlp.experts.127.down_proj", "model.layers.17.mlp.experts.128.down_proj", "model.layers.17.mlp.experts.129.down_proj", "model.layers.17.mlp.experts.130.down_proj", "model.layers.17.mlp.experts.131.down_proj", "model.layers.17.mlp.experts.132.down_proj", "model.layers.17.mlp.experts.133.down_proj", "model.layers.17.mlp.experts.134.down_proj", "model.layers.17.mlp.experts.135.down_proj", "model.layers.17.mlp.experts.136.down_proj", "model.layers.17.mlp.experts.137.down_proj", "model.layers.17.mlp.experts.138.down_proj", "model.layers.17.mlp.experts.139.down_proj", "model.layers.17.mlp.experts.140.down_proj", "model.layers.17.mlp.experts.141.down_proj", "model.layers.17.mlp.experts.142.down_proj", "model.layers.17.mlp.experts.143.down_proj", "model.layers.17.mlp.experts.144.down_proj", "model.layers.17.mlp.experts.145.down_proj", "model.layers.17.mlp.experts.146.down_proj", "model.layers.17.mlp.experts.147.down_proj", "model.layers.17.mlp.experts.148.down_proj", "model.layers.17.mlp.experts.149.down_proj", "model.layers.17.mlp.experts.150.down_proj", "model.layers.17.mlp.experts.151.down_proj", "model.layers.17.mlp.experts.152.down_proj", "model.layers.17.mlp.experts.153.down_proj", "model.layers.17.mlp.experts.154.down_proj", "model.layers.17.mlp.experts.155.down_proj", "model.layers.17.mlp.experts.156.down_proj", "model.layers.17.mlp.experts.157.down_proj", "model.layers.17.mlp.experts.158.down_proj", "model.layers.17.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0006970978341996753, "dbits": 1258291200 }, { "dkld": -0.0009425561875104904, "dbits": 2516582400 }, { "dkld": -0.0009421010501682869, "dbits": 3774873600 }, { "dkld": -0.0010118760168552482, "dbits": 6291456000 } ] }, { "idx": 105, "layers": [ "model.layers.18.self_attn.q_proj" ], "candidates": [ { "dkld": -3.4483522176748105e-05, "dbits": 62914560 }, { "dkld": 0.0002258906140923389, "dbits": 125829120 }, { "dkld": 0.0001391518861055263, "dbits": 188743680 }, { "dkld": 0.00022171288728713157, "dbits": 314572800 } ] }, { "idx": 106, "layers": [ "model.layers.18.self_attn.k_proj", "model.layers.18.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0010199104435741901, "dbits": 10485760 }, { "dkld": -0.0009138736873865183, "dbits": 20971520 }, { "dkld": -0.0010033261030912427, "dbits": 31457280 }, { "dkld": -0.0010879863053560312, "dbits": 52428800 } ] }, { "idx": 107, "layers": [ "model.layers.18.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00018266402184963226, "dbits": 62914560 }, { "dkld": -0.00019903723150492547, "dbits": 125829120 }, { "dkld": -0.0006886456161737525, "dbits": 188743680 }, { "dkld": -0.000563873536884793, "dbits": 314572800 } ] }, { "idx": 108, "layers": [ "model.layers.18.mlp.shared_experts.gate_proj", "model.layers.18.mlp.shared_experts.up_proj", "model.layers.18.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00039450414478778284, "dbits": 23592960 }, { "dkld": -0.0003716094419360244, "dbits": 47185920 }, { "dkld": -2.846363931895135e-05, "dbits": 70778880 }, { "dkld": -3.340104594827531e-05, "dbits": 117964800 } ] }, { "idx": 109, "layers": [ "model.layers.18.mlp.experts.0.gate_proj", "model.layers.18.mlp.experts.1.gate_proj", "model.layers.18.mlp.experts.2.gate_proj", "model.layers.18.mlp.experts.3.gate_proj", "model.layers.18.mlp.experts.4.gate_proj", "model.layers.18.mlp.experts.5.gate_proj", "model.layers.18.mlp.experts.6.gate_proj", "model.layers.18.mlp.experts.7.gate_proj", "model.layers.18.mlp.experts.8.gate_proj", "model.layers.18.mlp.experts.9.gate_proj", "model.layers.18.mlp.experts.10.gate_proj", "model.layers.18.mlp.experts.11.gate_proj", "model.layers.18.mlp.experts.12.gate_proj", "model.layers.18.mlp.experts.13.gate_proj", "model.layers.18.mlp.experts.14.gate_proj", "model.layers.18.mlp.experts.15.gate_proj", "model.layers.18.mlp.experts.16.gate_proj", "model.layers.18.mlp.experts.17.gate_proj", "model.layers.18.mlp.experts.18.gate_proj", "model.layers.18.mlp.experts.19.gate_proj", "model.layers.18.mlp.experts.20.gate_proj", "model.layers.18.mlp.experts.21.gate_proj", "model.layers.18.mlp.experts.22.gate_proj", "model.layers.18.mlp.experts.23.gate_proj", "model.layers.18.mlp.experts.24.gate_proj", "model.layers.18.mlp.experts.25.gate_proj", "model.layers.18.mlp.experts.26.gate_proj", "model.layers.18.mlp.experts.27.gate_proj", "model.layers.18.mlp.experts.28.gate_proj", "model.layers.18.mlp.experts.29.gate_proj", "model.layers.18.mlp.experts.30.gate_proj", "model.layers.18.mlp.experts.31.gate_proj", "model.layers.18.mlp.experts.32.gate_proj", "model.layers.18.mlp.experts.33.gate_proj", "model.layers.18.mlp.experts.34.gate_proj", "model.layers.18.mlp.experts.35.gate_proj", "model.layers.18.mlp.experts.36.gate_proj", "model.layers.18.mlp.experts.37.gate_proj", "model.layers.18.mlp.experts.38.gate_proj", "model.layers.18.mlp.experts.39.gate_proj", "model.layers.18.mlp.experts.40.gate_proj", "model.layers.18.mlp.experts.41.gate_proj", "model.layers.18.mlp.experts.42.gate_proj", "model.layers.18.mlp.experts.43.gate_proj", "model.layers.18.mlp.experts.44.gate_proj", "model.layers.18.mlp.experts.45.gate_proj", "model.layers.18.mlp.experts.46.gate_proj", "model.layers.18.mlp.experts.47.gate_proj", "model.layers.18.mlp.experts.48.gate_proj", "model.layers.18.mlp.experts.49.gate_proj", "model.layers.18.mlp.experts.50.gate_proj", "model.layers.18.mlp.experts.51.gate_proj", "model.layers.18.mlp.experts.52.gate_proj", "model.layers.18.mlp.experts.53.gate_proj", "model.layers.18.mlp.experts.54.gate_proj", "model.layers.18.mlp.experts.55.gate_proj", "model.layers.18.mlp.experts.56.gate_proj", "model.layers.18.mlp.experts.57.gate_proj", "model.layers.18.mlp.experts.58.gate_proj", "model.layers.18.mlp.experts.59.gate_proj", "model.layers.18.mlp.experts.60.gate_proj", "model.layers.18.mlp.experts.61.gate_proj", "model.layers.18.mlp.experts.62.gate_proj", "model.layers.18.mlp.experts.63.gate_proj", "model.layers.18.mlp.experts.64.gate_proj", "model.layers.18.mlp.experts.65.gate_proj", "model.layers.18.mlp.experts.66.gate_proj", "model.layers.18.mlp.experts.67.gate_proj", "model.layers.18.mlp.experts.68.gate_proj", "model.layers.18.mlp.experts.69.gate_proj", "model.layers.18.mlp.experts.70.gate_proj", "model.layers.18.mlp.experts.71.gate_proj", "model.layers.18.mlp.experts.72.gate_proj", "model.layers.18.mlp.experts.73.gate_proj", "model.layers.18.mlp.experts.74.gate_proj", "model.layers.18.mlp.experts.75.gate_proj", "model.layers.18.mlp.experts.76.gate_proj", "model.layers.18.mlp.experts.77.gate_proj", "model.layers.18.mlp.experts.78.gate_proj", "model.layers.18.mlp.experts.79.gate_proj", "model.layers.18.mlp.experts.80.gate_proj", "model.layers.18.mlp.experts.81.gate_proj", "model.layers.18.mlp.experts.82.gate_proj", "model.layers.18.mlp.experts.83.gate_proj", "model.layers.18.mlp.experts.84.gate_proj", "model.layers.18.mlp.experts.85.gate_proj", "model.layers.18.mlp.experts.86.gate_proj", "model.layers.18.mlp.experts.87.gate_proj", "model.layers.18.mlp.experts.88.gate_proj", "model.layers.18.mlp.experts.89.gate_proj", "model.layers.18.mlp.experts.90.gate_proj", "model.layers.18.mlp.experts.91.gate_proj", "model.layers.18.mlp.experts.92.gate_proj", "model.layers.18.mlp.experts.93.gate_proj", "model.layers.18.mlp.experts.94.gate_proj", "model.layers.18.mlp.experts.95.gate_proj", "model.layers.18.mlp.experts.96.gate_proj", "model.layers.18.mlp.experts.97.gate_proj", "model.layers.18.mlp.experts.98.gate_proj", "model.layers.18.mlp.experts.99.gate_proj", "model.layers.18.mlp.experts.100.gate_proj", "model.layers.18.mlp.experts.101.gate_proj", "model.layers.18.mlp.experts.102.gate_proj", "model.layers.18.mlp.experts.103.gate_proj", "model.layers.18.mlp.experts.104.gate_proj", "model.layers.18.mlp.experts.105.gate_proj", "model.layers.18.mlp.experts.106.gate_proj", "model.layers.18.mlp.experts.107.gate_proj", "model.layers.18.mlp.experts.108.gate_proj", "model.layers.18.mlp.experts.109.gate_proj", "model.layers.18.mlp.experts.110.gate_proj", "model.layers.18.mlp.experts.111.gate_proj", "model.layers.18.mlp.experts.112.gate_proj", "model.layers.18.mlp.experts.113.gate_proj", "model.layers.18.mlp.experts.114.gate_proj", "model.layers.18.mlp.experts.115.gate_proj", "model.layers.18.mlp.experts.116.gate_proj", "model.layers.18.mlp.experts.117.gate_proj", "model.layers.18.mlp.experts.118.gate_proj", "model.layers.18.mlp.experts.119.gate_proj", "model.layers.18.mlp.experts.120.gate_proj", "model.layers.18.mlp.experts.121.gate_proj", "model.layers.18.mlp.experts.122.gate_proj", "model.layers.18.mlp.experts.123.gate_proj", "model.layers.18.mlp.experts.124.gate_proj", "model.layers.18.mlp.experts.125.gate_proj", "model.layers.18.mlp.experts.126.gate_proj", "model.layers.18.mlp.experts.127.gate_proj", "model.layers.18.mlp.experts.128.gate_proj", "model.layers.18.mlp.experts.129.gate_proj", "model.layers.18.mlp.experts.130.gate_proj", "model.layers.18.mlp.experts.131.gate_proj", "model.layers.18.mlp.experts.132.gate_proj", "model.layers.18.mlp.experts.133.gate_proj", "model.layers.18.mlp.experts.134.gate_proj", "model.layers.18.mlp.experts.135.gate_proj", "model.layers.18.mlp.experts.136.gate_proj", "model.layers.18.mlp.experts.137.gate_proj", "model.layers.18.mlp.experts.138.gate_proj", "model.layers.18.mlp.experts.139.gate_proj", "model.layers.18.mlp.experts.140.gate_proj", "model.layers.18.mlp.experts.141.gate_proj", "model.layers.18.mlp.experts.142.gate_proj", "model.layers.18.mlp.experts.143.gate_proj", "model.layers.18.mlp.experts.144.gate_proj", "model.layers.18.mlp.experts.145.gate_proj", "model.layers.18.mlp.experts.146.gate_proj", "model.layers.18.mlp.experts.147.gate_proj", "model.layers.18.mlp.experts.148.gate_proj", "model.layers.18.mlp.experts.149.gate_proj", "model.layers.18.mlp.experts.150.gate_proj", "model.layers.18.mlp.experts.151.gate_proj", "model.layers.18.mlp.experts.152.gate_proj", "model.layers.18.mlp.experts.153.gate_proj", "model.layers.18.mlp.experts.154.gate_proj", "model.layers.18.mlp.experts.155.gate_proj", "model.layers.18.mlp.experts.156.gate_proj", "model.layers.18.mlp.experts.157.gate_proj", "model.layers.18.mlp.experts.158.gate_proj", "model.layers.18.mlp.experts.159.gate_proj", "model.layers.18.mlp.experts.0.up_proj", "model.layers.18.mlp.experts.1.up_proj", "model.layers.18.mlp.experts.2.up_proj", "model.layers.18.mlp.experts.3.up_proj", "model.layers.18.mlp.experts.4.up_proj", "model.layers.18.mlp.experts.5.up_proj", "model.layers.18.mlp.experts.6.up_proj", "model.layers.18.mlp.experts.7.up_proj", "model.layers.18.mlp.experts.8.up_proj", "model.layers.18.mlp.experts.9.up_proj", "model.layers.18.mlp.experts.10.up_proj", "model.layers.18.mlp.experts.11.up_proj", "model.layers.18.mlp.experts.12.up_proj", "model.layers.18.mlp.experts.13.up_proj", "model.layers.18.mlp.experts.14.up_proj", "model.layers.18.mlp.experts.15.up_proj", "model.layers.18.mlp.experts.16.up_proj", "model.layers.18.mlp.experts.17.up_proj", "model.layers.18.mlp.experts.18.up_proj", "model.layers.18.mlp.experts.19.up_proj", "model.layers.18.mlp.experts.20.up_proj", "model.layers.18.mlp.experts.21.up_proj", "model.layers.18.mlp.experts.22.up_proj", "model.layers.18.mlp.experts.23.up_proj", "model.layers.18.mlp.experts.24.up_proj", "model.layers.18.mlp.experts.25.up_proj", "model.layers.18.mlp.experts.26.up_proj", "model.layers.18.mlp.experts.27.up_proj", "model.layers.18.mlp.experts.28.up_proj", "model.layers.18.mlp.experts.29.up_proj", "model.layers.18.mlp.experts.30.up_proj", "model.layers.18.mlp.experts.31.up_proj", "model.layers.18.mlp.experts.32.up_proj", "model.layers.18.mlp.experts.33.up_proj", "model.layers.18.mlp.experts.34.up_proj", "model.layers.18.mlp.experts.35.up_proj", "model.layers.18.mlp.experts.36.up_proj", "model.layers.18.mlp.experts.37.up_proj", "model.layers.18.mlp.experts.38.up_proj", "model.layers.18.mlp.experts.39.up_proj", "model.layers.18.mlp.experts.40.up_proj", "model.layers.18.mlp.experts.41.up_proj", "model.layers.18.mlp.experts.42.up_proj", "model.layers.18.mlp.experts.43.up_proj", "model.layers.18.mlp.experts.44.up_proj", "model.layers.18.mlp.experts.45.up_proj", "model.layers.18.mlp.experts.46.up_proj", "model.layers.18.mlp.experts.47.up_proj", "model.layers.18.mlp.experts.48.up_proj", "model.layers.18.mlp.experts.49.up_proj", "model.layers.18.mlp.experts.50.up_proj", "model.layers.18.mlp.experts.51.up_proj", "model.layers.18.mlp.experts.52.up_proj", "model.layers.18.mlp.experts.53.up_proj", "model.layers.18.mlp.experts.54.up_proj", "model.layers.18.mlp.experts.55.up_proj", "model.layers.18.mlp.experts.56.up_proj", "model.layers.18.mlp.experts.57.up_proj", "model.layers.18.mlp.experts.58.up_proj", "model.layers.18.mlp.experts.59.up_proj", "model.layers.18.mlp.experts.60.up_proj", "model.layers.18.mlp.experts.61.up_proj", "model.layers.18.mlp.experts.62.up_proj", "model.layers.18.mlp.experts.63.up_proj", "model.layers.18.mlp.experts.64.up_proj", "model.layers.18.mlp.experts.65.up_proj", "model.layers.18.mlp.experts.66.up_proj", "model.layers.18.mlp.experts.67.up_proj", "model.layers.18.mlp.experts.68.up_proj", "model.layers.18.mlp.experts.69.up_proj", "model.layers.18.mlp.experts.70.up_proj", "model.layers.18.mlp.experts.71.up_proj", "model.layers.18.mlp.experts.72.up_proj", "model.layers.18.mlp.experts.73.up_proj", "model.layers.18.mlp.experts.74.up_proj", "model.layers.18.mlp.experts.75.up_proj", "model.layers.18.mlp.experts.76.up_proj", "model.layers.18.mlp.experts.77.up_proj", "model.layers.18.mlp.experts.78.up_proj", "model.layers.18.mlp.experts.79.up_proj", "model.layers.18.mlp.experts.80.up_proj", "model.layers.18.mlp.experts.81.up_proj", "model.layers.18.mlp.experts.82.up_proj", "model.layers.18.mlp.experts.83.up_proj", "model.layers.18.mlp.experts.84.up_proj", "model.layers.18.mlp.experts.85.up_proj", "model.layers.18.mlp.experts.86.up_proj", "model.layers.18.mlp.experts.87.up_proj", "model.layers.18.mlp.experts.88.up_proj", "model.layers.18.mlp.experts.89.up_proj", "model.layers.18.mlp.experts.90.up_proj", "model.layers.18.mlp.experts.91.up_proj", "model.layers.18.mlp.experts.92.up_proj", "model.layers.18.mlp.experts.93.up_proj", "model.layers.18.mlp.experts.94.up_proj", "model.layers.18.mlp.experts.95.up_proj", "model.layers.18.mlp.experts.96.up_proj", "model.layers.18.mlp.experts.97.up_proj", "model.layers.18.mlp.experts.98.up_proj", "model.layers.18.mlp.experts.99.up_proj", "model.layers.18.mlp.experts.100.up_proj", "model.layers.18.mlp.experts.101.up_proj", "model.layers.18.mlp.experts.102.up_proj", "model.layers.18.mlp.experts.103.up_proj", "model.layers.18.mlp.experts.104.up_proj", "model.layers.18.mlp.experts.105.up_proj", "model.layers.18.mlp.experts.106.up_proj", "model.layers.18.mlp.experts.107.up_proj", "model.layers.18.mlp.experts.108.up_proj", "model.layers.18.mlp.experts.109.up_proj", "model.layers.18.mlp.experts.110.up_proj", "model.layers.18.mlp.experts.111.up_proj", "model.layers.18.mlp.experts.112.up_proj", "model.layers.18.mlp.experts.113.up_proj", "model.layers.18.mlp.experts.114.up_proj", "model.layers.18.mlp.experts.115.up_proj", "model.layers.18.mlp.experts.116.up_proj", "model.layers.18.mlp.experts.117.up_proj", "model.layers.18.mlp.experts.118.up_proj", "model.layers.18.mlp.experts.119.up_proj", "model.layers.18.mlp.experts.120.up_proj", "model.layers.18.mlp.experts.121.up_proj", "model.layers.18.mlp.experts.122.up_proj", "model.layers.18.mlp.experts.123.up_proj", "model.layers.18.mlp.experts.124.up_proj", "model.layers.18.mlp.experts.125.up_proj", "model.layers.18.mlp.experts.126.up_proj", "model.layers.18.mlp.experts.127.up_proj", "model.layers.18.mlp.experts.128.up_proj", "model.layers.18.mlp.experts.129.up_proj", "model.layers.18.mlp.experts.130.up_proj", "model.layers.18.mlp.experts.131.up_proj", "model.layers.18.mlp.experts.132.up_proj", "model.layers.18.mlp.experts.133.up_proj", "model.layers.18.mlp.experts.134.up_proj", "model.layers.18.mlp.experts.135.up_proj", "model.layers.18.mlp.experts.136.up_proj", "model.layers.18.mlp.experts.137.up_proj", "model.layers.18.mlp.experts.138.up_proj", "model.layers.18.mlp.experts.139.up_proj", "model.layers.18.mlp.experts.140.up_proj", "model.layers.18.mlp.experts.141.up_proj", "model.layers.18.mlp.experts.142.up_proj", "model.layers.18.mlp.experts.143.up_proj", "model.layers.18.mlp.experts.144.up_proj", "model.layers.18.mlp.experts.145.up_proj", "model.layers.18.mlp.experts.146.up_proj", "model.layers.18.mlp.experts.147.up_proj", "model.layers.18.mlp.experts.148.up_proj", "model.layers.18.mlp.experts.149.up_proj", "model.layers.18.mlp.experts.150.up_proj", "model.layers.18.mlp.experts.151.up_proj", "model.layers.18.mlp.experts.152.up_proj", "model.layers.18.mlp.experts.153.up_proj", "model.layers.18.mlp.experts.154.up_proj", "model.layers.18.mlp.experts.155.up_proj", "model.layers.18.mlp.experts.156.up_proj", "model.layers.18.mlp.experts.157.up_proj", "model.layers.18.mlp.experts.158.up_proj", "model.layers.18.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000898025836795574, "dbits": 2516582400 }, { "dkld": -0.0009699675254523782, "dbits": 5033164800 }, { "dkld": -0.0010545150376856383, "dbits": 7549747200 }, { "dkld": -0.0010349531657993793, "dbits": 12582912000 } ] }, { "idx": 110, "layers": [ "model.layers.18.mlp.experts.0.down_proj", "model.layers.18.mlp.experts.1.down_proj", "model.layers.18.mlp.experts.2.down_proj", "model.layers.18.mlp.experts.3.down_proj", "model.layers.18.mlp.experts.4.down_proj", "model.layers.18.mlp.experts.5.down_proj", "model.layers.18.mlp.experts.6.down_proj", "model.layers.18.mlp.experts.7.down_proj", "model.layers.18.mlp.experts.8.down_proj", "model.layers.18.mlp.experts.9.down_proj", "model.layers.18.mlp.experts.10.down_proj", "model.layers.18.mlp.experts.11.down_proj", "model.layers.18.mlp.experts.12.down_proj", "model.layers.18.mlp.experts.13.down_proj", "model.layers.18.mlp.experts.14.down_proj", "model.layers.18.mlp.experts.15.down_proj", "model.layers.18.mlp.experts.16.down_proj", "model.layers.18.mlp.experts.17.down_proj", "model.layers.18.mlp.experts.18.down_proj", "model.layers.18.mlp.experts.19.down_proj", "model.layers.18.mlp.experts.20.down_proj", "model.layers.18.mlp.experts.21.down_proj", "model.layers.18.mlp.experts.22.down_proj", "model.layers.18.mlp.experts.23.down_proj", "model.layers.18.mlp.experts.24.down_proj", "model.layers.18.mlp.experts.25.down_proj", "model.layers.18.mlp.experts.26.down_proj", "model.layers.18.mlp.experts.27.down_proj", "model.layers.18.mlp.experts.28.down_proj", "model.layers.18.mlp.experts.29.down_proj", "model.layers.18.mlp.experts.30.down_proj", "model.layers.18.mlp.experts.31.down_proj", "model.layers.18.mlp.experts.32.down_proj", "model.layers.18.mlp.experts.33.down_proj", "model.layers.18.mlp.experts.34.down_proj", "model.layers.18.mlp.experts.35.down_proj", "model.layers.18.mlp.experts.36.down_proj", "model.layers.18.mlp.experts.37.down_proj", "model.layers.18.mlp.experts.38.down_proj", "model.layers.18.mlp.experts.39.down_proj", "model.layers.18.mlp.experts.40.down_proj", "model.layers.18.mlp.experts.41.down_proj", "model.layers.18.mlp.experts.42.down_proj", "model.layers.18.mlp.experts.43.down_proj", "model.layers.18.mlp.experts.44.down_proj", "model.layers.18.mlp.experts.45.down_proj", "model.layers.18.mlp.experts.46.down_proj", "model.layers.18.mlp.experts.47.down_proj", "model.layers.18.mlp.experts.48.down_proj", "model.layers.18.mlp.experts.49.down_proj", "model.layers.18.mlp.experts.50.down_proj", "model.layers.18.mlp.experts.51.down_proj", "model.layers.18.mlp.experts.52.down_proj", "model.layers.18.mlp.experts.53.down_proj", "model.layers.18.mlp.experts.54.down_proj", "model.layers.18.mlp.experts.55.down_proj", "model.layers.18.mlp.experts.56.down_proj", "model.layers.18.mlp.experts.57.down_proj", "model.layers.18.mlp.experts.58.down_proj", "model.layers.18.mlp.experts.59.down_proj", "model.layers.18.mlp.experts.60.down_proj", "model.layers.18.mlp.experts.61.down_proj", "model.layers.18.mlp.experts.62.down_proj", "model.layers.18.mlp.experts.63.down_proj", "model.layers.18.mlp.experts.64.down_proj", "model.layers.18.mlp.experts.65.down_proj", "model.layers.18.mlp.experts.66.down_proj", "model.layers.18.mlp.experts.67.down_proj", "model.layers.18.mlp.experts.68.down_proj", "model.layers.18.mlp.experts.69.down_proj", "model.layers.18.mlp.experts.70.down_proj", "model.layers.18.mlp.experts.71.down_proj", "model.layers.18.mlp.experts.72.down_proj", "model.layers.18.mlp.experts.73.down_proj", "model.layers.18.mlp.experts.74.down_proj", "model.layers.18.mlp.experts.75.down_proj", "model.layers.18.mlp.experts.76.down_proj", "model.layers.18.mlp.experts.77.down_proj", "model.layers.18.mlp.experts.78.down_proj", "model.layers.18.mlp.experts.79.down_proj", "model.layers.18.mlp.experts.80.down_proj", "model.layers.18.mlp.experts.81.down_proj", "model.layers.18.mlp.experts.82.down_proj", "model.layers.18.mlp.experts.83.down_proj", "model.layers.18.mlp.experts.84.down_proj", "model.layers.18.mlp.experts.85.down_proj", "model.layers.18.mlp.experts.86.down_proj", "model.layers.18.mlp.experts.87.down_proj", "model.layers.18.mlp.experts.88.down_proj", "model.layers.18.mlp.experts.89.down_proj", "model.layers.18.mlp.experts.90.down_proj", "model.layers.18.mlp.experts.91.down_proj", "model.layers.18.mlp.experts.92.down_proj", "model.layers.18.mlp.experts.93.down_proj", "model.layers.18.mlp.experts.94.down_proj", "model.layers.18.mlp.experts.95.down_proj", "model.layers.18.mlp.experts.96.down_proj", "model.layers.18.mlp.experts.97.down_proj", "model.layers.18.mlp.experts.98.down_proj", "model.layers.18.mlp.experts.99.down_proj", "model.layers.18.mlp.experts.100.down_proj", "model.layers.18.mlp.experts.101.down_proj", "model.layers.18.mlp.experts.102.down_proj", "model.layers.18.mlp.experts.103.down_proj", "model.layers.18.mlp.experts.104.down_proj", "model.layers.18.mlp.experts.105.down_proj", "model.layers.18.mlp.experts.106.down_proj", "model.layers.18.mlp.experts.107.down_proj", "model.layers.18.mlp.experts.108.down_proj", "model.layers.18.mlp.experts.109.down_proj", "model.layers.18.mlp.experts.110.down_proj", "model.layers.18.mlp.experts.111.down_proj", "model.layers.18.mlp.experts.112.down_proj", "model.layers.18.mlp.experts.113.down_proj", "model.layers.18.mlp.experts.114.down_proj", "model.layers.18.mlp.experts.115.down_proj", "model.layers.18.mlp.experts.116.down_proj", "model.layers.18.mlp.experts.117.down_proj", "model.layers.18.mlp.experts.118.down_proj", "model.layers.18.mlp.experts.119.down_proj", "model.layers.18.mlp.experts.120.down_proj", "model.layers.18.mlp.experts.121.down_proj", "model.layers.18.mlp.experts.122.down_proj", "model.layers.18.mlp.experts.123.down_proj", "model.layers.18.mlp.experts.124.down_proj", "model.layers.18.mlp.experts.125.down_proj", "model.layers.18.mlp.experts.126.down_proj", "model.layers.18.mlp.experts.127.down_proj", "model.layers.18.mlp.experts.128.down_proj", "model.layers.18.mlp.experts.129.down_proj", "model.layers.18.mlp.experts.130.down_proj", "model.layers.18.mlp.experts.131.down_proj", "model.layers.18.mlp.experts.132.down_proj", "model.layers.18.mlp.experts.133.down_proj", "model.layers.18.mlp.experts.134.down_proj", "model.layers.18.mlp.experts.135.down_proj", "model.layers.18.mlp.experts.136.down_proj", "model.layers.18.mlp.experts.137.down_proj", "model.layers.18.mlp.experts.138.down_proj", "model.layers.18.mlp.experts.139.down_proj", "model.layers.18.mlp.experts.140.down_proj", "model.layers.18.mlp.experts.141.down_proj", "model.layers.18.mlp.experts.142.down_proj", "model.layers.18.mlp.experts.143.down_proj", "model.layers.18.mlp.experts.144.down_proj", "model.layers.18.mlp.experts.145.down_proj", "model.layers.18.mlp.experts.146.down_proj", "model.layers.18.mlp.experts.147.down_proj", "model.layers.18.mlp.experts.148.down_proj", "model.layers.18.mlp.experts.149.down_proj", "model.layers.18.mlp.experts.150.down_proj", "model.layers.18.mlp.experts.151.down_proj", "model.layers.18.mlp.experts.152.down_proj", "model.layers.18.mlp.experts.153.down_proj", "model.layers.18.mlp.experts.154.down_proj", "model.layers.18.mlp.experts.155.down_proj", "model.layers.18.mlp.experts.156.down_proj", "model.layers.18.mlp.experts.157.down_proj", "model.layers.18.mlp.experts.158.down_proj", "model.layers.18.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00034283977001905996, "dbits": 1258291200 }, { "dkld": -0.0003990823403000887, "dbits": 2516582400 }, { "dkld": -0.0005258155055344188, "dbits": 3774873600 }, { "dkld": -0.0004948165267705945, "dbits": 6291456000 } ] }, { "idx": 111, "layers": [ "model.layers.19.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00021267496049404144, "dbits": 62914560 }, { "dkld": -3.8691703230148144e-05, "dbits": 125829120 }, { "dkld": 1.548808068035923e-05, "dbits": 188743680 }, { "dkld": -4.26903367042597e-05, "dbits": 314572800 } ] }, { "idx": 112, "layers": [ "model.layers.19.self_attn.k_proj", "model.layers.19.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0005873613990843296, "dbits": 10485760 }, { "dkld": 0.0007910178974270821, "dbits": 20971520 }, { "dkld": 0.0005598453804850495, "dbits": 31457280 }, { "dkld": 0.00043460000306366764, "dbits": 52428800 } ] }, { "idx": 113, "layers": [ "model.layers.19.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0007860947400331608, "dbits": 62914560 }, { "dkld": -0.0008889584802091177, "dbits": 125829120 }, { "dkld": -0.0009424891322851181, "dbits": 188743680 }, { "dkld": -0.0009264284744858742, "dbits": 314572800 } ] }, { "idx": 114, "layers": [ "model.layers.19.mlp.shared_experts.gate_proj", "model.layers.19.mlp.shared_experts.up_proj", "model.layers.19.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.001623982004821306, "dbits": 23592960 }, { "dkld": -0.000980689283460387, "dbits": 47185920 }, { "dkld": -0.0007026731036603451, "dbits": 70778880 }, { "dkld": -0.000811417214572438, "dbits": 117964800 } ] }, { "idx": 115, "layers": [ "model.layers.19.mlp.experts.0.gate_proj", "model.layers.19.mlp.experts.1.gate_proj", "model.layers.19.mlp.experts.2.gate_proj", "model.layers.19.mlp.experts.3.gate_proj", "model.layers.19.mlp.experts.4.gate_proj", "model.layers.19.mlp.experts.5.gate_proj", "model.layers.19.mlp.experts.6.gate_proj", "model.layers.19.mlp.experts.7.gate_proj", "model.layers.19.mlp.experts.8.gate_proj", "model.layers.19.mlp.experts.9.gate_proj", "model.layers.19.mlp.experts.10.gate_proj", "model.layers.19.mlp.experts.11.gate_proj", "model.layers.19.mlp.experts.12.gate_proj", "model.layers.19.mlp.experts.13.gate_proj", "model.layers.19.mlp.experts.14.gate_proj", "model.layers.19.mlp.experts.15.gate_proj", "model.layers.19.mlp.experts.16.gate_proj", "model.layers.19.mlp.experts.17.gate_proj", "model.layers.19.mlp.experts.18.gate_proj", "model.layers.19.mlp.experts.19.gate_proj", "model.layers.19.mlp.experts.20.gate_proj", "model.layers.19.mlp.experts.21.gate_proj", "model.layers.19.mlp.experts.22.gate_proj", "model.layers.19.mlp.experts.23.gate_proj", "model.layers.19.mlp.experts.24.gate_proj", "model.layers.19.mlp.experts.25.gate_proj", "model.layers.19.mlp.experts.26.gate_proj", "model.layers.19.mlp.experts.27.gate_proj", "model.layers.19.mlp.experts.28.gate_proj", "model.layers.19.mlp.experts.29.gate_proj", "model.layers.19.mlp.experts.30.gate_proj", "model.layers.19.mlp.experts.31.gate_proj", "model.layers.19.mlp.experts.32.gate_proj", "model.layers.19.mlp.experts.33.gate_proj", "model.layers.19.mlp.experts.34.gate_proj", "model.layers.19.mlp.experts.35.gate_proj", "model.layers.19.mlp.experts.36.gate_proj", "model.layers.19.mlp.experts.37.gate_proj", "model.layers.19.mlp.experts.38.gate_proj", "model.layers.19.mlp.experts.39.gate_proj", "model.layers.19.mlp.experts.40.gate_proj", "model.layers.19.mlp.experts.41.gate_proj", "model.layers.19.mlp.experts.42.gate_proj", "model.layers.19.mlp.experts.43.gate_proj", "model.layers.19.mlp.experts.44.gate_proj", "model.layers.19.mlp.experts.45.gate_proj", "model.layers.19.mlp.experts.46.gate_proj", "model.layers.19.mlp.experts.47.gate_proj", "model.layers.19.mlp.experts.48.gate_proj", "model.layers.19.mlp.experts.49.gate_proj", "model.layers.19.mlp.experts.50.gate_proj", "model.layers.19.mlp.experts.51.gate_proj", "model.layers.19.mlp.experts.52.gate_proj", "model.layers.19.mlp.experts.53.gate_proj", "model.layers.19.mlp.experts.54.gate_proj", "model.layers.19.mlp.experts.55.gate_proj", "model.layers.19.mlp.experts.56.gate_proj", "model.layers.19.mlp.experts.57.gate_proj", "model.layers.19.mlp.experts.58.gate_proj", "model.layers.19.mlp.experts.59.gate_proj", "model.layers.19.mlp.experts.60.gate_proj", "model.layers.19.mlp.experts.61.gate_proj", "model.layers.19.mlp.experts.62.gate_proj", "model.layers.19.mlp.experts.63.gate_proj", "model.layers.19.mlp.experts.64.gate_proj", "model.layers.19.mlp.experts.65.gate_proj", "model.layers.19.mlp.experts.66.gate_proj", "model.layers.19.mlp.experts.67.gate_proj", "model.layers.19.mlp.experts.68.gate_proj", "model.layers.19.mlp.experts.69.gate_proj", "model.layers.19.mlp.experts.70.gate_proj", "model.layers.19.mlp.experts.71.gate_proj", "model.layers.19.mlp.experts.72.gate_proj", "model.layers.19.mlp.experts.73.gate_proj", "model.layers.19.mlp.experts.74.gate_proj", "model.layers.19.mlp.experts.75.gate_proj", "model.layers.19.mlp.experts.76.gate_proj", "model.layers.19.mlp.experts.77.gate_proj", "model.layers.19.mlp.experts.78.gate_proj", "model.layers.19.mlp.experts.79.gate_proj", "model.layers.19.mlp.experts.80.gate_proj", "model.layers.19.mlp.experts.81.gate_proj", "model.layers.19.mlp.experts.82.gate_proj", "model.layers.19.mlp.experts.83.gate_proj", "model.layers.19.mlp.experts.84.gate_proj", "model.layers.19.mlp.experts.85.gate_proj", "model.layers.19.mlp.experts.86.gate_proj", "model.layers.19.mlp.experts.87.gate_proj", "model.layers.19.mlp.experts.88.gate_proj", "model.layers.19.mlp.experts.89.gate_proj", "model.layers.19.mlp.experts.90.gate_proj", "model.layers.19.mlp.experts.91.gate_proj", "model.layers.19.mlp.experts.92.gate_proj", "model.layers.19.mlp.experts.93.gate_proj", "model.layers.19.mlp.experts.94.gate_proj", "model.layers.19.mlp.experts.95.gate_proj", "model.layers.19.mlp.experts.96.gate_proj", "model.layers.19.mlp.experts.97.gate_proj", "model.layers.19.mlp.experts.98.gate_proj", "model.layers.19.mlp.experts.99.gate_proj", "model.layers.19.mlp.experts.100.gate_proj", "model.layers.19.mlp.experts.101.gate_proj", "model.layers.19.mlp.experts.102.gate_proj", "model.layers.19.mlp.experts.103.gate_proj", "model.layers.19.mlp.experts.104.gate_proj", "model.layers.19.mlp.experts.105.gate_proj", "model.layers.19.mlp.experts.106.gate_proj", "model.layers.19.mlp.experts.107.gate_proj", "model.layers.19.mlp.experts.108.gate_proj", "model.layers.19.mlp.experts.109.gate_proj", "model.layers.19.mlp.experts.110.gate_proj", "model.layers.19.mlp.experts.111.gate_proj", "model.layers.19.mlp.experts.112.gate_proj", "model.layers.19.mlp.experts.113.gate_proj", "model.layers.19.mlp.experts.114.gate_proj", "model.layers.19.mlp.experts.115.gate_proj", "model.layers.19.mlp.experts.116.gate_proj", "model.layers.19.mlp.experts.117.gate_proj", "model.layers.19.mlp.experts.118.gate_proj", "model.layers.19.mlp.experts.119.gate_proj", "model.layers.19.mlp.experts.120.gate_proj", "model.layers.19.mlp.experts.121.gate_proj", "model.layers.19.mlp.experts.122.gate_proj", "model.layers.19.mlp.experts.123.gate_proj", "model.layers.19.mlp.experts.124.gate_proj", "model.layers.19.mlp.experts.125.gate_proj", "model.layers.19.mlp.experts.126.gate_proj", "model.layers.19.mlp.experts.127.gate_proj", "model.layers.19.mlp.experts.128.gate_proj", "model.layers.19.mlp.experts.129.gate_proj", "model.layers.19.mlp.experts.130.gate_proj", "model.layers.19.mlp.experts.131.gate_proj", "model.layers.19.mlp.experts.132.gate_proj", "model.layers.19.mlp.experts.133.gate_proj", "model.layers.19.mlp.experts.134.gate_proj", "model.layers.19.mlp.experts.135.gate_proj", "model.layers.19.mlp.experts.136.gate_proj", "model.layers.19.mlp.experts.137.gate_proj", "model.layers.19.mlp.experts.138.gate_proj", "model.layers.19.mlp.experts.139.gate_proj", "model.layers.19.mlp.experts.140.gate_proj", "model.layers.19.mlp.experts.141.gate_proj", "model.layers.19.mlp.experts.142.gate_proj", "model.layers.19.mlp.experts.143.gate_proj", "model.layers.19.mlp.experts.144.gate_proj", "model.layers.19.mlp.experts.145.gate_proj", "model.layers.19.mlp.experts.146.gate_proj", "model.layers.19.mlp.experts.147.gate_proj", "model.layers.19.mlp.experts.148.gate_proj", "model.layers.19.mlp.experts.149.gate_proj", "model.layers.19.mlp.experts.150.gate_proj", "model.layers.19.mlp.experts.151.gate_proj", "model.layers.19.mlp.experts.152.gate_proj", "model.layers.19.mlp.experts.153.gate_proj", "model.layers.19.mlp.experts.154.gate_proj", "model.layers.19.mlp.experts.155.gate_proj", "model.layers.19.mlp.experts.156.gate_proj", "model.layers.19.mlp.experts.157.gate_proj", "model.layers.19.mlp.experts.158.gate_proj", "model.layers.19.mlp.experts.159.gate_proj", "model.layers.19.mlp.experts.0.up_proj", "model.layers.19.mlp.experts.1.up_proj", "model.layers.19.mlp.experts.2.up_proj", "model.layers.19.mlp.experts.3.up_proj", "model.layers.19.mlp.experts.4.up_proj", "model.layers.19.mlp.experts.5.up_proj", "model.layers.19.mlp.experts.6.up_proj", "model.layers.19.mlp.experts.7.up_proj", "model.layers.19.mlp.experts.8.up_proj", "model.layers.19.mlp.experts.9.up_proj", "model.layers.19.mlp.experts.10.up_proj", "model.layers.19.mlp.experts.11.up_proj", "model.layers.19.mlp.experts.12.up_proj", "model.layers.19.mlp.experts.13.up_proj", "model.layers.19.mlp.experts.14.up_proj", "model.layers.19.mlp.experts.15.up_proj", "model.layers.19.mlp.experts.16.up_proj", "model.layers.19.mlp.experts.17.up_proj", "model.layers.19.mlp.experts.18.up_proj", "model.layers.19.mlp.experts.19.up_proj", "model.layers.19.mlp.experts.20.up_proj", "model.layers.19.mlp.experts.21.up_proj", "model.layers.19.mlp.experts.22.up_proj", "model.layers.19.mlp.experts.23.up_proj", "model.layers.19.mlp.experts.24.up_proj", "model.layers.19.mlp.experts.25.up_proj", "model.layers.19.mlp.experts.26.up_proj", "model.layers.19.mlp.experts.27.up_proj", "model.layers.19.mlp.experts.28.up_proj", "model.layers.19.mlp.experts.29.up_proj", "model.layers.19.mlp.experts.30.up_proj", "model.layers.19.mlp.experts.31.up_proj", "model.layers.19.mlp.experts.32.up_proj", "model.layers.19.mlp.experts.33.up_proj", "model.layers.19.mlp.experts.34.up_proj", "model.layers.19.mlp.experts.35.up_proj", "model.layers.19.mlp.experts.36.up_proj", "model.layers.19.mlp.experts.37.up_proj", "model.layers.19.mlp.experts.38.up_proj", "model.layers.19.mlp.experts.39.up_proj", "model.layers.19.mlp.experts.40.up_proj", "model.layers.19.mlp.experts.41.up_proj", "model.layers.19.mlp.experts.42.up_proj", "model.layers.19.mlp.experts.43.up_proj", "model.layers.19.mlp.experts.44.up_proj", "model.layers.19.mlp.experts.45.up_proj", "model.layers.19.mlp.experts.46.up_proj", "model.layers.19.mlp.experts.47.up_proj", "model.layers.19.mlp.experts.48.up_proj", "model.layers.19.mlp.experts.49.up_proj", "model.layers.19.mlp.experts.50.up_proj", "model.layers.19.mlp.experts.51.up_proj", "model.layers.19.mlp.experts.52.up_proj", "model.layers.19.mlp.experts.53.up_proj", "model.layers.19.mlp.experts.54.up_proj", "model.layers.19.mlp.experts.55.up_proj", "model.layers.19.mlp.experts.56.up_proj", "model.layers.19.mlp.experts.57.up_proj", "model.layers.19.mlp.experts.58.up_proj", "model.layers.19.mlp.experts.59.up_proj", "model.layers.19.mlp.experts.60.up_proj", "model.layers.19.mlp.experts.61.up_proj", "model.layers.19.mlp.experts.62.up_proj", "model.layers.19.mlp.experts.63.up_proj", "model.layers.19.mlp.experts.64.up_proj", "model.layers.19.mlp.experts.65.up_proj", "model.layers.19.mlp.experts.66.up_proj", "model.layers.19.mlp.experts.67.up_proj", "model.layers.19.mlp.experts.68.up_proj", "model.layers.19.mlp.experts.69.up_proj", "model.layers.19.mlp.experts.70.up_proj", "model.layers.19.mlp.experts.71.up_proj", "model.layers.19.mlp.experts.72.up_proj", "model.layers.19.mlp.experts.73.up_proj", "model.layers.19.mlp.experts.74.up_proj", "model.layers.19.mlp.experts.75.up_proj", "model.layers.19.mlp.experts.76.up_proj", "model.layers.19.mlp.experts.77.up_proj", "model.layers.19.mlp.experts.78.up_proj", "model.layers.19.mlp.experts.79.up_proj", "model.layers.19.mlp.experts.80.up_proj", "model.layers.19.mlp.experts.81.up_proj", "model.layers.19.mlp.experts.82.up_proj", "model.layers.19.mlp.experts.83.up_proj", "model.layers.19.mlp.experts.84.up_proj", "model.layers.19.mlp.experts.85.up_proj", "model.layers.19.mlp.experts.86.up_proj", "model.layers.19.mlp.experts.87.up_proj", "model.layers.19.mlp.experts.88.up_proj", "model.layers.19.mlp.experts.89.up_proj", "model.layers.19.mlp.experts.90.up_proj", "model.layers.19.mlp.experts.91.up_proj", "model.layers.19.mlp.experts.92.up_proj", "model.layers.19.mlp.experts.93.up_proj", "model.layers.19.mlp.experts.94.up_proj", "model.layers.19.mlp.experts.95.up_proj", "model.layers.19.mlp.experts.96.up_proj", "model.layers.19.mlp.experts.97.up_proj", "model.layers.19.mlp.experts.98.up_proj", "model.layers.19.mlp.experts.99.up_proj", "model.layers.19.mlp.experts.100.up_proj", "model.layers.19.mlp.experts.101.up_proj", "model.layers.19.mlp.experts.102.up_proj", "model.layers.19.mlp.experts.103.up_proj", "model.layers.19.mlp.experts.104.up_proj", "model.layers.19.mlp.experts.105.up_proj", "model.layers.19.mlp.experts.106.up_proj", "model.layers.19.mlp.experts.107.up_proj", "model.layers.19.mlp.experts.108.up_proj", "model.layers.19.mlp.experts.109.up_proj", "model.layers.19.mlp.experts.110.up_proj", "model.layers.19.mlp.experts.111.up_proj", "model.layers.19.mlp.experts.112.up_proj", "model.layers.19.mlp.experts.113.up_proj", "model.layers.19.mlp.experts.114.up_proj", "model.layers.19.mlp.experts.115.up_proj", "model.layers.19.mlp.experts.116.up_proj", "model.layers.19.mlp.experts.117.up_proj", "model.layers.19.mlp.experts.118.up_proj", "model.layers.19.mlp.experts.119.up_proj", "model.layers.19.mlp.experts.120.up_proj", "model.layers.19.mlp.experts.121.up_proj", "model.layers.19.mlp.experts.122.up_proj", "model.layers.19.mlp.experts.123.up_proj", "model.layers.19.mlp.experts.124.up_proj", "model.layers.19.mlp.experts.125.up_proj", "model.layers.19.mlp.experts.126.up_proj", "model.layers.19.mlp.experts.127.up_proj", "model.layers.19.mlp.experts.128.up_proj", "model.layers.19.mlp.experts.129.up_proj", "model.layers.19.mlp.experts.130.up_proj", "model.layers.19.mlp.experts.131.up_proj", "model.layers.19.mlp.experts.132.up_proj", "model.layers.19.mlp.experts.133.up_proj", "model.layers.19.mlp.experts.134.up_proj", "model.layers.19.mlp.experts.135.up_proj", "model.layers.19.mlp.experts.136.up_proj", "model.layers.19.mlp.experts.137.up_proj", "model.layers.19.mlp.experts.138.up_proj", "model.layers.19.mlp.experts.139.up_proj", "model.layers.19.mlp.experts.140.up_proj", "model.layers.19.mlp.experts.141.up_proj", "model.layers.19.mlp.experts.142.up_proj", "model.layers.19.mlp.experts.143.up_proj", "model.layers.19.mlp.experts.144.up_proj", "model.layers.19.mlp.experts.145.up_proj", "model.layers.19.mlp.experts.146.up_proj", "model.layers.19.mlp.experts.147.up_proj", "model.layers.19.mlp.experts.148.up_proj", "model.layers.19.mlp.experts.149.up_proj", "model.layers.19.mlp.experts.150.up_proj", "model.layers.19.mlp.experts.151.up_proj", "model.layers.19.mlp.experts.152.up_proj", "model.layers.19.mlp.experts.153.up_proj", "model.layers.19.mlp.experts.154.up_proj", "model.layers.19.mlp.experts.155.up_proj", "model.layers.19.mlp.experts.156.up_proj", "model.layers.19.mlp.experts.157.up_proj", "model.layers.19.mlp.experts.158.up_proj", "model.layers.19.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0009639196097850911, "dbits": 2516582400 }, { "dkld": -0.00104956300929189, "dbits": 5033164800 }, { "dkld": -0.0010783595964312664, "dbits": 7549747200 }, { "dkld": -0.0009614605456590708, "dbits": 12582912000 } ] }, { "idx": 116, "layers": [ "model.layers.19.mlp.experts.0.down_proj", "model.layers.19.mlp.experts.1.down_proj", "model.layers.19.mlp.experts.2.down_proj", "model.layers.19.mlp.experts.3.down_proj", "model.layers.19.mlp.experts.4.down_proj", "model.layers.19.mlp.experts.5.down_proj", "model.layers.19.mlp.experts.6.down_proj", "model.layers.19.mlp.experts.7.down_proj", "model.layers.19.mlp.experts.8.down_proj", "model.layers.19.mlp.experts.9.down_proj", "model.layers.19.mlp.experts.10.down_proj", "model.layers.19.mlp.experts.11.down_proj", "model.layers.19.mlp.experts.12.down_proj", "model.layers.19.mlp.experts.13.down_proj", "model.layers.19.mlp.experts.14.down_proj", "model.layers.19.mlp.experts.15.down_proj", "model.layers.19.mlp.experts.16.down_proj", "model.layers.19.mlp.experts.17.down_proj", "model.layers.19.mlp.experts.18.down_proj", "model.layers.19.mlp.experts.19.down_proj", "model.layers.19.mlp.experts.20.down_proj", "model.layers.19.mlp.experts.21.down_proj", "model.layers.19.mlp.experts.22.down_proj", "model.layers.19.mlp.experts.23.down_proj", "model.layers.19.mlp.experts.24.down_proj", "model.layers.19.mlp.experts.25.down_proj", "model.layers.19.mlp.experts.26.down_proj", "model.layers.19.mlp.experts.27.down_proj", "model.layers.19.mlp.experts.28.down_proj", "model.layers.19.mlp.experts.29.down_proj", "model.layers.19.mlp.experts.30.down_proj", "model.layers.19.mlp.experts.31.down_proj", "model.layers.19.mlp.experts.32.down_proj", "model.layers.19.mlp.experts.33.down_proj", "model.layers.19.mlp.experts.34.down_proj", "model.layers.19.mlp.experts.35.down_proj", "model.layers.19.mlp.experts.36.down_proj", "model.layers.19.mlp.experts.37.down_proj", "model.layers.19.mlp.experts.38.down_proj", "model.layers.19.mlp.experts.39.down_proj", "model.layers.19.mlp.experts.40.down_proj", "model.layers.19.mlp.experts.41.down_proj", "model.layers.19.mlp.experts.42.down_proj", "model.layers.19.mlp.experts.43.down_proj", "model.layers.19.mlp.experts.44.down_proj", "model.layers.19.mlp.experts.45.down_proj", "model.layers.19.mlp.experts.46.down_proj", "model.layers.19.mlp.experts.47.down_proj", "model.layers.19.mlp.experts.48.down_proj", "model.layers.19.mlp.experts.49.down_proj", "model.layers.19.mlp.experts.50.down_proj", "model.layers.19.mlp.experts.51.down_proj", "model.layers.19.mlp.experts.52.down_proj", "model.layers.19.mlp.experts.53.down_proj", "model.layers.19.mlp.experts.54.down_proj", "model.layers.19.mlp.experts.55.down_proj", "model.layers.19.mlp.experts.56.down_proj", "model.layers.19.mlp.experts.57.down_proj", "model.layers.19.mlp.experts.58.down_proj", "model.layers.19.mlp.experts.59.down_proj", "model.layers.19.mlp.experts.60.down_proj", "model.layers.19.mlp.experts.61.down_proj", "model.layers.19.mlp.experts.62.down_proj", "model.layers.19.mlp.experts.63.down_proj", "model.layers.19.mlp.experts.64.down_proj", "model.layers.19.mlp.experts.65.down_proj", "model.layers.19.mlp.experts.66.down_proj", "model.layers.19.mlp.experts.67.down_proj", "model.layers.19.mlp.experts.68.down_proj", "model.layers.19.mlp.experts.69.down_proj", "model.layers.19.mlp.experts.70.down_proj", "model.layers.19.mlp.experts.71.down_proj", "model.layers.19.mlp.experts.72.down_proj", "model.layers.19.mlp.experts.73.down_proj", "model.layers.19.mlp.experts.74.down_proj", "model.layers.19.mlp.experts.75.down_proj", "model.layers.19.mlp.experts.76.down_proj", "model.layers.19.mlp.experts.77.down_proj", "model.layers.19.mlp.experts.78.down_proj", "model.layers.19.mlp.experts.79.down_proj", "model.layers.19.mlp.experts.80.down_proj", "model.layers.19.mlp.experts.81.down_proj", "model.layers.19.mlp.experts.82.down_proj", "model.layers.19.mlp.experts.83.down_proj", "model.layers.19.mlp.experts.84.down_proj", "model.layers.19.mlp.experts.85.down_proj", "model.layers.19.mlp.experts.86.down_proj", "model.layers.19.mlp.experts.87.down_proj", "model.layers.19.mlp.experts.88.down_proj", "model.layers.19.mlp.experts.89.down_proj", "model.layers.19.mlp.experts.90.down_proj", "model.layers.19.mlp.experts.91.down_proj", "model.layers.19.mlp.experts.92.down_proj", "model.layers.19.mlp.experts.93.down_proj", "model.layers.19.mlp.experts.94.down_proj", "model.layers.19.mlp.experts.95.down_proj", "model.layers.19.mlp.experts.96.down_proj", "model.layers.19.mlp.experts.97.down_proj", "model.layers.19.mlp.experts.98.down_proj", "model.layers.19.mlp.experts.99.down_proj", "model.layers.19.mlp.experts.100.down_proj", "model.layers.19.mlp.experts.101.down_proj", "model.layers.19.mlp.experts.102.down_proj", "model.layers.19.mlp.experts.103.down_proj", "model.layers.19.mlp.experts.104.down_proj", "model.layers.19.mlp.experts.105.down_proj", "model.layers.19.mlp.experts.106.down_proj", "model.layers.19.mlp.experts.107.down_proj", "model.layers.19.mlp.experts.108.down_proj", "model.layers.19.mlp.experts.109.down_proj", "model.layers.19.mlp.experts.110.down_proj", "model.layers.19.mlp.experts.111.down_proj", "model.layers.19.mlp.experts.112.down_proj", "model.layers.19.mlp.experts.113.down_proj", "model.layers.19.mlp.experts.114.down_proj", "model.layers.19.mlp.experts.115.down_proj", "model.layers.19.mlp.experts.116.down_proj", "model.layers.19.mlp.experts.117.down_proj", "model.layers.19.mlp.experts.118.down_proj", "model.layers.19.mlp.experts.119.down_proj", "model.layers.19.mlp.experts.120.down_proj", "model.layers.19.mlp.experts.121.down_proj", "model.layers.19.mlp.experts.122.down_proj", "model.layers.19.mlp.experts.123.down_proj", "model.layers.19.mlp.experts.124.down_proj", "model.layers.19.mlp.experts.125.down_proj", "model.layers.19.mlp.experts.126.down_proj", "model.layers.19.mlp.experts.127.down_proj", "model.layers.19.mlp.experts.128.down_proj", "model.layers.19.mlp.experts.129.down_proj", "model.layers.19.mlp.experts.130.down_proj", "model.layers.19.mlp.experts.131.down_proj", "model.layers.19.mlp.experts.132.down_proj", "model.layers.19.mlp.experts.133.down_proj", "model.layers.19.mlp.experts.134.down_proj", "model.layers.19.mlp.experts.135.down_proj", "model.layers.19.mlp.experts.136.down_proj", "model.layers.19.mlp.experts.137.down_proj", "model.layers.19.mlp.experts.138.down_proj", "model.layers.19.mlp.experts.139.down_proj", "model.layers.19.mlp.experts.140.down_proj", "model.layers.19.mlp.experts.141.down_proj", "model.layers.19.mlp.experts.142.down_proj", "model.layers.19.mlp.experts.143.down_proj", "model.layers.19.mlp.experts.144.down_proj", "model.layers.19.mlp.experts.145.down_proj", "model.layers.19.mlp.experts.146.down_proj", "model.layers.19.mlp.experts.147.down_proj", "model.layers.19.mlp.experts.148.down_proj", "model.layers.19.mlp.experts.149.down_proj", "model.layers.19.mlp.experts.150.down_proj", "model.layers.19.mlp.experts.151.down_proj", "model.layers.19.mlp.experts.152.down_proj", "model.layers.19.mlp.experts.153.down_proj", "model.layers.19.mlp.experts.154.down_proj", "model.layers.19.mlp.experts.155.down_proj", "model.layers.19.mlp.experts.156.down_proj", "model.layers.19.mlp.experts.157.down_proj", "model.layers.19.mlp.experts.158.down_proj", "model.layers.19.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0006534045562148205, "dbits": 1258291200 }, { "dkld": -0.0007199697196483695, "dbits": 2516582400 }, { "dkld": -0.000787006784230479, "dbits": 3774873600 }, { "dkld": -0.0008346537128090942, "dbits": 6291456000 } ] }, { "idx": 117, "layers": [ "model.layers.20.self_attn.q_proj" ], "candidates": [ { "dkld": -5.513429641723633e-07, "dbits": 62914560 }, { "dkld": 0.00015732999891042432, "dbits": 125829120 }, { "dkld": 0.0001897482201456968, "dbits": 188743680 }, { "dkld": 0.00018116552382707596, "dbits": 314572800 } ] }, { "idx": 118, "layers": [ "model.layers.20.self_attn.k_proj", "model.layers.20.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00022634472697972974, "dbits": 10485760 }, { "dkld": -0.00037890458479523936, "dbits": 20971520 }, { "dkld": 6.316378712653559e-05, "dbits": 31457280 }, { "dkld": 2.6267580687994174e-05, "dbits": 52428800 } ] }, { "idx": 119, "layers": [ "model.layers.20.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0006834926083683912, "dbits": 62914560 }, { "dkld": -2.2273324429999963e-05, "dbits": 125829120 }, { "dkld": 0.00020058620721101483, "dbits": 188743680 }, { "dkld": 0.0001891091465949929, "dbits": 314572800 } ] }, { "idx": 120, "layers": [ "model.layers.20.mlp.shared_experts.gate_proj", "model.layers.20.mlp.shared_experts.up_proj", "model.layers.20.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0009838653728365954, "dbits": 23592960 }, { "dkld": -0.0015296029858291177, "dbits": 47185920 }, { "dkld": -0.0015638560056686485, "dbits": 70778880 }, { "dkld": -0.0015874405391514412, "dbits": 117964800 } ] }, { "idx": 121, "layers": [ "model.layers.20.mlp.experts.0.gate_proj", "model.layers.20.mlp.experts.1.gate_proj", "model.layers.20.mlp.experts.2.gate_proj", "model.layers.20.mlp.experts.3.gate_proj", "model.layers.20.mlp.experts.4.gate_proj", "model.layers.20.mlp.experts.5.gate_proj", "model.layers.20.mlp.experts.6.gate_proj", "model.layers.20.mlp.experts.7.gate_proj", "model.layers.20.mlp.experts.8.gate_proj", "model.layers.20.mlp.experts.9.gate_proj", "model.layers.20.mlp.experts.10.gate_proj", "model.layers.20.mlp.experts.11.gate_proj", "model.layers.20.mlp.experts.12.gate_proj", "model.layers.20.mlp.experts.13.gate_proj", "model.layers.20.mlp.experts.14.gate_proj", "model.layers.20.mlp.experts.15.gate_proj", "model.layers.20.mlp.experts.16.gate_proj", "model.layers.20.mlp.experts.17.gate_proj", "model.layers.20.mlp.experts.18.gate_proj", "model.layers.20.mlp.experts.19.gate_proj", "model.layers.20.mlp.experts.20.gate_proj", "model.layers.20.mlp.experts.21.gate_proj", "model.layers.20.mlp.experts.22.gate_proj", "model.layers.20.mlp.experts.23.gate_proj", "model.layers.20.mlp.experts.24.gate_proj", "model.layers.20.mlp.experts.25.gate_proj", "model.layers.20.mlp.experts.26.gate_proj", "model.layers.20.mlp.experts.27.gate_proj", "model.layers.20.mlp.experts.28.gate_proj", "model.layers.20.mlp.experts.29.gate_proj", "model.layers.20.mlp.experts.30.gate_proj", "model.layers.20.mlp.experts.31.gate_proj", "model.layers.20.mlp.experts.32.gate_proj", "model.layers.20.mlp.experts.33.gate_proj", "model.layers.20.mlp.experts.34.gate_proj", "model.layers.20.mlp.experts.35.gate_proj", "model.layers.20.mlp.experts.36.gate_proj", "model.layers.20.mlp.experts.37.gate_proj", "model.layers.20.mlp.experts.38.gate_proj", "model.layers.20.mlp.experts.39.gate_proj", "model.layers.20.mlp.experts.40.gate_proj", "model.layers.20.mlp.experts.41.gate_proj", "model.layers.20.mlp.experts.42.gate_proj", "model.layers.20.mlp.experts.43.gate_proj", "model.layers.20.mlp.experts.44.gate_proj", "model.layers.20.mlp.experts.45.gate_proj", "model.layers.20.mlp.experts.46.gate_proj", "model.layers.20.mlp.experts.47.gate_proj", "model.layers.20.mlp.experts.48.gate_proj", "model.layers.20.mlp.experts.49.gate_proj", "model.layers.20.mlp.experts.50.gate_proj", "model.layers.20.mlp.experts.51.gate_proj", "model.layers.20.mlp.experts.52.gate_proj", "model.layers.20.mlp.experts.53.gate_proj", "model.layers.20.mlp.experts.54.gate_proj", "model.layers.20.mlp.experts.55.gate_proj", "model.layers.20.mlp.experts.56.gate_proj", "model.layers.20.mlp.experts.57.gate_proj", "model.layers.20.mlp.experts.58.gate_proj", "model.layers.20.mlp.experts.59.gate_proj", "model.layers.20.mlp.experts.60.gate_proj", "model.layers.20.mlp.experts.61.gate_proj", "model.layers.20.mlp.experts.62.gate_proj", "model.layers.20.mlp.experts.63.gate_proj", "model.layers.20.mlp.experts.64.gate_proj", "model.layers.20.mlp.experts.65.gate_proj", "model.layers.20.mlp.experts.66.gate_proj", "model.layers.20.mlp.experts.67.gate_proj", "model.layers.20.mlp.experts.68.gate_proj", "model.layers.20.mlp.experts.69.gate_proj", "model.layers.20.mlp.experts.70.gate_proj", "model.layers.20.mlp.experts.71.gate_proj", "model.layers.20.mlp.experts.72.gate_proj", "model.layers.20.mlp.experts.73.gate_proj", "model.layers.20.mlp.experts.74.gate_proj", "model.layers.20.mlp.experts.75.gate_proj", "model.layers.20.mlp.experts.76.gate_proj", "model.layers.20.mlp.experts.77.gate_proj", "model.layers.20.mlp.experts.78.gate_proj", "model.layers.20.mlp.experts.79.gate_proj", "model.layers.20.mlp.experts.80.gate_proj", "model.layers.20.mlp.experts.81.gate_proj", "model.layers.20.mlp.experts.82.gate_proj", "model.layers.20.mlp.experts.83.gate_proj", "model.layers.20.mlp.experts.84.gate_proj", "model.layers.20.mlp.experts.85.gate_proj", "model.layers.20.mlp.experts.86.gate_proj", "model.layers.20.mlp.experts.87.gate_proj", "model.layers.20.mlp.experts.88.gate_proj", "model.layers.20.mlp.experts.89.gate_proj", "model.layers.20.mlp.experts.90.gate_proj", "model.layers.20.mlp.experts.91.gate_proj", "model.layers.20.mlp.experts.92.gate_proj", "model.layers.20.mlp.experts.93.gate_proj", "model.layers.20.mlp.experts.94.gate_proj", "model.layers.20.mlp.experts.95.gate_proj", "model.layers.20.mlp.experts.96.gate_proj", "model.layers.20.mlp.experts.97.gate_proj", "model.layers.20.mlp.experts.98.gate_proj", "model.layers.20.mlp.experts.99.gate_proj", "model.layers.20.mlp.experts.100.gate_proj", "model.layers.20.mlp.experts.101.gate_proj", "model.layers.20.mlp.experts.102.gate_proj", "model.layers.20.mlp.experts.103.gate_proj", "model.layers.20.mlp.experts.104.gate_proj", "model.layers.20.mlp.experts.105.gate_proj", "model.layers.20.mlp.experts.106.gate_proj", "model.layers.20.mlp.experts.107.gate_proj", "model.layers.20.mlp.experts.108.gate_proj", "model.layers.20.mlp.experts.109.gate_proj", "model.layers.20.mlp.experts.110.gate_proj", "model.layers.20.mlp.experts.111.gate_proj", "model.layers.20.mlp.experts.112.gate_proj", "model.layers.20.mlp.experts.113.gate_proj", "model.layers.20.mlp.experts.114.gate_proj", "model.layers.20.mlp.experts.115.gate_proj", "model.layers.20.mlp.experts.116.gate_proj", "model.layers.20.mlp.experts.117.gate_proj", "model.layers.20.mlp.experts.118.gate_proj", "model.layers.20.mlp.experts.119.gate_proj", "model.layers.20.mlp.experts.120.gate_proj", "model.layers.20.mlp.experts.121.gate_proj", "model.layers.20.mlp.experts.122.gate_proj", "model.layers.20.mlp.experts.123.gate_proj", "model.layers.20.mlp.experts.124.gate_proj", "model.layers.20.mlp.experts.125.gate_proj", "model.layers.20.mlp.experts.126.gate_proj", "model.layers.20.mlp.experts.127.gate_proj", "model.layers.20.mlp.experts.128.gate_proj", "model.layers.20.mlp.experts.129.gate_proj", "model.layers.20.mlp.experts.130.gate_proj", "model.layers.20.mlp.experts.131.gate_proj", "model.layers.20.mlp.experts.132.gate_proj", "model.layers.20.mlp.experts.133.gate_proj", "model.layers.20.mlp.experts.134.gate_proj", "model.layers.20.mlp.experts.135.gate_proj", "model.layers.20.mlp.experts.136.gate_proj", "model.layers.20.mlp.experts.137.gate_proj", "model.layers.20.mlp.experts.138.gate_proj", "model.layers.20.mlp.experts.139.gate_proj", "model.layers.20.mlp.experts.140.gate_proj", "model.layers.20.mlp.experts.141.gate_proj", "model.layers.20.mlp.experts.142.gate_proj", "model.layers.20.mlp.experts.143.gate_proj", "model.layers.20.mlp.experts.144.gate_proj", "model.layers.20.mlp.experts.145.gate_proj", "model.layers.20.mlp.experts.146.gate_proj", "model.layers.20.mlp.experts.147.gate_proj", "model.layers.20.mlp.experts.148.gate_proj", "model.layers.20.mlp.experts.149.gate_proj", "model.layers.20.mlp.experts.150.gate_proj", "model.layers.20.mlp.experts.151.gate_proj", "model.layers.20.mlp.experts.152.gate_proj", "model.layers.20.mlp.experts.153.gate_proj", "model.layers.20.mlp.experts.154.gate_proj", "model.layers.20.mlp.experts.155.gate_proj", "model.layers.20.mlp.experts.156.gate_proj", "model.layers.20.mlp.experts.157.gate_proj", "model.layers.20.mlp.experts.158.gate_proj", "model.layers.20.mlp.experts.159.gate_proj", "model.layers.20.mlp.experts.0.up_proj", "model.layers.20.mlp.experts.1.up_proj", "model.layers.20.mlp.experts.2.up_proj", "model.layers.20.mlp.experts.3.up_proj", "model.layers.20.mlp.experts.4.up_proj", "model.layers.20.mlp.experts.5.up_proj", "model.layers.20.mlp.experts.6.up_proj", "model.layers.20.mlp.experts.7.up_proj", "model.layers.20.mlp.experts.8.up_proj", "model.layers.20.mlp.experts.9.up_proj", "model.layers.20.mlp.experts.10.up_proj", "model.layers.20.mlp.experts.11.up_proj", "model.layers.20.mlp.experts.12.up_proj", "model.layers.20.mlp.experts.13.up_proj", "model.layers.20.mlp.experts.14.up_proj", "model.layers.20.mlp.experts.15.up_proj", "model.layers.20.mlp.experts.16.up_proj", "model.layers.20.mlp.experts.17.up_proj", "model.layers.20.mlp.experts.18.up_proj", "model.layers.20.mlp.experts.19.up_proj", "model.layers.20.mlp.experts.20.up_proj", "model.layers.20.mlp.experts.21.up_proj", "model.layers.20.mlp.experts.22.up_proj", "model.layers.20.mlp.experts.23.up_proj", "model.layers.20.mlp.experts.24.up_proj", "model.layers.20.mlp.experts.25.up_proj", "model.layers.20.mlp.experts.26.up_proj", "model.layers.20.mlp.experts.27.up_proj", "model.layers.20.mlp.experts.28.up_proj", "model.layers.20.mlp.experts.29.up_proj", "model.layers.20.mlp.experts.30.up_proj", "model.layers.20.mlp.experts.31.up_proj", "model.layers.20.mlp.experts.32.up_proj", "model.layers.20.mlp.experts.33.up_proj", "model.layers.20.mlp.experts.34.up_proj", "model.layers.20.mlp.experts.35.up_proj", "model.layers.20.mlp.experts.36.up_proj", "model.layers.20.mlp.experts.37.up_proj", "model.layers.20.mlp.experts.38.up_proj", "model.layers.20.mlp.experts.39.up_proj", "model.layers.20.mlp.experts.40.up_proj", "model.layers.20.mlp.experts.41.up_proj", "model.layers.20.mlp.experts.42.up_proj", "model.layers.20.mlp.experts.43.up_proj", "model.layers.20.mlp.experts.44.up_proj", "model.layers.20.mlp.experts.45.up_proj", "model.layers.20.mlp.experts.46.up_proj", "model.layers.20.mlp.experts.47.up_proj", "model.layers.20.mlp.experts.48.up_proj", "model.layers.20.mlp.experts.49.up_proj", "model.layers.20.mlp.experts.50.up_proj", "model.layers.20.mlp.experts.51.up_proj", "model.layers.20.mlp.experts.52.up_proj", "model.layers.20.mlp.experts.53.up_proj", "model.layers.20.mlp.experts.54.up_proj", "model.layers.20.mlp.experts.55.up_proj", "model.layers.20.mlp.experts.56.up_proj", "model.layers.20.mlp.experts.57.up_proj", "model.layers.20.mlp.experts.58.up_proj", "model.layers.20.mlp.experts.59.up_proj", "model.layers.20.mlp.experts.60.up_proj", "model.layers.20.mlp.experts.61.up_proj", "model.layers.20.mlp.experts.62.up_proj", "model.layers.20.mlp.experts.63.up_proj", "model.layers.20.mlp.experts.64.up_proj", "model.layers.20.mlp.experts.65.up_proj", "model.layers.20.mlp.experts.66.up_proj", "model.layers.20.mlp.experts.67.up_proj", "model.layers.20.mlp.experts.68.up_proj", "model.layers.20.mlp.experts.69.up_proj", "model.layers.20.mlp.experts.70.up_proj", "model.layers.20.mlp.experts.71.up_proj", "model.layers.20.mlp.experts.72.up_proj", "model.layers.20.mlp.experts.73.up_proj", "model.layers.20.mlp.experts.74.up_proj", "model.layers.20.mlp.experts.75.up_proj", "model.layers.20.mlp.experts.76.up_proj", "model.layers.20.mlp.experts.77.up_proj", "model.layers.20.mlp.experts.78.up_proj", "model.layers.20.mlp.experts.79.up_proj", "model.layers.20.mlp.experts.80.up_proj", "model.layers.20.mlp.experts.81.up_proj", "model.layers.20.mlp.experts.82.up_proj", "model.layers.20.mlp.experts.83.up_proj", "model.layers.20.mlp.experts.84.up_proj", "model.layers.20.mlp.experts.85.up_proj", "model.layers.20.mlp.experts.86.up_proj", "model.layers.20.mlp.experts.87.up_proj", "model.layers.20.mlp.experts.88.up_proj", "model.layers.20.mlp.experts.89.up_proj", "model.layers.20.mlp.experts.90.up_proj", "model.layers.20.mlp.experts.91.up_proj", "model.layers.20.mlp.experts.92.up_proj", "model.layers.20.mlp.experts.93.up_proj", "model.layers.20.mlp.experts.94.up_proj", "model.layers.20.mlp.experts.95.up_proj", "model.layers.20.mlp.experts.96.up_proj", "model.layers.20.mlp.experts.97.up_proj", "model.layers.20.mlp.experts.98.up_proj", "model.layers.20.mlp.experts.99.up_proj", "model.layers.20.mlp.experts.100.up_proj", "model.layers.20.mlp.experts.101.up_proj", "model.layers.20.mlp.experts.102.up_proj", "model.layers.20.mlp.experts.103.up_proj", "model.layers.20.mlp.experts.104.up_proj", "model.layers.20.mlp.experts.105.up_proj", "model.layers.20.mlp.experts.106.up_proj", "model.layers.20.mlp.experts.107.up_proj", "model.layers.20.mlp.experts.108.up_proj", "model.layers.20.mlp.experts.109.up_proj", "model.layers.20.mlp.experts.110.up_proj", "model.layers.20.mlp.experts.111.up_proj", "model.layers.20.mlp.experts.112.up_proj", "model.layers.20.mlp.experts.113.up_proj", "model.layers.20.mlp.experts.114.up_proj", "model.layers.20.mlp.experts.115.up_proj", "model.layers.20.mlp.experts.116.up_proj", "model.layers.20.mlp.experts.117.up_proj", "model.layers.20.mlp.experts.118.up_proj", "model.layers.20.mlp.experts.119.up_proj", "model.layers.20.mlp.experts.120.up_proj", "model.layers.20.mlp.experts.121.up_proj", "model.layers.20.mlp.experts.122.up_proj", "model.layers.20.mlp.experts.123.up_proj", "model.layers.20.mlp.experts.124.up_proj", "model.layers.20.mlp.experts.125.up_proj", "model.layers.20.mlp.experts.126.up_proj", "model.layers.20.mlp.experts.127.up_proj", "model.layers.20.mlp.experts.128.up_proj", "model.layers.20.mlp.experts.129.up_proj", "model.layers.20.mlp.experts.130.up_proj", "model.layers.20.mlp.experts.131.up_proj", "model.layers.20.mlp.experts.132.up_proj", "model.layers.20.mlp.experts.133.up_proj", "model.layers.20.mlp.experts.134.up_proj", "model.layers.20.mlp.experts.135.up_proj", "model.layers.20.mlp.experts.136.up_proj", "model.layers.20.mlp.experts.137.up_proj", "model.layers.20.mlp.experts.138.up_proj", "model.layers.20.mlp.experts.139.up_proj", "model.layers.20.mlp.experts.140.up_proj", "model.layers.20.mlp.experts.141.up_proj", "model.layers.20.mlp.experts.142.up_proj", "model.layers.20.mlp.experts.143.up_proj", "model.layers.20.mlp.experts.144.up_proj", "model.layers.20.mlp.experts.145.up_proj", "model.layers.20.mlp.experts.146.up_proj", "model.layers.20.mlp.experts.147.up_proj", "model.layers.20.mlp.experts.148.up_proj", "model.layers.20.mlp.experts.149.up_proj", "model.layers.20.mlp.experts.150.up_proj", "model.layers.20.mlp.experts.151.up_proj", "model.layers.20.mlp.experts.152.up_proj", "model.layers.20.mlp.experts.153.up_proj", "model.layers.20.mlp.experts.154.up_proj", "model.layers.20.mlp.experts.155.up_proj", "model.layers.20.mlp.experts.156.up_proj", "model.layers.20.mlp.experts.157.up_proj", "model.layers.20.mlp.experts.158.up_proj", "model.layers.20.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0007003111764788683, "dbits": 2516582400 }, { "dkld": -0.0008550105616450393, "dbits": 5033164800 }, { "dkld": -0.0009101872332394206, "dbits": 7549747200 }, { "dkld": -0.0009545358829200351, "dbits": 12582912000 } ] }, { "idx": 122, "layers": [ "model.layers.20.mlp.experts.0.down_proj", "model.layers.20.mlp.experts.1.down_proj", "model.layers.20.mlp.experts.2.down_proj", "model.layers.20.mlp.experts.3.down_proj", "model.layers.20.mlp.experts.4.down_proj", "model.layers.20.mlp.experts.5.down_proj", "model.layers.20.mlp.experts.6.down_proj", "model.layers.20.mlp.experts.7.down_proj", "model.layers.20.mlp.experts.8.down_proj", "model.layers.20.mlp.experts.9.down_proj", "model.layers.20.mlp.experts.10.down_proj", "model.layers.20.mlp.experts.11.down_proj", "model.layers.20.mlp.experts.12.down_proj", "model.layers.20.mlp.experts.13.down_proj", "model.layers.20.mlp.experts.14.down_proj", "model.layers.20.mlp.experts.15.down_proj", "model.layers.20.mlp.experts.16.down_proj", "model.layers.20.mlp.experts.17.down_proj", "model.layers.20.mlp.experts.18.down_proj", "model.layers.20.mlp.experts.19.down_proj", "model.layers.20.mlp.experts.20.down_proj", "model.layers.20.mlp.experts.21.down_proj", "model.layers.20.mlp.experts.22.down_proj", "model.layers.20.mlp.experts.23.down_proj", "model.layers.20.mlp.experts.24.down_proj", "model.layers.20.mlp.experts.25.down_proj", "model.layers.20.mlp.experts.26.down_proj", "model.layers.20.mlp.experts.27.down_proj", "model.layers.20.mlp.experts.28.down_proj", "model.layers.20.mlp.experts.29.down_proj", "model.layers.20.mlp.experts.30.down_proj", "model.layers.20.mlp.experts.31.down_proj", "model.layers.20.mlp.experts.32.down_proj", "model.layers.20.mlp.experts.33.down_proj", "model.layers.20.mlp.experts.34.down_proj", "model.layers.20.mlp.experts.35.down_proj", "model.layers.20.mlp.experts.36.down_proj", "model.layers.20.mlp.experts.37.down_proj", "model.layers.20.mlp.experts.38.down_proj", "model.layers.20.mlp.experts.39.down_proj", "model.layers.20.mlp.experts.40.down_proj", "model.layers.20.mlp.experts.41.down_proj", "model.layers.20.mlp.experts.42.down_proj", "model.layers.20.mlp.experts.43.down_proj", "model.layers.20.mlp.experts.44.down_proj", "model.layers.20.mlp.experts.45.down_proj", "model.layers.20.mlp.experts.46.down_proj", "model.layers.20.mlp.experts.47.down_proj", "model.layers.20.mlp.experts.48.down_proj", "model.layers.20.mlp.experts.49.down_proj", "model.layers.20.mlp.experts.50.down_proj", "model.layers.20.mlp.experts.51.down_proj", "model.layers.20.mlp.experts.52.down_proj", "model.layers.20.mlp.experts.53.down_proj", "model.layers.20.mlp.experts.54.down_proj", "model.layers.20.mlp.experts.55.down_proj", "model.layers.20.mlp.experts.56.down_proj", "model.layers.20.mlp.experts.57.down_proj", "model.layers.20.mlp.experts.58.down_proj", "model.layers.20.mlp.experts.59.down_proj", "model.layers.20.mlp.experts.60.down_proj", "model.layers.20.mlp.experts.61.down_proj", "model.layers.20.mlp.experts.62.down_proj", "model.layers.20.mlp.experts.63.down_proj", "model.layers.20.mlp.experts.64.down_proj", "model.layers.20.mlp.experts.65.down_proj", "model.layers.20.mlp.experts.66.down_proj", "model.layers.20.mlp.experts.67.down_proj", "model.layers.20.mlp.experts.68.down_proj", "model.layers.20.mlp.experts.69.down_proj", "model.layers.20.mlp.experts.70.down_proj", "model.layers.20.mlp.experts.71.down_proj", "model.layers.20.mlp.experts.72.down_proj", "model.layers.20.mlp.experts.73.down_proj", "model.layers.20.mlp.experts.74.down_proj", "model.layers.20.mlp.experts.75.down_proj", "model.layers.20.mlp.experts.76.down_proj", "model.layers.20.mlp.experts.77.down_proj", "model.layers.20.mlp.experts.78.down_proj", "model.layers.20.mlp.experts.79.down_proj", "model.layers.20.mlp.experts.80.down_proj", "model.layers.20.mlp.experts.81.down_proj", "model.layers.20.mlp.experts.82.down_proj", "model.layers.20.mlp.experts.83.down_proj", "model.layers.20.mlp.experts.84.down_proj", "model.layers.20.mlp.experts.85.down_proj", "model.layers.20.mlp.experts.86.down_proj", "model.layers.20.mlp.experts.87.down_proj", "model.layers.20.mlp.experts.88.down_proj", "model.layers.20.mlp.experts.89.down_proj", "model.layers.20.mlp.experts.90.down_proj", "model.layers.20.mlp.experts.91.down_proj", "model.layers.20.mlp.experts.92.down_proj", "model.layers.20.mlp.experts.93.down_proj", "model.layers.20.mlp.experts.94.down_proj", "model.layers.20.mlp.experts.95.down_proj", "model.layers.20.mlp.experts.96.down_proj", "model.layers.20.mlp.experts.97.down_proj", "model.layers.20.mlp.experts.98.down_proj", "model.layers.20.mlp.experts.99.down_proj", "model.layers.20.mlp.experts.100.down_proj", "model.layers.20.mlp.experts.101.down_proj", "model.layers.20.mlp.experts.102.down_proj", "model.layers.20.mlp.experts.103.down_proj", "model.layers.20.mlp.experts.104.down_proj", "model.layers.20.mlp.experts.105.down_proj", "model.layers.20.mlp.experts.106.down_proj", "model.layers.20.mlp.experts.107.down_proj", "model.layers.20.mlp.experts.108.down_proj", "model.layers.20.mlp.experts.109.down_proj", "model.layers.20.mlp.experts.110.down_proj", "model.layers.20.mlp.experts.111.down_proj", "model.layers.20.mlp.experts.112.down_proj", "model.layers.20.mlp.experts.113.down_proj", "model.layers.20.mlp.experts.114.down_proj", "model.layers.20.mlp.experts.115.down_proj", "model.layers.20.mlp.experts.116.down_proj", "model.layers.20.mlp.experts.117.down_proj", "model.layers.20.mlp.experts.118.down_proj", "model.layers.20.mlp.experts.119.down_proj", "model.layers.20.mlp.experts.120.down_proj", "model.layers.20.mlp.experts.121.down_proj", "model.layers.20.mlp.experts.122.down_proj", "model.layers.20.mlp.experts.123.down_proj", "model.layers.20.mlp.experts.124.down_proj", "model.layers.20.mlp.experts.125.down_proj", "model.layers.20.mlp.experts.126.down_proj", "model.layers.20.mlp.experts.127.down_proj", "model.layers.20.mlp.experts.128.down_proj", "model.layers.20.mlp.experts.129.down_proj", "model.layers.20.mlp.experts.130.down_proj", "model.layers.20.mlp.experts.131.down_proj", "model.layers.20.mlp.experts.132.down_proj", "model.layers.20.mlp.experts.133.down_proj", "model.layers.20.mlp.experts.134.down_proj", "model.layers.20.mlp.experts.135.down_proj", "model.layers.20.mlp.experts.136.down_proj", "model.layers.20.mlp.experts.137.down_proj", "model.layers.20.mlp.experts.138.down_proj", "model.layers.20.mlp.experts.139.down_proj", "model.layers.20.mlp.experts.140.down_proj", "model.layers.20.mlp.experts.141.down_proj", "model.layers.20.mlp.experts.142.down_proj", "model.layers.20.mlp.experts.143.down_proj", "model.layers.20.mlp.experts.144.down_proj", "model.layers.20.mlp.experts.145.down_proj", "model.layers.20.mlp.experts.146.down_proj", "model.layers.20.mlp.experts.147.down_proj", "model.layers.20.mlp.experts.148.down_proj", "model.layers.20.mlp.experts.149.down_proj", "model.layers.20.mlp.experts.150.down_proj", "model.layers.20.mlp.experts.151.down_proj", "model.layers.20.mlp.experts.152.down_proj", "model.layers.20.mlp.experts.153.down_proj", "model.layers.20.mlp.experts.154.down_proj", "model.layers.20.mlp.experts.155.down_proj", "model.layers.20.mlp.experts.156.down_proj", "model.layers.20.mlp.experts.157.down_proj", "model.layers.20.mlp.experts.158.down_proj", "model.layers.20.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0008531026542186737, "dbits": 1258291200 }, { "dkld": -0.0008382737636566245, "dbits": 2516582400 }, { "dkld": -0.0008313516154885292, "dbits": 3774873600 }, { "dkld": -0.0008749213069677436, "dbits": 6291456000 } ] }, { "idx": 123, "layers": [ "model.layers.21.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002686345949769048, "dbits": 62914560 }, { "dkld": -0.0002536002546548871, "dbits": 125829120 }, { "dkld": -0.00021363086998463232, "dbits": 188743680 }, { "dkld": -0.0001218779943883419, "dbits": 314572800 } ] }, { "idx": 124, "layers": [ "model.layers.21.self_attn.k_proj", "model.layers.21.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0002593880519270869, "dbits": 10485760 }, { "dkld": 0.0002733279019594137, "dbits": 20971520 }, { "dkld": 7.522050291298987e-05, "dbits": 31457280 }, { "dkld": 0.00019479440525173308, "dbits": 52428800 } ] }, { "idx": 125, "layers": [ "model.layers.21.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002245923504233388, "dbits": 62914560 }, { "dkld": 0.0002568878233432659, "dbits": 125829120 }, { "dkld": 0.00025049252435564717, "dbits": 188743680 }, { "dkld": 0.0002948084846138843, "dbits": 314572800 } ] }, { "idx": 126, "layers": [ "model.layers.21.mlp.shared_experts.gate_proj", "model.layers.21.mlp.shared_experts.up_proj", "model.layers.21.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00011117532849312661, "dbits": 23592960 }, { "dkld": 0.0007168337702751076, "dbits": 47185920 }, { "dkld": 0.0004775899462401839, "dbits": 70778880 }, { "dkld": 0.0003984379582107067, "dbits": 117964800 } ] }, { "idx": 127, "layers": [ "model.layers.21.mlp.experts.0.gate_proj", "model.layers.21.mlp.experts.1.gate_proj", "model.layers.21.mlp.experts.2.gate_proj", "model.layers.21.mlp.experts.3.gate_proj", "model.layers.21.mlp.experts.4.gate_proj", "model.layers.21.mlp.experts.5.gate_proj", "model.layers.21.mlp.experts.6.gate_proj", "model.layers.21.mlp.experts.7.gate_proj", "model.layers.21.mlp.experts.8.gate_proj", "model.layers.21.mlp.experts.9.gate_proj", "model.layers.21.mlp.experts.10.gate_proj", "model.layers.21.mlp.experts.11.gate_proj", "model.layers.21.mlp.experts.12.gate_proj", "model.layers.21.mlp.experts.13.gate_proj", "model.layers.21.mlp.experts.14.gate_proj", "model.layers.21.mlp.experts.15.gate_proj", "model.layers.21.mlp.experts.16.gate_proj", "model.layers.21.mlp.experts.17.gate_proj", "model.layers.21.mlp.experts.18.gate_proj", "model.layers.21.mlp.experts.19.gate_proj", "model.layers.21.mlp.experts.20.gate_proj", "model.layers.21.mlp.experts.21.gate_proj", "model.layers.21.mlp.experts.22.gate_proj", "model.layers.21.mlp.experts.23.gate_proj", "model.layers.21.mlp.experts.24.gate_proj", "model.layers.21.mlp.experts.25.gate_proj", "model.layers.21.mlp.experts.26.gate_proj", "model.layers.21.mlp.experts.27.gate_proj", "model.layers.21.mlp.experts.28.gate_proj", "model.layers.21.mlp.experts.29.gate_proj", "model.layers.21.mlp.experts.30.gate_proj", "model.layers.21.mlp.experts.31.gate_proj", "model.layers.21.mlp.experts.32.gate_proj", "model.layers.21.mlp.experts.33.gate_proj", "model.layers.21.mlp.experts.34.gate_proj", "model.layers.21.mlp.experts.35.gate_proj", "model.layers.21.mlp.experts.36.gate_proj", "model.layers.21.mlp.experts.37.gate_proj", "model.layers.21.mlp.experts.38.gate_proj", "model.layers.21.mlp.experts.39.gate_proj", "model.layers.21.mlp.experts.40.gate_proj", "model.layers.21.mlp.experts.41.gate_proj", "model.layers.21.mlp.experts.42.gate_proj", "model.layers.21.mlp.experts.43.gate_proj", "model.layers.21.mlp.experts.44.gate_proj", "model.layers.21.mlp.experts.45.gate_proj", "model.layers.21.mlp.experts.46.gate_proj", "model.layers.21.mlp.experts.47.gate_proj", "model.layers.21.mlp.experts.48.gate_proj", "model.layers.21.mlp.experts.49.gate_proj", "model.layers.21.mlp.experts.50.gate_proj", "model.layers.21.mlp.experts.51.gate_proj", "model.layers.21.mlp.experts.52.gate_proj", "model.layers.21.mlp.experts.53.gate_proj", "model.layers.21.mlp.experts.54.gate_proj", "model.layers.21.mlp.experts.55.gate_proj", "model.layers.21.mlp.experts.56.gate_proj", "model.layers.21.mlp.experts.57.gate_proj", "model.layers.21.mlp.experts.58.gate_proj", "model.layers.21.mlp.experts.59.gate_proj", "model.layers.21.mlp.experts.60.gate_proj", "model.layers.21.mlp.experts.61.gate_proj", "model.layers.21.mlp.experts.62.gate_proj", "model.layers.21.mlp.experts.63.gate_proj", "model.layers.21.mlp.experts.64.gate_proj", "model.layers.21.mlp.experts.65.gate_proj", "model.layers.21.mlp.experts.66.gate_proj", "model.layers.21.mlp.experts.67.gate_proj", "model.layers.21.mlp.experts.68.gate_proj", "model.layers.21.mlp.experts.69.gate_proj", "model.layers.21.mlp.experts.70.gate_proj", "model.layers.21.mlp.experts.71.gate_proj", "model.layers.21.mlp.experts.72.gate_proj", "model.layers.21.mlp.experts.73.gate_proj", "model.layers.21.mlp.experts.74.gate_proj", "model.layers.21.mlp.experts.75.gate_proj", "model.layers.21.mlp.experts.76.gate_proj", "model.layers.21.mlp.experts.77.gate_proj", "model.layers.21.mlp.experts.78.gate_proj", "model.layers.21.mlp.experts.79.gate_proj", "model.layers.21.mlp.experts.80.gate_proj", "model.layers.21.mlp.experts.81.gate_proj", "model.layers.21.mlp.experts.82.gate_proj", "model.layers.21.mlp.experts.83.gate_proj", "model.layers.21.mlp.experts.84.gate_proj", "model.layers.21.mlp.experts.85.gate_proj", "model.layers.21.mlp.experts.86.gate_proj", "model.layers.21.mlp.experts.87.gate_proj", "model.layers.21.mlp.experts.88.gate_proj", "model.layers.21.mlp.experts.89.gate_proj", "model.layers.21.mlp.experts.90.gate_proj", "model.layers.21.mlp.experts.91.gate_proj", "model.layers.21.mlp.experts.92.gate_proj", "model.layers.21.mlp.experts.93.gate_proj", "model.layers.21.mlp.experts.94.gate_proj", "model.layers.21.mlp.experts.95.gate_proj", "model.layers.21.mlp.experts.96.gate_proj", "model.layers.21.mlp.experts.97.gate_proj", "model.layers.21.mlp.experts.98.gate_proj", "model.layers.21.mlp.experts.99.gate_proj", "model.layers.21.mlp.experts.100.gate_proj", "model.layers.21.mlp.experts.101.gate_proj", "model.layers.21.mlp.experts.102.gate_proj", "model.layers.21.mlp.experts.103.gate_proj", "model.layers.21.mlp.experts.104.gate_proj", "model.layers.21.mlp.experts.105.gate_proj", "model.layers.21.mlp.experts.106.gate_proj", "model.layers.21.mlp.experts.107.gate_proj", "model.layers.21.mlp.experts.108.gate_proj", "model.layers.21.mlp.experts.109.gate_proj", "model.layers.21.mlp.experts.110.gate_proj", "model.layers.21.mlp.experts.111.gate_proj", "model.layers.21.mlp.experts.112.gate_proj", "model.layers.21.mlp.experts.113.gate_proj", "model.layers.21.mlp.experts.114.gate_proj", "model.layers.21.mlp.experts.115.gate_proj", "model.layers.21.mlp.experts.116.gate_proj", "model.layers.21.mlp.experts.117.gate_proj", "model.layers.21.mlp.experts.118.gate_proj", "model.layers.21.mlp.experts.119.gate_proj", "model.layers.21.mlp.experts.120.gate_proj", "model.layers.21.mlp.experts.121.gate_proj", "model.layers.21.mlp.experts.122.gate_proj", "model.layers.21.mlp.experts.123.gate_proj", "model.layers.21.mlp.experts.124.gate_proj", "model.layers.21.mlp.experts.125.gate_proj", "model.layers.21.mlp.experts.126.gate_proj", "model.layers.21.mlp.experts.127.gate_proj", "model.layers.21.mlp.experts.128.gate_proj", "model.layers.21.mlp.experts.129.gate_proj", "model.layers.21.mlp.experts.130.gate_proj", "model.layers.21.mlp.experts.131.gate_proj", "model.layers.21.mlp.experts.132.gate_proj", "model.layers.21.mlp.experts.133.gate_proj", "model.layers.21.mlp.experts.134.gate_proj", "model.layers.21.mlp.experts.135.gate_proj", "model.layers.21.mlp.experts.136.gate_proj", "model.layers.21.mlp.experts.137.gate_proj", "model.layers.21.mlp.experts.138.gate_proj", "model.layers.21.mlp.experts.139.gate_proj", "model.layers.21.mlp.experts.140.gate_proj", "model.layers.21.mlp.experts.141.gate_proj", "model.layers.21.mlp.experts.142.gate_proj", "model.layers.21.mlp.experts.143.gate_proj", "model.layers.21.mlp.experts.144.gate_proj", "model.layers.21.mlp.experts.145.gate_proj", "model.layers.21.mlp.experts.146.gate_proj", "model.layers.21.mlp.experts.147.gate_proj", "model.layers.21.mlp.experts.148.gate_proj", "model.layers.21.mlp.experts.149.gate_proj", "model.layers.21.mlp.experts.150.gate_proj", "model.layers.21.mlp.experts.151.gate_proj", "model.layers.21.mlp.experts.152.gate_proj", "model.layers.21.mlp.experts.153.gate_proj", "model.layers.21.mlp.experts.154.gate_proj", "model.layers.21.mlp.experts.155.gate_proj", "model.layers.21.mlp.experts.156.gate_proj", "model.layers.21.mlp.experts.157.gate_proj", "model.layers.21.mlp.experts.158.gate_proj", "model.layers.21.mlp.experts.159.gate_proj", "model.layers.21.mlp.experts.0.up_proj", "model.layers.21.mlp.experts.1.up_proj", "model.layers.21.mlp.experts.2.up_proj", "model.layers.21.mlp.experts.3.up_proj", "model.layers.21.mlp.experts.4.up_proj", "model.layers.21.mlp.experts.5.up_proj", "model.layers.21.mlp.experts.6.up_proj", "model.layers.21.mlp.experts.7.up_proj", "model.layers.21.mlp.experts.8.up_proj", "model.layers.21.mlp.experts.9.up_proj", "model.layers.21.mlp.experts.10.up_proj", "model.layers.21.mlp.experts.11.up_proj", "model.layers.21.mlp.experts.12.up_proj", "model.layers.21.mlp.experts.13.up_proj", "model.layers.21.mlp.experts.14.up_proj", "model.layers.21.mlp.experts.15.up_proj", "model.layers.21.mlp.experts.16.up_proj", "model.layers.21.mlp.experts.17.up_proj", "model.layers.21.mlp.experts.18.up_proj", "model.layers.21.mlp.experts.19.up_proj", "model.layers.21.mlp.experts.20.up_proj", "model.layers.21.mlp.experts.21.up_proj", "model.layers.21.mlp.experts.22.up_proj", "model.layers.21.mlp.experts.23.up_proj", "model.layers.21.mlp.experts.24.up_proj", "model.layers.21.mlp.experts.25.up_proj", "model.layers.21.mlp.experts.26.up_proj", "model.layers.21.mlp.experts.27.up_proj", "model.layers.21.mlp.experts.28.up_proj", "model.layers.21.mlp.experts.29.up_proj", "model.layers.21.mlp.experts.30.up_proj", "model.layers.21.mlp.experts.31.up_proj", "model.layers.21.mlp.experts.32.up_proj", "model.layers.21.mlp.experts.33.up_proj", "model.layers.21.mlp.experts.34.up_proj", "model.layers.21.mlp.experts.35.up_proj", "model.layers.21.mlp.experts.36.up_proj", "model.layers.21.mlp.experts.37.up_proj", "model.layers.21.mlp.experts.38.up_proj", "model.layers.21.mlp.experts.39.up_proj", "model.layers.21.mlp.experts.40.up_proj", "model.layers.21.mlp.experts.41.up_proj", "model.layers.21.mlp.experts.42.up_proj", "model.layers.21.mlp.experts.43.up_proj", "model.layers.21.mlp.experts.44.up_proj", "model.layers.21.mlp.experts.45.up_proj", "model.layers.21.mlp.experts.46.up_proj", "model.layers.21.mlp.experts.47.up_proj", "model.layers.21.mlp.experts.48.up_proj", "model.layers.21.mlp.experts.49.up_proj", "model.layers.21.mlp.experts.50.up_proj", "model.layers.21.mlp.experts.51.up_proj", "model.layers.21.mlp.experts.52.up_proj", "model.layers.21.mlp.experts.53.up_proj", "model.layers.21.mlp.experts.54.up_proj", "model.layers.21.mlp.experts.55.up_proj", "model.layers.21.mlp.experts.56.up_proj", "model.layers.21.mlp.experts.57.up_proj", "model.layers.21.mlp.experts.58.up_proj", "model.layers.21.mlp.experts.59.up_proj", "model.layers.21.mlp.experts.60.up_proj", "model.layers.21.mlp.experts.61.up_proj", "model.layers.21.mlp.experts.62.up_proj", "model.layers.21.mlp.experts.63.up_proj", "model.layers.21.mlp.experts.64.up_proj", "model.layers.21.mlp.experts.65.up_proj", "model.layers.21.mlp.experts.66.up_proj", "model.layers.21.mlp.experts.67.up_proj", "model.layers.21.mlp.experts.68.up_proj", "model.layers.21.mlp.experts.69.up_proj", "model.layers.21.mlp.experts.70.up_proj", "model.layers.21.mlp.experts.71.up_proj", "model.layers.21.mlp.experts.72.up_proj", "model.layers.21.mlp.experts.73.up_proj", "model.layers.21.mlp.experts.74.up_proj", "model.layers.21.mlp.experts.75.up_proj", "model.layers.21.mlp.experts.76.up_proj", "model.layers.21.mlp.experts.77.up_proj", "model.layers.21.mlp.experts.78.up_proj", "model.layers.21.mlp.experts.79.up_proj", "model.layers.21.mlp.experts.80.up_proj", "model.layers.21.mlp.experts.81.up_proj", "model.layers.21.mlp.experts.82.up_proj", "model.layers.21.mlp.experts.83.up_proj", "model.layers.21.mlp.experts.84.up_proj", "model.layers.21.mlp.experts.85.up_proj", "model.layers.21.mlp.experts.86.up_proj", "model.layers.21.mlp.experts.87.up_proj", "model.layers.21.mlp.experts.88.up_proj", "model.layers.21.mlp.experts.89.up_proj", "model.layers.21.mlp.experts.90.up_proj", "model.layers.21.mlp.experts.91.up_proj", "model.layers.21.mlp.experts.92.up_proj", "model.layers.21.mlp.experts.93.up_proj", "model.layers.21.mlp.experts.94.up_proj", "model.layers.21.mlp.experts.95.up_proj", "model.layers.21.mlp.experts.96.up_proj", "model.layers.21.mlp.experts.97.up_proj", "model.layers.21.mlp.experts.98.up_proj", "model.layers.21.mlp.experts.99.up_proj", "model.layers.21.mlp.experts.100.up_proj", "model.layers.21.mlp.experts.101.up_proj", "model.layers.21.mlp.experts.102.up_proj", "model.layers.21.mlp.experts.103.up_proj", "model.layers.21.mlp.experts.104.up_proj", "model.layers.21.mlp.experts.105.up_proj", "model.layers.21.mlp.experts.106.up_proj", "model.layers.21.mlp.experts.107.up_proj", "model.layers.21.mlp.experts.108.up_proj", "model.layers.21.mlp.experts.109.up_proj", "model.layers.21.mlp.experts.110.up_proj", "model.layers.21.mlp.experts.111.up_proj", "model.layers.21.mlp.experts.112.up_proj", "model.layers.21.mlp.experts.113.up_proj", "model.layers.21.mlp.experts.114.up_proj", "model.layers.21.mlp.experts.115.up_proj", "model.layers.21.mlp.experts.116.up_proj", "model.layers.21.mlp.experts.117.up_proj", "model.layers.21.mlp.experts.118.up_proj", "model.layers.21.mlp.experts.119.up_proj", "model.layers.21.mlp.experts.120.up_proj", "model.layers.21.mlp.experts.121.up_proj", "model.layers.21.mlp.experts.122.up_proj", "model.layers.21.mlp.experts.123.up_proj", "model.layers.21.mlp.experts.124.up_proj", "model.layers.21.mlp.experts.125.up_proj", "model.layers.21.mlp.experts.126.up_proj", "model.layers.21.mlp.experts.127.up_proj", "model.layers.21.mlp.experts.128.up_proj", "model.layers.21.mlp.experts.129.up_proj", "model.layers.21.mlp.experts.130.up_proj", "model.layers.21.mlp.experts.131.up_proj", "model.layers.21.mlp.experts.132.up_proj", "model.layers.21.mlp.experts.133.up_proj", "model.layers.21.mlp.experts.134.up_proj", "model.layers.21.mlp.experts.135.up_proj", "model.layers.21.mlp.experts.136.up_proj", "model.layers.21.mlp.experts.137.up_proj", "model.layers.21.mlp.experts.138.up_proj", "model.layers.21.mlp.experts.139.up_proj", "model.layers.21.mlp.experts.140.up_proj", "model.layers.21.mlp.experts.141.up_proj", "model.layers.21.mlp.experts.142.up_proj", "model.layers.21.mlp.experts.143.up_proj", "model.layers.21.mlp.experts.144.up_proj", "model.layers.21.mlp.experts.145.up_proj", "model.layers.21.mlp.experts.146.up_proj", "model.layers.21.mlp.experts.147.up_proj", "model.layers.21.mlp.experts.148.up_proj", "model.layers.21.mlp.experts.149.up_proj", "model.layers.21.mlp.experts.150.up_proj", "model.layers.21.mlp.experts.151.up_proj", "model.layers.21.mlp.experts.152.up_proj", "model.layers.21.mlp.experts.153.up_proj", "model.layers.21.mlp.experts.154.up_proj", "model.layers.21.mlp.experts.155.up_proj", "model.layers.21.mlp.experts.156.up_proj", "model.layers.21.mlp.experts.157.up_proj", "model.layers.21.mlp.experts.158.up_proj", "model.layers.21.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0005552025511860875, "dbits": 2516582400 }, { "dkld": -0.0008925462141633062, "dbits": 5033164800 }, { "dkld": -0.0008576113730669049, "dbits": 7549747200 }, { "dkld": -0.0007931969128549127, "dbits": 12582912000 } ] }, { "idx": 128, "layers": [ "model.layers.21.mlp.experts.0.down_proj", "model.layers.21.mlp.experts.1.down_proj", "model.layers.21.mlp.experts.2.down_proj", "model.layers.21.mlp.experts.3.down_proj", "model.layers.21.mlp.experts.4.down_proj", "model.layers.21.mlp.experts.5.down_proj", "model.layers.21.mlp.experts.6.down_proj", "model.layers.21.mlp.experts.7.down_proj", "model.layers.21.mlp.experts.8.down_proj", "model.layers.21.mlp.experts.9.down_proj", "model.layers.21.mlp.experts.10.down_proj", "model.layers.21.mlp.experts.11.down_proj", "model.layers.21.mlp.experts.12.down_proj", "model.layers.21.mlp.experts.13.down_proj", "model.layers.21.mlp.experts.14.down_proj", "model.layers.21.mlp.experts.15.down_proj", "model.layers.21.mlp.experts.16.down_proj", "model.layers.21.mlp.experts.17.down_proj", "model.layers.21.mlp.experts.18.down_proj", "model.layers.21.mlp.experts.19.down_proj", "model.layers.21.mlp.experts.20.down_proj", "model.layers.21.mlp.experts.21.down_proj", "model.layers.21.mlp.experts.22.down_proj", "model.layers.21.mlp.experts.23.down_proj", "model.layers.21.mlp.experts.24.down_proj", "model.layers.21.mlp.experts.25.down_proj", "model.layers.21.mlp.experts.26.down_proj", "model.layers.21.mlp.experts.27.down_proj", "model.layers.21.mlp.experts.28.down_proj", "model.layers.21.mlp.experts.29.down_proj", "model.layers.21.mlp.experts.30.down_proj", "model.layers.21.mlp.experts.31.down_proj", "model.layers.21.mlp.experts.32.down_proj", "model.layers.21.mlp.experts.33.down_proj", "model.layers.21.mlp.experts.34.down_proj", "model.layers.21.mlp.experts.35.down_proj", "model.layers.21.mlp.experts.36.down_proj", "model.layers.21.mlp.experts.37.down_proj", "model.layers.21.mlp.experts.38.down_proj", "model.layers.21.mlp.experts.39.down_proj", "model.layers.21.mlp.experts.40.down_proj", "model.layers.21.mlp.experts.41.down_proj", "model.layers.21.mlp.experts.42.down_proj", "model.layers.21.mlp.experts.43.down_proj", "model.layers.21.mlp.experts.44.down_proj", "model.layers.21.mlp.experts.45.down_proj", "model.layers.21.mlp.experts.46.down_proj", "model.layers.21.mlp.experts.47.down_proj", "model.layers.21.mlp.experts.48.down_proj", "model.layers.21.mlp.experts.49.down_proj", "model.layers.21.mlp.experts.50.down_proj", "model.layers.21.mlp.experts.51.down_proj", "model.layers.21.mlp.experts.52.down_proj", "model.layers.21.mlp.experts.53.down_proj", "model.layers.21.mlp.experts.54.down_proj", "model.layers.21.mlp.experts.55.down_proj", "model.layers.21.mlp.experts.56.down_proj", "model.layers.21.mlp.experts.57.down_proj", "model.layers.21.mlp.experts.58.down_proj", "model.layers.21.mlp.experts.59.down_proj", "model.layers.21.mlp.experts.60.down_proj", "model.layers.21.mlp.experts.61.down_proj", "model.layers.21.mlp.experts.62.down_proj", "model.layers.21.mlp.experts.63.down_proj", "model.layers.21.mlp.experts.64.down_proj", "model.layers.21.mlp.experts.65.down_proj", "model.layers.21.mlp.experts.66.down_proj", "model.layers.21.mlp.experts.67.down_proj", "model.layers.21.mlp.experts.68.down_proj", "model.layers.21.mlp.experts.69.down_proj", "model.layers.21.mlp.experts.70.down_proj", "model.layers.21.mlp.experts.71.down_proj", "model.layers.21.mlp.experts.72.down_proj", "model.layers.21.mlp.experts.73.down_proj", "model.layers.21.mlp.experts.74.down_proj", "model.layers.21.mlp.experts.75.down_proj", "model.layers.21.mlp.experts.76.down_proj", "model.layers.21.mlp.experts.77.down_proj", "model.layers.21.mlp.experts.78.down_proj", "model.layers.21.mlp.experts.79.down_proj", "model.layers.21.mlp.experts.80.down_proj", "model.layers.21.mlp.experts.81.down_proj", "model.layers.21.mlp.experts.82.down_proj", "model.layers.21.mlp.experts.83.down_proj", "model.layers.21.mlp.experts.84.down_proj", "model.layers.21.mlp.experts.85.down_proj", "model.layers.21.mlp.experts.86.down_proj", "model.layers.21.mlp.experts.87.down_proj", "model.layers.21.mlp.experts.88.down_proj", "model.layers.21.mlp.experts.89.down_proj", "model.layers.21.mlp.experts.90.down_proj", "model.layers.21.mlp.experts.91.down_proj", "model.layers.21.mlp.experts.92.down_proj", "model.layers.21.mlp.experts.93.down_proj", "model.layers.21.mlp.experts.94.down_proj", "model.layers.21.mlp.experts.95.down_proj", "model.layers.21.mlp.experts.96.down_proj", "model.layers.21.mlp.experts.97.down_proj", "model.layers.21.mlp.experts.98.down_proj", "model.layers.21.mlp.experts.99.down_proj", "model.layers.21.mlp.experts.100.down_proj", "model.layers.21.mlp.experts.101.down_proj", "model.layers.21.mlp.experts.102.down_proj", "model.layers.21.mlp.experts.103.down_proj", "model.layers.21.mlp.experts.104.down_proj", "model.layers.21.mlp.experts.105.down_proj", "model.layers.21.mlp.experts.106.down_proj", "model.layers.21.mlp.experts.107.down_proj", "model.layers.21.mlp.experts.108.down_proj", "model.layers.21.mlp.experts.109.down_proj", "model.layers.21.mlp.experts.110.down_proj", "model.layers.21.mlp.experts.111.down_proj", "model.layers.21.mlp.experts.112.down_proj", "model.layers.21.mlp.experts.113.down_proj", "model.layers.21.mlp.experts.114.down_proj", "model.layers.21.mlp.experts.115.down_proj", "model.layers.21.mlp.experts.116.down_proj", "model.layers.21.mlp.experts.117.down_proj", "model.layers.21.mlp.experts.118.down_proj", "model.layers.21.mlp.experts.119.down_proj", "model.layers.21.mlp.experts.120.down_proj", "model.layers.21.mlp.experts.121.down_proj", "model.layers.21.mlp.experts.122.down_proj", "model.layers.21.mlp.experts.123.down_proj", "model.layers.21.mlp.experts.124.down_proj", "model.layers.21.mlp.experts.125.down_proj", "model.layers.21.mlp.experts.126.down_proj", "model.layers.21.mlp.experts.127.down_proj", "model.layers.21.mlp.experts.128.down_proj", "model.layers.21.mlp.experts.129.down_proj", "model.layers.21.mlp.experts.130.down_proj", "model.layers.21.mlp.experts.131.down_proj", "model.layers.21.mlp.experts.132.down_proj", "model.layers.21.mlp.experts.133.down_proj", "model.layers.21.mlp.experts.134.down_proj", "model.layers.21.mlp.experts.135.down_proj", "model.layers.21.mlp.experts.136.down_proj", "model.layers.21.mlp.experts.137.down_proj", "model.layers.21.mlp.experts.138.down_proj", "model.layers.21.mlp.experts.139.down_proj", "model.layers.21.mlp.experts.140.down_proj", "model.layers.21.mlp.experts.141.down_proj", "model.layers.21.mlp.experts.142.down_proj", "model.layers.21.mlp.experts.143.down_proj", "model.layers.21.mlp.experts.144.down_proj", "model.layers.21.mlp.experts.145.down_proj", "model.layers.21.mlp.experts.146.down_proj", "model.layers.21.mlp.experts.147.down_proj", "model.layers.21.mlp.experts.148.down_proj", "model.layers.21.mlp.experts.149.down_proj", "model.layers.21.mlp.experts.150.down_proj", "model.layers.21.mlp.experts.151.down_proj", "model.layers.21.mlp.experts.152.down_proj", "model.layers.21.mlp.experts.153.down_proj", "model.layers.21.mlp.experts.154.down_proj", "model.layers.21.mlp.experts.155.down_proj", "model.layers.21.mlp.experts.156.down_proj", "model.layers.21.mlp.experts.157.down_proj", "model.layers.21.mlp.experts.158.down_proj", "model.layers.21.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00016568563878537057, "dbits": 1258291200 }, { "dkld": -0.00041924696415662766, "dbits": 2516582400 }, { "dkld": -0.000403569545596838, "dbits": 3774873600 }, { "dkld": -0.00035798195749521533, "dbits": 6291456000 } ] }, { "idx": 129, "layers": [ "model.layers.22.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00027466071769595424, "dbits": 62914560 }, { "dkld": -0.00029268879443407336, "dbits": 125829120 }, { "dkld": -0.0003801155835390091, "dbits": 188743680 }, { "dkld": -0.0003661200404167231, "dbits": 314572800 } ] }, { "idx": 130, "layers": [ "model.layers.22.self_attn.k_proj", "model.layers.22.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0004205134697258389, "dbits": 10485760 }, { "dkld": 1.1165440082547273e-05, "dbits": 20971520 }, { "dkld": 0.0003694671206176253, "dbits": 31457280 }, { "dkld": 0.0003065940923988736, "dbits": 52428800 } ] }, { "idx": 131, "layers": [ "model.layers.22.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0004835834726691357, "dbits": 62914560 }, { "dkld": -0.0003372650593519294, "dbits": 125829120 }, { "dkld": -0.00027176756411791125, "dbits": 188743680 }, { "dkld": -0.0002606497146189213, "dbits": 314572800 } ] }, { "idx": 132, "layers": [ "model.layers.22.mlp.shared_experts.gate_proj", "model.layers.22.mlp.shared_experts.up_proj", "model.layers.22.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0009415533393621334, "dbits": 23592960 }, { "dkld": 0.0008453944697976029, "dbits": 47185920 }, { "dkld": 0.0006436346098780549, "dbits": 70778880 }, { "dkld": 0.0005845166742801666, "dbits": 117964800 } ] }, { "idx": 133, "layers": [ "model.layers.22.mlp.experts.0.gate_proj", "model.layers.22.mlp.experts.1.gate_proj", "model.layers.22.mlp.experts.2.gate_proj", "model.layers.22.mlp.experts.3.gate_proj", "model.layers.22.mlp.experts.4.gate_proj", "model.layers.22.mlp.experts.5.gate_proj", "model.layers.22.mlp.experts.6.gate_proj", "model.layers.22.mlp.experts.7.gate_proj", "model.layers.22.mlp.experts.8.gate_proj", "model.layers.22.mlp.experts.9.gate_proj", "model.layers.22.mlp.experts.10.gate_proj", "model.layers.22.mlp.experts.11.gate_proj", "model.layers.22.mlp.experts.12.gate_proj", "model.layers.22.mlp.experts.13.gate_proj", "model.layers.22.mlp.experts.14.gate_proj", "model.layers.22.mlp.experts.15.gate_proj", "model.layers.22.mlp.experts.16.gate_proj", "model.layers.22.mlp.experts.17.gate_proj", "model.layers.22.mlp.experts.18.gate_proj", "model.layers.22.mlp.experts.19.gate_proj", "model.layers.22.mlp.experts.20.gate_proj", "model.layers.22.mlp.experts.21.gate_proj", "model.layers.22.mlp.experts.22.gate_proj", "model.layers.22.mlp.experts.23.gate_proj", "model.layers.22.mlp.experts.24.gate_proj", "model.layers.22.mlp.experts.25.gate_proj", "model.layers.22.mlp.experts.26.gate_proj", "model.layers.22.mlp.experts.27.gate_proj", "model.layers.22.mlp.experts.28.gate_proj", "model.layers.22.mlp.experts.29.gate_proj", "model.layers.22.mlp.experts.30.gate_proj", "model.layers.22.mlp.experts.31.gate_proj", "model.layers.22.mlp.experts.32.gate_proj", "model.layers.22.mlp.experts.33.gate_proj", "model.layers.22.mlp.experts.34.gate_proj", "model.layers.22.mlp.experts.35.gate_proj", "model.layers.22.mlp.experts.36.gate_proj", "model.layers.22.mlp.experts.37.gate_proj", "model.layers.22.mlp.experts.38.gate_proj", "model.layers.22.mlp.experts.39.gate_proj", "model.layers.22.mlp.experts.40.gate_proj", "model.layers.22.mlp.experts.41.gate_proj", "model.layers.22.mlp.experts.42.gate_proj", "model.layers.22.mlp.experts.43.gate_proj", "model.layers.22.mlp.experts.44.gate_proj", "model.layers.22.mlp.experts.45.gate_proj", "model.layers.22.mlp.experts.46.gate_proj", "model.layers.22.mlp.experts.47.gate_proj", "model.layers.22.mlp.experts.48.gate_proj", "model.layers.22.mlp.experts.49.gate_proj", "model.layers.22.mlp.experts.50.gate_proj", "model.layers.22.mlp.experts.51.gate_proj", "model.layers.22.mlp.experts.52.gate_proj", "model.layers.22.mlp.experts.53.gate_proj", "model.layers.22.mlp.experts.54.gate_proj", "model.layers.22.mlp.experts.55.gate_proj", "model.layers.22.mlp.experts.56.gate_proj", "model.layers.22.mlp.experts.57.gate_proj", "model.layers.22.mlp.experts.58.gate_proj", "model.layers.22.mlp.experts.59.gate_proj", "model.layers.22.mlp.experts.60.gate_proj", "model.layers.22.mlp.experts.61.gate_proj", "model.layers.22.mlp.experts.62.gate_proj", "model.layers.22.mlp.experts.63.gate_proj", "model.layers.22.mlp.experts.64.gate_proj", "model.layers.22.mlp.experts.65.gate_proj", "model.layers.22.mlp.experts.66.gate_proj", "model.layers.22.mlp.experts.67.gate_proj", "model.layers.22.mlp.experts.68.gate_proj", "model.layers.22.mlp.experts.69.gate_proj", "model.layers.22.mlp.experts.70.gate_proj", "model.layers.22.mlp.experts.71.gate_proj", "model.layers.22.mlp.experts.72.gate_proj", "model.layers.22.mlp.experts.73.gate_proj", "model.layers.22.mlp.experts.74.gate_proj", "model.layers.22.mlp.experts.75.gate_proj", "model.layers.22.mlp.experts.76.gate_proj", "model.layers.22.mlp.experts.77.gate_proj", "model.layers.22.mlp.experts.78.gate_proj", "model.layers.22.mlp.experts.79.gate_proj", "model.layers.22.mlp.experts.80.gate_proj", "model.layers.22.mlp.experts.81.gate_proj", "model.layers.22.mlp.experts.82.gate_proj", "model.layers.22.mlp.experts.83.gate_proj", "model.layers.22.mlp.experts.84.gate_proj", "model.layers.22.mlp.experts.85.gate_proj", "model.layers.22.mlp.experts.86.gate_proj", "model.layers.22.mlp.experts.87.gate_proj", "model.layers.22.mlp.experts.88.gate_proj", "model.layers.22.mlp.experts.89.gate_proj", "model.layers.22.mlp.experts.90.gate_proj", "model.layers.22.mlp.experts.91.gate_proj", "model.layers.22.mlp.experts.92.gate_proj", "model.layers.22.mlp.experts.93.gate_proj", "model.layers.22.mlp.experts.94.gate_proj", "model.layers.22.mlp.experts.95.gate_proj", "model.layers.22.mlp.experts.96.gate_proj", "model.layers.22.mlp.experts.97.gate_proj", "model.layers.22.mlp.experts.98.gate_proj", "model.layers.22.mlp.experts.99.gate_proj", "model.layers.22.mlp.experts.100.gate_proj", "model.layers.22.mlp.experts.101.gate_proj", "model.layers.22.mlp.experts.102.gate_proj", "model.layers.22.mlp.experts.103.gate_proj", "model.layers.22.mlp.experts.104.gate_proj", "model.layers.22.mlp.experts.105.gate_proj", "model.layers.22.mlp.experts.106.gate_proj", "model.layers.22.mlp.experts.107.gate_proj", "model.layers.22.mlp.experts.108.gate_proj", "model.layers.22.mlp.experts.109.gate_proj", "model.layers.22.mlp.experts.110.gate_proj", "model.layers.22.mlp.experts.111.gate_proj", "model.layers.22.mlp.experts.112.gate_proj", "model.layers.22.mlp.experts.113.gate_proj", "model.layers.22.mlp.experts.114.gate_proj", "model.layers.22.mlp.experts.115.gate_proj", "model.layers.22.mlp.experts.116.gate_proj", "model.layers.22.mlp.experts.117.gate_proj", "model.layers.22.mlp.experts.118.gate_proj", "model.layers.22.mlp.experts.119.gate_proj", "model.layers.22.mlp.experts.120.gate_proj", "model.layers.22.mlp.experts.121.gate_proj", "model.layers.22.mlp.experts.122.gate_proj", "model.layers.22.mlp.experts.123.gate_proj", "model.layers.22.mlp.experts.124.gate_proj", "model.layers.22.mlp.experts.125.gate_proj", "model.layers.22.mlp.experts.126.gate_proj", "model.layers.22.mlp.experts.127.gate_proj", "model.layers.22.mlp.experts.128.gate_proj", "model.layers.22.mlp.experts.129.gate_proj", "model.layers.22.mlp.experts.130.gate_proj", "model.layers.22.mlp.experts.131.gate_proj", "model.layers.22.mlp.experts.132.gate_proj", "model.layers.22.mlp.experts.133.gate_proj", "model.layers.22.mlp.experts.134.gate_proj", "model.layers.22.mlp.experts.135.gate_proj", "model.layers.22.mlp.experts.136.gate_proj", "model.layers.22.mlp.experts.137.gate_proj", "model.layers.22.mlp.experts.138.gate_proj", "model.layers.22.mlp.experts.139.gate_proj", "model.layers.22.mlp.experts.140.gate_proj", "model.layers.22.mlp.experts.141.gate_proj", "model.layers.22.mlp.experts.142.gate_proj", "model.layers.22.mlp.experts.143.gate_proj", "model.layers.22.mlp.experts.144.gate_proj", "model.layers.22.mlp.experts.145.gate_proj", "model.layers.22.mlp.experts.146.gate_proj", "model.layers.22.mlp.experts.147.gate_proj", "model.layers.22.mlp.experts.148.gate_proj", "model.layers.22.mlp.experts.149.gate_proj", "model.layers.22.mlp.experts.150.gate_proj", "model.layers.22.mlp.experts.151.gate_proj", "model.layers.22.mlp.experts.152.gate_proj", "model.layers.22.mlp.experts.153.gate_proj", "model.layers.22.mlp.experts.154.gate_proj", "model.layers.22.mlp.experts.155.gate_proj", "model.layers.22.mlp.experts.156.gate_proj", "model.layers.22.mlp.experts.157.gate_proj", "model.layers.22.mlp.experts.158.gate_proj", "model.layers.22.mlp.experts.159.gate_proj", "model.layers.22.mlp.experts.0.up_proj", "model.layers.22.mlp.experts.1.up_proj", "model.layers.22.mlp.experts.2.up_proj", "model.layers.22.mlp.experts.3.up_proj", "model.layers.22.mlp.experts.4.up_proj", "model.layers.22.mlp.experts.5.up_proj", "model.layers.22.mlp.experts.6.up_proj", "model.layers.22.mlp.experts.7.up_proj", "model.layers.22.mlp.experts.8.up_proj", "model.layers.22.mlp.experts.9.up_proj", "model.layers.22.mlp.experts.10.up_proj", "model.layers.22.mlp.experts.11.up_proj", "model.layers.22.mlp.experts.12.up_proj", "model.layers.22.mlp.experts.13.up_proj", "model.layers.22.mlp.experts.14.up_proj", "model.layers.22.mlp.experts.15.up_proj", "model.layers.22.mlp.experts.16.up_proj", "model.layers.22.mlp.experts.17.up_proj", "model.layers.22.mlp.experts.18.up_proj", "model.layers.22.mlp.experts.19.up_proj", "model.layers.22.mlp.experts.20.up_proj", "model.layers.22.mlp.experts.21.up_proj", "model.layers.22.mlp.experts.22.up_proj", "model.layers.22.mlp.experts.23.up_proj", "model.layers.22.mlp.experts.24.up_proj", "model.layers.22.mlp.experts.25.up_proj", "model.layers.22.mlp.experts.26.up_proj", "model.layers.22.mlp.experts.27.up_proj", "model.layers.22.mlp.experts.28.up_proj", "model.layers.22.mlp.experts.29.up_proj", "model.layers.22.mlp.experts.30.up_proj", "model.layers.22.mlp.experts.31.up_proj", "model.layers.22.mlp.experts.32.up_proj", "model.layers.22.mlp.experts.33.up_proj", "model.layers.22.mlp.experts.34.up_proj", "model.layers.22.mlp.experts.35.up_proj", "model.layers.22.mlp.experts.36.up_proj", "model.layers.22.mlp.experts.37.up_proj", "model.layers.22.mlp.experts.38.up_proj", "model.layers.22.mlp.experts.39.up_proj", "model.layers.22.mlp.experts.40.up_proj", "model.layers.22.mlp.experts.41.up_proj", "model.layers.22.mlp.experts.42.up_proj", "model.layers.22.mlp.experts.43.up_proj", "model.layers.22.mlp.experts.44.up_proj", "model.layers.22.mlp.experts.45.up_proj", "model.layers.22.mlp.experts.46.up_proj", "model.layers.22.mlp.experts.47.up_proj", "model.layers.22.mlp.experts.48.up_proj", "model.layers.22.mlp.experts.49.up_proj", "model.layers.22.mlp.experts.50.up_proj", "model.layers.22.mlp.experts.51.up_proj", "model.layers.22.mlp.experts.52.up_proj", "model.layers.22.mlp.experts.53.up_proj", "model.layers.22.mlp.experts.54.up_proj", "model.layers.22.mlp.experts.55.up_proj", "model.layers.22.mlp.experts.56.up_proj", "model.layers.22.mlp.experts.57.up_proj", "model.layers.22.mlp.experts.58.up_proj", "model.layers.22.mlp.experts.59.up_proj", "model.layers.22.mlp.experts.60.up_proj", "model.layers.22.mlp.experts.61.up_proj", "model.layers.22.mlp.experts.62.up_proj", "model.layers.22.mlp.experts.63.up_proj", "model.layers.22.mlp.experts.64.up_proj", "model.layers.22.mlp.experts.65.up_proj", "model.layers.22.mlp.experts.66.up_proj", "model.layers.22.mlp.experts.67.up_proj", "model.layers.22.mlp.experts.68.up_proj", "model.layers.22.mlp.experts.69.up_proj", "model.layers.22.mlp.experts.70.up_proj", "model.layers.22.mlp.experts.71.up_proj", "model.layers.22.mlp.experts.72.up_proj", "model.layers.22.mlp.experts.73.up_proj", "model.layers.22.mlp.experts.74.up_proj", "model.layers.22.mlp.experts.75.up_proj", "model.layers.22.mlp.experts.76.up_proj", "model.layers.22.mlp.experts.77.up_proj", "model.layers.22.mlp.experts.78.up_proj", "model.layers.22.mlp.experts.79.up_proj", "model.layers.22.mlp.experts.80.up_proj", "model.layers.22.mlp.experts.81.up_proj", "model.layers.22.mlp.experts.82.up_proj", "model.layers.22.mlp.experts.83.up_proj", "model.layers.22.mlp.experts.84.up_proj", "model.layers.22.mlp.experts.85.up_proj", "model.layers.22.mlp.experts.86.up_proj", "model.layers.22.mlp.experts.87.up_proj", "model.layers.22.mlp.experts.88.up_proj", "model.layers.22.mlp.experts.89.up_proj", "model.layers.22.mlp.experts.90.up_proj", "model.layers.22.mlp.experts.91.up_proj", "model.layers.22.mlp.experts.92.up_proj", "model.layers.22.mlp.experts.93.up_proj", "model.layers.22.mlp.experts.94.up_proj", "model.layers.22.mlp.experts.95.up_proj", "model.layers.22.mlp.experts.96.up_proj", "model.layers.22.mlp.experts.97.up_proj", "model.layers.22.mlp.experts.98.up_proj", "model.layers.22.mlp.experts.99.up_proj", "model.layers.22.mlp.experts.100.up_proj", "model.layers.22.mlp.experts.101.up_proj", "model.layers.22.mlp.experts.102.up_proj", "model.layers.22.mlp.experts.103.up_proj", "model.layers.22.mlp.experts.104.up_proj", "model.layers.22.mlp.experts.105.up_proj", "model.layers.22.mlp.experts.106.up_proj", "model.layers.22.mlp.experts.107.up_proj", "model.layers.22.mlp.experts.108.up_proj", "model.layers.22.mlp.experts.109.up_proj", "model.layers.22.mlp.experts.110.up_proj", "model.layers.22.mlp.experts.111.up_proj", "model.layers.22.mlp.experts.112.up_proj", "model.layers.22.mlp.experts.113.up_proj", "model.layers.22.mlp.experts.114.up_proj", "model.layers.22.mlp.experts.115.up_proj", "model.layers.22.mlp.experts.116.up_proj", "model.layers.22.mlp.experts.117.up_proj", "model.layers.22.mlp.experts.118.up_proj", "model.layers.22.mlp.experts.119.up_proj", "model.layers.22.mlp.experts.120.up_proj", "model.layers.22.mlp.experts.121.up_proj", "model.layers.22.mlp.experts.122.up_proj", "model.layers.22.mlp.experts.123.up_proj", "model.layers.22.mlp.experts.124.up_proj", "model.layers.22.mlp.experts.125.up_proj", "model.layers.22.mlp.experts.126.up_proj", "model.layers.22.mlp.experts.127.up_proj", "model.layers.22.mlp.experts.128.up_proj", "model.layers.22.mlp.experts.129.up_proj", "model.layers.22.mlp.experts.130.up_proj", "model.layers.22.mlp.experts.131.up_proj", "model.layers.22.mlp.experts.132.up_proj", "model.layers.22.mlp.experts.133.up_proj", "model.layers.22.mlp.experts.134.up_proj", "model.layers.22.mlp.experts.135.up_proj", "model.layers.22.mlp.experts.136.up_proj", "model.layers.22.mlp.experts.137.up_proj", "model.layers.22.mlp.experts.138.up_proj", "model.layers.22.mlp.experts.139.up_proj", "model.layers.22.mlp.experts.140.up_proj", "model.layers.22.mlp.experts.141.up_proj", "model.layers.22.mlp.experts.142.up_proj", "model.layers.22.mlp.experts.143.up_proj", "model.layers.22.mlp.experts.144.up_proj", "model.layers.22.mlp.experts.145.up_proj", "model.layers.22.mlp.experts.146.up_proj", "model.layers.22.mlp.experts.147.up_proj", "model.layers.22.mlp.experts.148.up_proj", "model.layers.22.mlp.experts.149.up_proj", "model.layers.22.mlp.experts.150.up_proj", "model.layers.22.mlp.experts.151.up_proj", "model.layers.22.mlp.experts.152.up_proj", "model.layers.22.mlp.experts.153.up_proj", "model.layers.22.mlp.experts.154.up_proj", "model.layers.22.mlp.experts.155.up_proj", "model.layers.22.mlp.experts.156.up_proj", "model.layers.22.mlp.experts.157.up_proj", "model.layers.22.mlp.experts.158.up_proj", "model.layers.22.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0004477288573980359, "dbits": 2516582400 }, { "dkld": -0.0005808314308524132, "dbits": 5033164800 }, { "dkld": -0.0005601430311799049, "dbits": 7549747200 }, { "dkld": -0.0006389514543116148, "dbits": 12582912000 } ] }, { "idx": 134, "layers": [ "model.layers.22.mlp.experts.0.down_proj", "model.layers.22.mlp.experts.1.down_proj", "model.layers.22.mlp.experts.2.down_proj", "model.layers.22.mlp.experts.3.down_proj", "model.layers.22.mlp.experts.4.down_proj", "model.layers.22.mlp.experts.5.down_proj", "model.layers.22.mlp.experts.6.down_proj", "model.layers.22.mlp.experts.7.down_proj", "model.layers.22.mlp.experts.8.down_proj", "model.layers.22.mlp.experts.9.down_proj", "model.layers.22.mlp.experts.10.down_proj", "model.layers.22.mlp.experts.11.down_proj", "model.layers.22.mlp.experts.12.down_proj", "model.layers.22.mlp.experts.13.down_proj", "model.layers.22.mlp.experts.14.down_proj", "model.layers.22.mlp.experts.15.down_proj", "model.layers.22.mlp.experts.16.down_proj", "model.layers.22.mlp.experts.17.down_proj", "model.layers.22.mlp.experts.18.down_proj", "model.layers.22.mlp.experts.19.down_proj", "model.layers.22.mlp.experts.20.down_proj", "model.layers.22.mlp.experts.21.down_proj", "model.layers.22.mlp.experts.22.down_proj", "model.layers.22.mlp.experts.23.down_proj", "model.layers.22.mlp.experts.24.down_proj", "model.layers.22.mlp.experts.25.down_proj", "model.layers.22.mlp.experts.26.down_proj", "model.layers.22.mlp.experts.27.down_proj", "model.layers.22.mlp.experts.28.down_proj", "model.layers.22.mlp.experts.29.down_proj", "model.layers.22.mlp.experts.30.down_proj", "model.layers.22.mlp.experts.31.down_proj", "model.layers.22.mlp.experts.32.down_proj", "model.layers.22.mlp.experts.33.down_proj", "model.layers.22.mlp.experts.34.down_proj", "model.layers.22.mlp.experts.35.down_proj", "model.layers.22.mlp.experts.36.down_proj", "model.layers.22.mlp.experts.37.down_proj", "model.layers.22.mlp.experts.38.down_proj", "model.layers.22.mlp.experts.39.down_proj", "model.layers.22.mlp.experts.40.down_proj", "model.layers.22.mlp.experts.41.down_proj", "model.layers.22.mlp.experts.42.down_proj", "model.layers.22.mlp.experts.43.down_proj", "model.layers.22.mlp.experts.44.down_proj", "model.layers.22.mlp.experts.45.down_proj", "model.layers.22.mlp.experts.46.down_proj", "model.layers.22.mlp.experts.47.down_proj", "model.layers.22.mlp.experts.48.down_proj", "model.layers.22.mlp.experts.49.down_proj", "model.layers.22.mlp.experts.50.down_proj", "model.layers.22.mlp.experts.51.down_proj", "model.layers.22.mlp.experts.52.down_proj", "model.layers.22.mlp.experts.53.down_proj", "model.layers.22.mlp.experts.54.down_proj", "model.layers.22.mlp.experts.55.down_proj", "model.layers.22.mlp.experts.56.down_proj", "model.layers.22.mlp.experts.57.down_proj", "model.layers.22.mlp.experts.58.down_proj", "model.layers.22.mlp.experts.59.down_proj", "model.layers.22.mlp.experts.60.down_proj", "model.layers.22.mlp.experts.61.down_proj", "model.layers.22.mlp.experts.62.down_proj", "model.layers.22.mlp.experts.63.down_proj", "model.layers.22.mlp.experts.64.down_proj", "model.layers.22.mlp.experts.65.down_proj", "model.layers.22.mlp.experts.66.down_proj", "model.layers.22.mlp.experts.67.down_proj", "model.layers.22.mlp.experts.68.down_proj", "model.layers.22.mlp.experts.69.down_proj", "model.layers.22.mlp.experts.70.down_proj", "model.layers.22.mlp.experts.71.down_proj", "model.layers.22.mlp.experts.72.down_proj", "model.layers.22.mlp.experts.73.down_proj", "model.layers.22.mlp.experts.74.down_proj", "model.layers.22.mlp.experts.75.down_proj", "model.layers.22.mlp.experts.76.down_proj", "model.layers.22.mlp.experts.77.down_proj", "model.layers.22.mlp.experts.78.down_proj", "model.layers.22.mlp.experts.79.down_proj", "model.layers.22.mlp.experts.80.down_proj", "model.layers.22.mlp.experts.81.down_proj", "model.layers.22.mlp.experts.82.down_proj", "model.layers.22.mlp.experts.83.down_proj", "model.layers.22.mlp.experts.84.down_proj", "model.layers.22.mlp.experts.85.down_proj", "model.layers.22.mlp.experts.86.down_proj", "model.layers.22.mlp.experts.87.down_proj", "model.layers.22.mlp.experts.88.down_proj", "model.layers.22.mlp.experts.89.down_proj", "model.layers.22.mlp.experts.90.down_proj", "model.layers.22.mlp.experts.91.down_proj", "model.layers.22.mlp.experts.92.down_proj", "model.layers.22.mlp.experts.93.down_proj", "model.layers.22.mlp.experts.94.down_proj", "model.layers.22.mlp.experts.95.down_proj", "model.layers.22.mlp.experts.96.down_proj", "model.layers.22.mlp.experts.97.down_proj", "model.layers.22.mlp.experts.98.down_proj", "model.layers.22.mlp.experts.99.down_proj", "model.layers.22.mlp.experts.100.down_proj", "model.layers.22.mlp.experts.101.down_proj", "model.layers.22.mlp.experts.102.down_proj", "model.layers.22.mlp.experts.103.down_proj", "model.layers.22.mlp.experts.104.down_proj", "model.layers.22.mlp.experts.105.down_proj", "model.layers.22.mlp.experts.106.down_proj", "model.layers.22.mlp.experts.107.down_proj", "model.layers.22.mlp.experts.108.down_proj", "model.layers.22.mlp.experts.109.down_proj", "model.layers.22.mlp.experts.110.down_proj", "model.layers.22.mlp.experts.111.down_proj", "model.layers.22.mlp.experts.112.down_proj", "model.layers.22.mlp.experts.113.down_proj", "model.layers.22.mlp.experts.114.down_proj", "model.layers.22.mlp.experts.115.down_proj", "model.layers.22.mlp.experts.116.down_proj", "model.layers.22.mlp.experts.117.down_proj", "model.layers.22.mlp.experts.118.down_proj", "model.layers.22.mlp.experts.119.down_proj", "model.layers.22.mlp.experts.120.down_proj", "model.layers.22.mlp.experts.121.down_proj", "model.layers.22.mlp.experts.122.down_proj", "model.layers.22.mlp.experts.123.down_proj", "model.layers.22.mlp.experts.124.down_proj", "model.layers.22.mlp.experts.125.down_proj", "model.layers.22.mlp.experts.126.down_proj", "model.layers.22.mlp.experts.127.down_proj", "model.layers.22.mlp.experts.128.down_proj", "model.layers.22.mlp.experts.129.down_proj", "model.layers.22.mlp.experts.130.down_proj", "model.layers.22.mlp.experts.131.down_proj", "model.layers.22.mlp.experts.132.down_proj", "model.layers.22.mlp.experts.133.down_proj", "model.layers.22.mlp.experts.134.down_proj", "model.layers.22.mlp.experts.135.down_proj", "model.layers.22.mlp.experts.136.down_proj", "model.layers.22.mlp.experts.137.down_proj", "model.layers.22.mlp.experts.138.down_proj", "model.layers.22.mlp.experts.139.down_proj", "model.layers.22.mlp.experts.140.down_proj", "model.layers.22.mlp.experts.141.down_proj", "model.layers.22.mlp.experts.142.down_proj", "model.layers.22.mlp.experts.143.down_proj", "model.layers.22.mlp.experts.144.down_proj", "model.layers.22.mlp.experts.145.down_proj", "model.layers.22.mlp.experts.146.down_proj", "model.layers.22.mlp.experts.147.down_proj", "model.layers.22.mlp.experts.148.down_proj", "model.layers.22.mlp.experts.149.down_proj", "model.layers.22.mlp.experts.150.down_proj", "model.layers.22.mlp.experts.151.down_proj", "model.layers.22.mlp.experts.152.down_proj", "model.layers.22.mlp.experts.153.down_proj", "model.layers.22.mlp.experts.154.down_proj", "model.layers.22.mlp.experts.155.down_proj", "model.layers.22.mlp.experts.156.down_proj", "model.layers.22.mlp.experts.157.down_proj", "model.layers.22.mlp.experts.158.down_proj", "model.layers.22.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00022560022771359045, "dbits": 1258291200 }, { "dkld": -0.000331104360520848, "dbits": 2516582400 }, { "dkld": -0.000331162940710783, "dbits": 3774873600 }, { "dkld": -0.0003575649112463053, "dbits": 6291456000 } ] }, { "idx": 135, "layers": [ "model.layers.23.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0006711697205901146, "dbits": 62914560 }, { "dkld": -0.0006315836682915799, "dbits": 125829120 }, { "dkld": -0.0005352464504539967, "dbits": 188743680 }, { "dkld": -0.000628636684268713, "dbits": 314572800 } ] }, { "idx": 136, "layers": [ "model.layers.23.self_attn.k_proj", "model.layers.23.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00013491595163941938, "dbits": 10485760 }, { "dkld": -0.0005178587511181915, "dbits": 20971520 }, { "dkld": -0.0001767035573720932, "dbits": 31457280 }, { "dkld": -0.00034958999603987295, "dbits": 52428800 } ] }, { "idx": 137, "layers": [ "model.layers.23.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0004958899691701046, "dbits": 62914560 }, { "dkld": 7.38089904189082e-05, "dbits": 125829120 }, { "dkld": -0.00016076918691397546, "dbits": 188743680 }, { "dkld": -0.00019063604995608885, "dbits": 314572800 } ] }, { "idx": 138, "layers": [ "model.layers.23.mlp.shared_experts.gate_proj", "model.layers.23.mlp.shared_experts.up_proj", "model.layers.23.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.000942845083773125, "dbits": 23592960 }, { "dkld": 0.0006470702588558114, "dbits": 47185920 }, { "dkld": 0.0006886973977088817, "dbits": 70778880 }, { "dkld": 0.0008318576961755669, "dbits": 117964800 } ] }, { "idx": 139, "layers": [ "model.layers.23.mlp.experts.0.gate_proj", "model.layers.23.mlp.experts.1.gate_proj", "model.layers.23.mlp.experts.2.gate_proj", "model.layers.23.mlp.experts.3.gate_proj", "model.layers.23.mlp.experts.4.gate_proj", "model.layers.23.mlp.experts.5.gate_proj", "model.layers.23.mlp.experts.6.gate_proj", "model.layers.23.mlp.experts.7.gate_proj", "model.layers.23.mlp.experts.8.gate_proj", "model.layers.23.mlp.experts.9.gate_proj", "model.layers.23.mlp.experts.10.gate_proj", "model.layers.23.mlp.experts.11.gate_proj", "model.layers.23.mlp.experts.12.gate_proj", "model.layers.23.mlp.experts.13.gate_proj", "model.layers.23.mlp.experts.14.gate_proj", "model.layers.23.mlp.experts.15.gate_proj", "model.layers.23.mlp.experts.16.gate_proj", "model.layers.23.mlp.experts.17.gate_proj", "model.layers.23.mlp.experts.18.gate_proj", "model.layers.23.mlp.experts.19.gate_proj", "model.layers.23.mlp.experts.20.gate_proj", "model.layers.23.mlp.experts.21.gate_proj", "model.layers.23.mlp.experts.22.gate_proj", "model.layers.23.mlp.experts.23.gate_proj", "model.layers.23.mlp.experts.24.gate_proj", "model.layers.23.mlp.experts.25.gate_proj", "model.layers.23.mlp.experts.26.gate_proj", "model.layers.23.mlp.experts.27.gate_proj", "model.layers.23.mlp.experts.28.gate_proj", "model.layers.23.mlp.experts.29.gate_proj", "model.layers.23.mlp.experts.30.gate_proj", "model.layers.23.mlp.experts.31.gate_proj", "model.layers.23.mlp.experts.32.gate_proj", "model.layers.23.mlp.experts.33.gate_proj", "model.layers.23.mlp.experts.34.gate_proj", "model.layers.23.mlp.experts.35.gate_proj", "model.layers.23.mlp.experts.36.gate_proj", "model.layers.23.mlp.experts.37.gate_proj", "model.layers.23.mlp.experts.38.gate_proj", "model.layers.23.mlp.experts.39.gate_proj", "model.layers.23.mlp.experts.40.gate_proj", "model.layers.23.mlp.experts.41.gate_proj", "model.layers.23.mlp.experts.42.gate_proj", "model.layers.23.mlp.experts.43.gate_proj", "model.layers.23.mlp.experts.44.gate_proj", "model.layers.23.mlp.experts.45.gate_proj", "model.layers.23.mlp.experts.46.gate_proj", "model.layers.23.mlp.experts.47.gate_proj", "model.layers.23.mlp.experts.48.gate_proj", "model.layers.23.mlp.experts.49.gate_proj", "model.layers.23.mlp.experts.50.gate_proj", "model.layers.23.mlp.experts.51.gate_proj", "model.layers.23.mlp.experts.52.gate_proj", "model.layers.23.mlp.experts.53.gate_proj", "model.layers.23.mlp.experts.54.gate_proj", "model.layers.23.mlp.experts.55.gate_proj", "model.layers.23.mlp.experts.56.gate_proj", "model.layers.23.mlp.experts.57.gate_proj", "model.layers.23.mlp.experts.58.gate_proj", "model.layers.23.mlp.experts.59.gate_proj", "model.layers.23.mlp.experts.60.gate_proj", "model.layers.23.mlp.experts.61.gate_proj", "model.layers.23.mlp.experts.62.gate_proj", "model.layers.23.mlp.experts.63.gate_proj", "model.layers.23.mlp.experts.64.gate_proj", "model.layers.23.mlp.experts.65.gate_proj", "model.layers.23.mlp.experts.66.gate_proj", "model.layers.23.mlp.experts.67.gate_proj", "model.layers.23.mlp.experts.68.gate_proj", "model.layers.23.mlp.experts.69.gate_proj", "model.layers.23.mlp.experts.70.gate_proj", "model.layers.23.mlp.experts.71.gate_proj", "model.layers.23.mlp.experts.72.gate_proj", "model.layers.23.mlp.experts.73.gate_proj", "model.layers.23.mlp.experts.74.gate_proj", "model.layers.23.mlp.experts.75.gate_proj", "model.layers.23.mlp.experts.76.gate_proj", "model.layers.23.mlp.experts.77.gate_proj", "model.layers.23.mlp.experts.78.gate_proj", "model.layers.23.mlp.experts.79.gate_proj", "model.layers.23.mlp.experts.80.gate_proj", "model.layers.23.mlp.experts.81.gate_proj", "model.layers.23.mlp.experts.82.gate_proj", "model.layers.23.mlp.experts.83.gate_proj", "model.layers.23.mlp.experts.84.gate_proj", "model.layers.23.mlp.experts.85.gate_proj", "model.layers.23.mlp.experts.86.gate_proj", "model.layers.23.mlp.experts.87.gate_proj", "model.layers.23.mlp.experts.88.gate_proj", "model.layers.23.mlp.experts.89.gate_proj", "model.layers.23.mlp.experts.90.gate_proj", "model.layers.23.mlp.experts.91.gate_proj", "model.layers.23.mlp.experts.92.gate_proj", "model.layers.23.mlp.experts.93.gate_proj", "model.layers.23.mlp.experts.94.gate_proj", "model.layers.23.mlp.experts.95.gate_proj", "model.layers.23.mlp.experts.96.gate_proj", "model.layers.23.mlp.experts.97.gate_proj", "model.layers.23.mlp.experts.98.gate_proj", "model.layers.23.mlp.experts.99.gate_proj", "model.layers.23.mlp.experts.100.gate_proj", "model.layers.23.mlp.experts.101.gate_proj", "model.layers.23.mlp.experts.102.gate_proj", "model.layers.23.mlp.experts.103.gate_proj", "model.layers.23.mlp.experts.104.gate_proj", "model.layers.23.mlp.experts.105.gate_proj", "model.layers.23.mlp.experts.106.gate_proj", "model.layers.23.mlp.experts.107.gate_proj", "model.layers.23.mlp.experts.108.gate_proj", "model.layers.23.mlp.experts.109.gate_proj", "model.layers.23.mlp.experts.110.gate_proj", "model.layers.23.mlp.experts.111.gate_proj", "model.layers.23.mlp.experts.112.gate_proj", "model.layers.23.mlp.experts.113.gate_proj", "model.layers.23.mlp.experts.114.gate_proj", "model.layers.23.mlp.experts.115.gate_proj", "model.layers.23.mlp.experts.116.gate_proj", "model.layers.23.mlp.experts.117.gate_proj", "model.layers.23.mlp.experts.118.gate_proj", "model.layers.23.mlp.experts.119.gate_proj", "model.layers.23.mlp.experts.120.gate_proj", "model.layers.23.mlp.experts.121.gate_proj", "model.layers.23.mlp.experts.122.gate_proj", "model.layers.23.mlp.experts.123.gate_proj", "model.layers.23.mlp.experts.124.gate_proj", "model.layers.23.mlp.experts.125.gate_proj", "model.layers.23.mlp.experts.126.gate_proj", "model.layers.23.mlp.experts.127.gate_proj", "model.layers.23.mlp.experts.128.gate_proj", "model.layers.23.mlp.experts.129.gate_proj", "model.layers.23.mlp.experts.130.gate_proj", "model.layers.23.mlp.experts.131.gate_proj", "model.layers.23.mlp.experts.132.gate_proj", "model.layers.23.mlp.experts.133.gate_proj", "model.layers.23.mlp.experts.134.gate_proj", "model.layers.23.mlp.experts.135.gate_proj", "model.layers.23.mlp.experts.136.gate_proj", "model.layers.23.mlp.experts.137.gate_proj", "model.layers.23.mlp.experts.138.gate_proj", "model.layers.23.mlp.experts.139.gate_proj", "model.layers.23.mlp.experts.140.gate_proj", "model.layers.23.mlp.experts.141.gate_proj", "model.layers.23.mlp.experts.142.gate_proj", "model.layers.23.mlp.experts.143.gate_proj", "model.layers.23.mlp.experts.144.gate_proj", "model.layers.23.mlp.experts.145.gate_proj", "model.layers.23.mlp.experts.146.gate_proj", "model.layers.23.mlp.experts.147.gate_proj", "model.layers.23.mlp.experts.148.gate_proj", "model.layers.23.mlp.experts.149.gate_proj", "model.layers.23.mlp.experts.150.gate_proj", "model.layers.23.mlp.experts.151.gate_proj", "model.layers.23.mlp.experts.152.gate_proj", "model.layers.23.mlp.experts.153.gate_proj", "model.layers.23.mlp.experts.154.gate_proj", "model.layers.23.mlp.experts.155.gate_proj", "model.layers.23.mlp.experts.156.gate_proj", "model.layers.23.mlp.experts.157.gate_proj", "model.layers.23.mlp.experts.158.gate_proj", "model.layers.23.mlp.experts.159.gate_proj", "model.layers.23.mlp.experts.0.up_proj", "model.layers.23.mlp.experts.1.up_proj", "model.layers.23.mlp.experts.2.up_proj", "model.layers.23.mlp.experts.3.up_proj", "model.layers.23.mlp.experts.4.up_proj", "model.layers.23.mlp.experts.5.up_proj", "model.layers.23.mlp.experts.6.up_proj", "model.layers.23.mlp.experts.7.up_proj", "model.layers.23.mlp.experts.8.up_proj", "model.layers.23.mlp.experts.9.up_proj", "model.layers.23.mlp.experts.10.up_proj", "model.layers.23.mlp.experts.11.up_proj", "model.layers.23.mlp.experts.12.up_proj", "model.layers.23.mlp.experts.13.up_proj", "model.layers.23.mlp.experts.14.up_proj", "model.layers.23.mlp.experts.15.up_proj", "model.layers.23.mlp.experts.16.up_proj", "model.layers.23.mlp.experts.17.up_proj", "model.layers.23.mlp.experts.18.up_proj", "model.layers.23.mlp.experts.19.up_proj", "model.layers.23.mlp.experts.20.up_proj", "model.layers.23.mlp.experts.21.up_proj", "model.layers.23.mlp.experts.22.up_proj", "model.layers.23.mlp.experts.23.up_proj", "model.layers.23.mlp.experts.24.up_proj", "model.layers.23.mlp.experts.25.up_proj", "model.layers.23.mlp.experts.26.up_proj", "model.layers.23.mlp.experts.27.up_proj", "model.layers.23.mlp.experts.28.up_proj", "model.layers.23.mlp.experts.29.up_proj", "model.layers.23.mlp.experts.30.up_proj", "model.layers.23.mlp.experts.31.up_proj", "model.layers.23.mlp.experts.32.up_proj", "model.layers.23.mlp.experts.33.up_proj", "model.layers.23.mlp.experts.34.up_proj", "model.layers.23.mlp.experts.35.up_proj", "model.layers.23.mlp.experts.36.up_proj", "model.layers.23.mlp.experts.37.up_proj", "model.layers.23.mlp.experts.38.up_proj", "model.layers.23.mlp.experts.39.up_proj", "model.layers.23.mlp.experts.40.up_proj", "model.layers.23.mlp.experts.41.up_proj", "model.layers.23.mlp.experts.42.up_proj", "model.layers.23.mlp.experts.43.up_proj", "model.layers.23.mlp.experts.44.up_proj", "model.layers.23.mlp.experts.45.up_proj", "model.layers.23.mlp.experts.46.up_proj", "model.layers.23.mlp.experts.47.up_proj", "model.layers.23.mlp.experts.48.up_proj", "model.layers.23.mlp.experts.49.up_proj", "model.layers.23.mlp.experts.50.up_proj", "model.layers.23.mlp.experts.51.up_proj", "model.layers.23.mlp.experts.52.up_proj", "model.layers.23.mlp.experts.53.up_proj", "model.layers.23.mlp.experts.54.up_proj", "model.layers.23.mlp.experts.55.up_proj", "model.layers.23.mlp.experts.56.up_proj", "model.layers.23.mlp.experts.57.up_proj", "model.layers.23.mlp.experts.58.up_proj", "model.layers.23.mlp.experts.59.up_proj", "model.layers.23.mlp.experts.60.up_proj", "model.layers.23.mlp.experts.61.up_proj", "model.layers.23.mlp.experts.62.up_proj", "model.layers.23.mlp.experts.63.up_proj", "model.layers.23.mlp.experts.64.up_proj", "model.layers.23.mlp.experts.65.up_proj", "model.layers.23.mlp.experts.66.up_proj", "model.layers.23.mlp.experts.67.up_proj", "model.layers.23.mlp.experts.68.up_proj", "model.layers.23.mlp.experts.69.up_proj", "model.layers.23.mlp.experts.70.up_proj", "model.layers.23.mlp.experts.71.up_proj", "model.layers.23.mlp.experts.72.up_proj", "model.layers.23.mlp.experts.73.up_proj", "model.layers.23.mlp.experts.74.up_proj", "model.layers.23.mlp.experts.75.up_proj", "model.layers.23.mlp.experts.76.up_proj", "model.layers.23.mlp.experts.77.up_proj", "model.layers.23.mlp.experts.78.up_proj", "model.layers.23.mlp.experts.79.up_proj", "model.layers.23.mlp.experts.80.up_proj", "model.layers.23.mlp.experts.81.up_proj", "model.layers.23.mlp.experts.82.up_proj", "model.layers.23.mlp.experts.83.up_proj", "model.layers.23.mlp.experts.84.up_proj", "model.layers.23.mlp.experts.85.up_proj", "model.layers.23.mlp.experts.86.up_proj", "model.layers.23.mlp.experts.87.up_proj", "model.layers.23.mlp.experts.88.up_proj", "model.layers.23.mlp.experts.89.up_proj", "model.layers.23.mlp.experts.90.up_proj", "model.layers.23.mlp.experts.91.up_proj", "model.layers.23.mlp.experts.92.up_proj", "model.layers.23.mlp.experts.93.up_proj", "model.layers.23.mlp.experts.94.up_proj", "model.layers.23.mlp.experts.95.up_proj", "model.layers.23.mlp.experts.96.up_proj", "model.layers.23.mlp.experts.97.up_proj", "model.layers.23.mlp.experts.98.up_proj", "model.layers.23.mlp.experts.99.up_proj", "model.layers.23.mlp.experts.100.up_proj", "model.layers.23.mlp.experts.101.up_proj", "model.layers.23.mlp.experts.102.up_proj", "model.layers.23.mlp.experts.103.up_proj", "model.layers.23.mlp.experts.104.up_proj", "model.layers.23.mlp.experts.105.up_proj", "model.layers.23.mlp.experts.106.up_proj", "model.layers.23.mlp.experts.107.up_proj", "model.layers.23.mlp.experts.108.up_proj", "model.layers.23.mlp.experts.109.up_proj", "model.layers.23.mlp.experts.110.up_proj", "model.layers.23.mlp.experts.111.up_proj", "model.layers.23.mlp.experts.112.up_proj", "model.layers.23.mlp.experts.113.up_proj", "model.layers.23.mlp.experts.114.up_proj", "model.layers.23.mlp.experts.115.up_proj", "model.layers.23.mlp.experts.116.up_proj", "model.layers.23.mlp.experts.117.up_proj", "model.layers.23.mlp.experts.118.up_proj", "model.layers.23.mlp.experts.119.up_proj", "model.layers.23.mlp.experts.120.up_proj", "model.layers.23.mlp.experts.121.up_proj", "model.layers.23.mlp.experts.122.up_proj", "model.layers.23.mlp.experts.123.up_proj", "model.layers.23.mlp.experts.124.up_proj", "model.layers.23.mlp.experts.125.up_proj", "model.layers.23.mlp.experts.126.up_proj", "model.layers.23.mlp.experts.127.up_proj", "model.layers.23.mlp.experts.128.up_proj", "model.layers.23.mlp.experts.129.up_proj", "model.layers.23.mlp.experts.130.up_proj", "model.layers.23.mlp.experts.131.up_proj", "model.layers.23.mlp.experts.132.up_proj", "model.layers.23.mlp.experts.133.up_proj", "model.layers.23.mlp.experts.134.up_proj", "model.layers.23.mlp.experts.135.up_proj", "model.layers.23.mlp.experts.136.up_proj", "model.layers.23.mlp.experts.137.up_proj", "model.layers.23.mlp.experts.138.up_proj", "model.layers.23.mlp.experts.139.up_proj", "model.layers.23.mlp.experts.140.up_proj", "model.layers.23.mlp.experts.141.up_proj", "model.layers.23.mlp.experts.142.up_proj", "model.layers.23.mlp.experts.143.up_proj", "model.layers.23.mlp.experts.144.up_proj", "model.layers.23.mlp.experts.145.up_proj", "model.layers.23.mlp.experts.146.up_proj", "model.layers.23.mlp.experts.147.up_proj", "model.layers.23.mlp.experts.148.up_proj", "model.layers.23.mlp.experts.149.up_proj", "model.layers.23.mlp.experts.150.up_proj", "model.layers.23.mlp.experts.151.up_proj", "model.layers.23.mlp.experts.152.up_proj", "model.layers.23.mlp.experts.153.up_proj", "model.layers.23.mlp.experts.154.up_proj", "model.layers.23.mlp.experts.155.up_proj", "model.layers.23.mlp.experts.156.up_proj", "model.layers.23.mlp.experts.157.up_proj", "model.layers.23.mlp.experts.158.up_proj", "model.layers.23.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0008187808096408872, "dbits": 2516582400 }, { "dkld": -0.0010117914527654759, "dbits": 5033164800 }, { "dkld": -0.001009310595691204, "dbits": 7549747200 }, { "dkld": -0.0009419829584658146, "dbits": 12582912000 } ] }, { "idx": 140, "layers": [ "model.layers.23.mlp.experts.0.down_proj", "model.layers.23.mlp.experts.1.down_proj", "model.layers.23.mlp.experts.2.down_proj", "model.layers.23.mlp.experts.3.down_proj", "model.layers.23.mlp.experts.4.down_proj", "model.layers.23.mlp.experts.5.down_proj", "model.layers.23.mlp.experts.6.down_proj", "model.layers.23.mlp.experts.7.down_proj", "model.layers.23.mlp.experts.8.down_proj", "model.layers.23.mlp.experts.9.down_proj", "model.layers.23.mlp.experts.10.down_proj", "model.layers.23.mlp.experts.11.down_proj", "model.layers.23.mlp.experts.12.down_proj", "model.layers.23.mlp.experts.13.down_proj", "model.layers.23.mlp.experts.14.down_proj", "model.layers.23.mlp.experts.15.down_proj", "model.layers.23.mlp.experts.16.down_proj", "model.layers.23.mlp.experts.17.down_proj", "model.layers.23.mlp.experts.18.down_proj", "model.layers.23.mlp.experts.19.down_proj", "model.layers.23.mlp.experts.20.down_proj", "model.layers.23.mlp.experts.21.down_proj", "model.layers.23.mlp.experts.22.down_proj", "model.layers.23.mlp.experts.23.down_proj", "model.layers.23.mlp.experts.24.down_proj", "model.layers.23.mlp.experts.25.down_proj", "model.layers.23.mlp.experts.26.down_proj", "model.layers.23.mlp.experts.27.down_proj", "model.layers.23.mlp.experts.28.down_proj", "model.layers.23.mlp.experts.29.down_proj", "model.layers.23.mlp.experts.30.down_proj", "model.layers.23.mlp.experts.31.down_proj", "model.layers.23.mlp.experts.32.down_proj", "model.layers.23.mlp.experts.33.down_proj", "model.layers.23.mlp.experts.34.down_proj", "model.layers.23.mlp.experts.35.down_proj", "model.layers.23.mlp.experts.36.down_proj", "model.layers.23.mlp.experts.37.down_proj", "model.layers.23.mlp.experts.38.down_proj", "model.layers.23.mlp.experts.39.down_proj", "model.layers.23.mlp.experts.40.down_proj", "model.layers.23.mlp.experts.41.down_proj", "model.layers.23.mlp.experts.42.down_proj", "model.layers.23.mlp.experts.43.down_proj", "model.layers.23.mlp.experts.44.down_proj", "model.layers.23.mlp.experts.45.down_proj", "model.layers.23.mlp.experts.46.down_proj", "model.layers.23.mlp.experts.47.down_proj", "model.layers.23.mlp.experts.48.down_proj", "model.layers.23.mlp.experts.49.down_proj", "model.layers.23.mlp.experts.50.down_proj", "model.layers.23.mlp.experts.51.down_proj", "model.layers.23.mlp.experts.52.down_proj", "model.layers.23.mlp.experts.53.down_proj", "model.layers.23.mlp.experts.54.down_proj", "model.layers.23.mlp.experts.55.down_proj", "model.layers.23.mlp.experts.56.down_proj", "model.layers.23.mlp.experts.57.down_proj", "model.layers.23.mlp.experts.58.down_proj", "model.layers.23.mlp.experts.59.down_proj", "model.layers.23.mlp.experts.60.down_proj", "model.layers.23.mlp.experts.61.down_proj", "model.layers.23.mlp.experts.62.down_proj", "model.layers.23.mlp.experts.63.down_proj", "model.layers.23.mlp.experts.64.down_proj", "model.layers.23.mlp.experts.65.down_proj", "model.layers.23.mlp.experts.66.down_proj", "model.layers.23.mlp.experts.67.down_proj", "model.layers.23.mlp.experts.68.down_proj", "model.layers.23.mlp.experts.69.down_proj", "model.layers.23.mlp.experts.70.down_proj", "model.layers.23.mlp.experts.71.down_proj", "model.layers.23.mlp.experts.72.down_proj", "model.layers.23.mlp.experts.73.down_proj", "model.layers.23.mlp.experts.74.down_proj", "model.layers.23.mlp.experts.75.down_proj", "model.layers.23.mlp.experts.76.down_proj", "model.layers.23.mlp.experts.77.down_proj", "model.layers.23.mlp.experts.78.down_proj", "model.layers.23.mlp.experts.79.down_proj", "model.layers.23.mlp.experts.80.down_proj", "model.layers.23.mlp.experts.81.down_proj", "model.layers.23.mlp.experts.82.down_proj", "model.layers.23.mlp.experts.83.down_proj", "model.layers.23.mlp.experts.84.down_proj", "model.layers.23.mlp.experts.85.down_proj", "model.layers.23.mlp.experts.86.down_proj", "model.layers.23.mlp.experts.87.down_proj", "model.layers.23.mlp.experts.88.down_proj", "model.layers.23.mlp.experts.89.down_proj", "model.layers.23.mlp.experts.90.down_proj", "model.layers.23.mlp.experts.91.down_proj", "model.layers.23.mlp.experts.92.down_proj", "model.layers.23.mlp.experts.93.down_proj", "model.layers.23.mlp.experts.94.down_proj", "model.layers.23.mlp.experts.95.down_proj", "model.layers.23.mlp.experts.96.down_proj", "model.layers.23.mlp.experts.97.down_proj", "model.layers.23.mlp.experts.98.down_proj", "model.layers.23.mlp.experts.99.down_proj", "model.layers.23.mlp.experts.100.down_proj", "model.layers.23.mlp.experts.101.down_proj", "model.layers.23.mlp.experts.102.down_proj", "model.layers.23.mlp.experts.103.down_proj", "model.layers.23.mlp.experts.104.down_proj", "model.layers.23.mlp.experts.105.down_proj", "model.layers.23.mlp.experts.106.down_proj", "model.layers.23.mlp.experts.107.down_proj", "model.layers.23.mlp.experts.108.down_proj", "model.layers.23.mlp.experts.109.down_proj", "model.layers.23.mlp.experts.110.down_proj", "model.layers.23.mlp.experts.111.down_proj", "model.layers.23.mlp.experts.112.down_proj", "model.layers.23.mlp.experts.113.down_proj", "model.layers.23.mlp.experts.114.down_proj", "model.layers.23.mlp.experts.115.down_proj", "model.layers.23.mlp.experts.116.down_proj", "model.layers.23.mlp.experts.117.down_proj", "model.layers.23.mlp.experts.118.down_proj", "model.layers.23.mlp.experts.119.down_proj", "model.layers.23.mlp.experts.120.down_proj", "model.layers.23.mlp.experts.121.down_proj", "model.layers.23.mlp.experts.122.down_proj", "model.layers.23.mlp.experts.123.down_proj", "model.layers.23.mlp.experts.124.down_proj", "model.layers.23.mlp.experts.125.down_proj", "model.layers.23.mlp.experts.126.down_proj", "model.layers.23.mlp.experts.127.down_proj", "model.layers.23.mlp.experts.128.down_proj", "model.layers.23.mlp.experts.129.down_proj", "model.layers.23.mlp.experts.130.down_proj", "model.layers.23.mlp.experts.131.down_proj", "model.layers.23.mlp.experts.132.down_proj", "model.layers.23.mlp.experts.133.down_proj", "model.layers.23.mlp.experts.134.down_proj", "model.layers.23.mlp.experts.135.down_proj", "model.layers.23.mlp.experts.136.down_proj", "model.layers.23.mlp.experts.137.down_proj", "model.layers.23.mlp.experts.138.down_proj", "model.layers.23.mlp.experts.139.down_proj", "model.layers.23.mlp.experts.140.down_proj", "model.layers.23.mlp.experts.141.down_proj", "model.layers.23.mlp.experts.142.down_proj", "model.layers.23.mlp.experts.143.down_proj", "model.layers.23.mlp.experts.144.down_proj", "model.layers.23.mlp.experts.145.down_proj", "model.layers.23.mlp.experts.146.down_proj", "model.layers.23.mlp.experts.147.down_proj", "model.layers.23.mlp.experts.148.down_proj", "model.layers.23.mlp.experts.149.down_proj", "model.layers.23.mlp.experts.150.down_proj", "model.layers.23.mlp.experts.151.down_proj", "model.layers.23.mlp.experts.152.down_proj", "model.layers.23.mlp.experts.153.down_proj", "model.layers.23.mlp.experts.154.down_proj", "model.layers.23.mlp.experts.155.down_proj", "model.layers.23.mlp.experts.156.down_proj", "model.layers.23.mlp.experts.157.down_proj", "model.layers.23.mlp.experts.158.down_proj", "model.layers.23.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003431858494877815, "dbits": 1258291200 }, { "dkld": -0.0004642217420041672, "dbits": 2516582400 }, { "dkld": -0.0004371253773570172, "dbits": 3774873600 }, { "dkld": -0.0005072270520031508, "dbits": 6291456000 } ] }, { "idx": 141, "layers": [ "model.layers.24.self_attn.q_proj" ], "candidates": [ { "dkld": -7.117092609405795e-05, "dbits": 62914560 }, { "dkld": -0.0003264584578573787, "dbits": 125829120 }, { "dkld": -0.0003901662304997555, "dbits": 188743680 }, { "dkld": -0.00033979490399360657, "dbits": 314572800 } ] }, { "idx": 142, "layers": [ "model.layers.24.self_attn.k_proj", "model.layers.24.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0008159587159752846, "dbits": 10485760 }, { "dkld": -0.0009975654073059587, "dbits": 20971520 }, { "dkld": -0.0012171451002359474, "dbits": 31457280 }, { "dkld": -0.0011422866955399513, "dbits": 52428800 } ] }, { "idx": 143, "layers": [ "model.layers.24.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002259758301079301, "dbits": 62914560 }, { "dkld": -6.698500365019677e-05, "dbits": 125829120 }, { "dkld": -0.00036633946001529694, "dbits": 188743680 }, { "dkld": -0.00026661884039641537, "dbits": 314572800 } ] }, { "idx": 144, "layers": [ "model.layers.24.mlp.shared_experts.gate_proj", "model.layers.24.mlp.shared_experts.up_proj", "model.layers.24.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005781529471278191, "dbits": 23592960 }, { "dkld": -0.0004977061413228567, "dbits": 47185920 }, { "dkld": -0.0007272252812981633, "dbits": 70778880 }, { "dkld": -0.0006045632995665129, "dbits": 117964800 } ] }, { "idx": 145, "layers": [ "model.layers.24.mlp.experts.0.gate_proj", "model.layers.24.mlp.experts.1.gate_proj", "model.layers.24.mlp.experts.2.gate_proj", "model.layers.24.mlp.experts.3.gate_proj", "model.layers.24.mlp.experts.4.gate_proj", "model.layers.24.mlp.experts.5.gate_proj", "model.layers.24.mlp.experts.6.gate_proj", "model.layers.24.mlp.experts.7.gate_proj", "model.layers.24.mlp.experts.8.gate_proj", "model.layers.24.mlp.experts.9.gate_proj", "model.layers.24.mlp.experts.10.gate_proj", "model.layers.24.mlp.experts.11.gate_proj", "model.layers.24.mlp.experts.12.gate_proj", "model.layers.24.mlp.experts.13.gate_proj", "model.layers.24.mlp.experts.14.gate_proj", "model.layers.24.mlp.experts.15.gate_proj", "model.layers.24.mlp.experts.16.gate_proj", "model.layers.24.mlp.experts.17.gate_proj", "model.layers.24.mlp.experts.18.gate_proj", "model.layers.24.mlp.experts.19.gate_proj", "model.layers.24.mlp.experts.20.gate_proj", "model.layers.24.mlp.experts.21.gate_proj", "model.layers.24.mlp.experts.22.gate_proj", "model.layers.24.mlp.experts.23.gate_proj", "model.layers.24.mlp.experts.24.gate_proj", "model.layers.24.mlp.experts.25.gate_proj", "model.layers.24.mlp.experts.26.gate_proj", "model.layers.24.mlp.experts.27.gate_proj", "model.layers.24.mlp.experts.28.gate_proj", "model.layers.24.mlp.experts.29.gate_proj", "model.layers.24.mlp.experts.30.gate_proj", "model.layers.24.mlp.experts.31.gate_proj", "model.layers.24.mlp.experts.32.gate_proj", "model.layers.24.mlp.experts.33.gate_proj", "model.layers.24.mlp.experts.34.gate_proj", "model.layers.24.mlp.experts.35.gate_proj", "model.layers.24.mlp.experts.36.gate_proj", "model.layers.24.mlp.experts.37.gate_proj", "model.layers.24.mlp.experts.38.gate_proj", "model.layers.24.mlp.experts.39.gate_proj", "model.layers.24.mlp.experts.40.gate_proj", "model.layers.24.mlp.experts.41.gate_proj", "model.layers.24.mlp.experts.42.gate_proj", "model.layers.24.mlp.experts.43.gate_proj", "model.layers.24.mlp.experts.44.gate_proj", "model.layers.24.mlp.experts.45.gate_proj", "model.layers.24.mlp.experts.46.gate_proj", "model.layers.24.mlp.experts.47.gate_proj", "model.layers.24.mlp.experts.48.gate_proj", "model.layers.24.mlp.experts.49.gate_proj", "model.layers.24.mlp.experts.50.gate_proj", "model.layers.24.mlp.experts.51.gate_proj", "model.layers.24.mlp.experts.52.gate_proj", "model.layers.24.mlp.experts.53.gate_proj", "model.layers.24.mlp.experts.54.gate_proj", "model.layers.24.mlp.experts.55.gate_proj", "model.layers.24.mlp.experts.56.gate_proj", "model.layers.24.mlp.experts.57.gate_proj", "model.layers.24.mlp.experts.58.gate_proj", "model.layers.24.mlp.experts.59.gate_proj", "model.layers.24.mlp.experts.60.gate_proj", "model.layers.24.mlp.experts.61.gate_proj", "model.layers.24.mlp.experts.62.gate_proj", "model.layers.24.mlp.experts.63.gate_proj", "model.layers.24.mlp.experts.64.gate_proj", "model.layers.24.mlp.experts.65.gate_proj", "model.layers.24.mlp.experts.66.gate_proj", "model.layers.24.mlp.experts.67.gate_proj", "model.layers.24.mlp.experts.68.gate_proj", "model.layers.24.mlp.experts.69.gate_proj", "model.layers.24.mlp.experts.70.gate_proj", "model.layers.24.mlp.experts.71.gate_proj", "model.layers.24.mlp.experts.72.gate_proj", "model.layers.24.mlp.experts.73.gate_proj", "model.layers.24.mlp.experts.74.gate_proj", "model.layers.24.mlp.experts.75.gate_proj", "model.layers.24.mlp.experts.76.gate_proj", "model.layers.24.mlp.experts.77.gate_proj", "model.layers.24.mlp.experts.78.gate_proj", "model.layers.24.mlp.experts.79.gate_proj", "model.layers.24.mlp.experts.80.gate_proj", "model.layers.24.mlp.experts.81.gate_proj", "model.layers.24.mlp.experts.82.gate_proj", "model.layers.24.mlp.experts.83.gate_proj", "model.layers.24.mlp.experts.84.gate_proj", "model.layers.24.mlp.experts.85.gate_proj", "model.layers.24.mlp.experts.86.gate_proj", "model.layers.24.mlp.experts.87.gate_proj", "model.layers.24.mlp.experts.88.gate_proj", "model.layers.24.mlp.experts.89.gate_proj", "model.layers.24.mlp.experts.90.gate_proj", "model.layers.24.mlp.experts.91.gate_proj", "model.layers.24.mlp.experts.92.gate_proj", "model.layers.24.mlp.experts.93.gate_proj", "model.layers.24.mlp.experts.94.gate_proj", "model.layers.24.mlp.experts.95.gate_proj", "model.layers.24.mlp.experts.96.gate_proj", "model.layers.24.mlp.experts.97.gate_proj", "model.layers.24.mlp.experts.98.gate_proj", "model.layers.24.mlp.experts.99.gate_proj", "model.layers.24.mlp.experts.100.gate_proj", "model.layers.24.mlp.experts.101.gate_proj", "model.layers.24.mlp.experts.102.gate_proj", "model.layers.24.mlp.experts.103.gate_proj", "model.layers.24.mlp.experts.104.gate_proj", "model.layers.24.mlp.experts.105.gate_proj", "model.layers.24.mlp.experts.106.gate_proj", "model.layers.24.mlp.experts.107.gate_proj", "model.layers.24.mlp.experts.108.gate_proj", "model.layers.24.mlp.experts.109.gate_proj", "model.layers.24.mlp.experts.110.gate_proj", "model.layers.24.mlp.experts.111.gate_proj", "model.layers.24.mlp.experts.112.gate_proj", "model.layers.24.mlp.experts.113.gate_proj", "model.layers.24.mlp.experts.114.gate_proj", "model.layers.24.mlp.experts.115.gate_proj", "model.layers.24.mlp.experts.116.gate_proj", "model.layers.24.mlp.experts.117.gate_proj", "model.layers.24.mlp.experts.118.gate_proj", "model.layers.24.mlp.experts.119.gate_proj", "model.layers.24.mlp.experts.120.gate_proj", "model.layers.24.mlp.experts.121.gate_proj", "model.layers.24.mlp.experts.122.gate_proj", "model.layers.24.mlp.experts.123.gate_proj", "model.layers.24.mlp.experts.124.gate_proj", "model.layers.24.mlp.experts.125.gate_proj", "model.layers.24.mlp.experts.126.gate_proj", "model.layers.24.mlp.experts.127.gate_proj", "model.layers.24.mlp.experts.128.gate_proj", "model.layers.24.mlp.experts.129.gate_proj", "model.layers.24.mlp.experts.130.gate_proj", "model.layers.24.mlp.experts.131.gate_proj", "model.layers.24.mlp.experts.132.gate_proj", "model.layers.24.mlp.experts.133.gate_proj", "model.layers.24.mlp.experts.134.gate_proj", "model.layers.24.mlp.experts.135.gate_proj", "model.layers.24.mlp.experts.136.gate_proj", "model.layers.24.mlp.experts.137.gate_proj", "model.layers.24.mlp.experts.138.gate_proj", "model.layers.24.mlp.experts.139.gate_proj", "model.layers.24.mlp.experts.140.gate_proj", "model.layers.24.mlp.experts.141.gate_proj", "model.layers.24.mlp.experts.142.gate_proj", "model.layers.24.mlp.experts.143.gate_proj", "model.layers.24.mlp.experts.144.gate_proj", "model.layers.24.mlp.experts.145.gate_proj", "model.layers.24.mlp.experts.146.gate_proj", "model.layers.24.mlp.experts.147.gate_proj", "model.layers.24.mlp.experts.148.gate_proj", "model.layers.24.mlp.experts.149.gate_proj", "model.layers.24.mlp.experts.150.gate_proj", "model.layers.24.mlp.experts.151.gate_proj", "model.layers.24.mlp.experts.152.gate_proj", "model.layers.24.mlp.experts.153.gate_proj", "model.layers.24.mlp.experts.154.gate_proj", "model.layers.24.mlp.experts.155.gate_proj", "model.layers.24.mlp.experts.156.gate_proj", "model.layers.24.mlp.experts.157.gate_proj", "model.layers.24.mlp.experts.158.gate_proj", "model.layers.24.mlp.experts.159.gate_proj", "model.layers.24.mlp.experts.0.up_proj", "model.layers.24.mlp.experts.1.up_proj", "model.layers.24.mlp.experts.2.up_proj", "model.layers.24.mlp.experts.3.up_proj", "model.layers.24.mlp.experts.4.up_proj", "model.layers.24.mlp.experts.5.up_proj", "model.layers.24.mlp.experts.6.up_proj", "model.layers.24.mlp.experts.7.up_proj", "model.layers.24.mlp.experts.8.up_proj", "model.layers.24.mlp.experts.9.up_proj", "model.layers.24.mlp.experts.10.up_proj", "model.layers.24.mlp.experts.11.up_proj", "model.layers.24.mlp.experts.12.up_proj", "model.layers.24.mlp.experts.13.up_proj", "model.layers.24.mlp.experts.14.up_proj", "model.layers.24.mlp.experts.15.up_proj", "model.layers.24.mlp.experts.16.up_proj", "model.layers.24.mlp.experts.17.up_proj", "model.layers.24.mlp.experts.18.up_proj", "model.layers.24.mlp.experts.19.up_proj", "model.layers.24.mlp.experts.20.up_proj", "model.layers.24.mlp.experts.21.up_proj", "model.layers.24.mlp.experts.22.up_proj", "model.layers.24.mlp.experts.23.up_proj", "model.layers.24.mlp.experts.24.up_proj", "model.layers.24.mlp.experts.25.up_proj", "model.layers.24.mlp.experts.26.up_proj", "model.layers.24.mlp.experts.27.up_proj", "model.layers.24.mlp.experts.28.up_proj", "model.layers.24.mlp.experts.29.up_proj", "model.layers.24.mlp.experts.30.up_proj", "model.layers.24.mlp.experts.31.up_proj", "model.layers.24.mlp.experts.32.up_proj", "model.layers.24.mlp.experts.33.up_proj", "model.layers.24.mlp.experts.34.up_proj", "model.layers.24.mlp.experts.35.up_proj", "model.layers.24.mlp.experts.36.up_proj", "model.layers.24.mlp.experts.37.up_proj", "model.layers.24.mlp.experts.38.up_proj", "model.layers.24.mlp.experts.39.up_proj", "model.layers.24.mlp.experts.40.up_proj", "model.layers.24.mlp.experts.41.up_proj", "model.layers.24.mlp.experts.42.up_proj", "model.layers.24.mlp.experts.43.up_proj", "model.layers.24.mlp.experts.44.up_proj", "model.layers.24.mlp.experts.45.up_proj", "model.layers.24.mlp.experts.46.up_proj", "model.layers.24.mlp.experts.47.up_proj", "model.layers.24.mlp.experts.48.up_proj", "model.layers.24.mlp.experts.49.up_proj", "model.layers.24.mlp.experts.50.up_proj", "model.layers.24.mlp.experts.51.up_proj", "model.layers.24.mlp.experts.52.up_proj", "model.layers.24.mlp.experts.53.up_proj", "model.layers.24.mlp.experts.54.up_proj", "model.layers.24.mlp.experts.55.up_proj", "model.layers.24.mlp.experts.56.up_proj", "model.layers.24.mlp.experts.57.up_proj", "model.layers.24.mlp.experts.58.up_proj", "model.layers.24.mlp.experts.59.up_proj", "model.layers.24.mlp.experts.60.up_proj", "model.layers.24.mlp.experts.61.up_proj", "model.layers.24.mlp.experts.62.up_proj", "model.layers.24.mlp.experts.63.up_proj", "model.layers.24.mlp.experts.64.up_proj", "model.layers.24.mlp.experts.65.up_proj", "model.layers.24.mlp.experts.66.up_proj", "model.layers.24.mlp.experts.67.up_proj", "model.layers.24.mlp.experts.68.up_proj", "model.layers.24.mlp.experts.69.up_proj", "model.layers.24.mlp.experts.70.up_proj", "model.layers.24.mlp.experts.71.up_proj", "model.layers.24.mlp.experts.72.up_proj", "model.layers.24.mlp.experts.73.up_proj", "model.layers.24.mlp.experts.74.up_proj", "model.layers.24.mlp.experts.75.up_proj", "model.layers.24.mlp.experts.76.up_proj", "model.layers.24.mlp.experts.77.up_proj", "model.layers.24.mlp.experts.78.up_proj", "model.layers.24.mlp.experts.79.up_proj", "model.layers.24.mlp.experts.80.up_proj", "model.layers.24.mlp.experts.81.up_proj", "model.layers.24.mlp.experts.82.up_proj", "model.layers.24.mlp.experts.83.up_proj", "model.layers.24.mlp.experts.84.up_proj", "model.layers.24.mlp.experts.85.up_proj", "model.layers.24.mlp.experts.86.up_proj", "model.layers.24.mlp.experts.87.up_proj", "model.layers.24.mlp.experts.88.up_proj", "model.layers.24.mlp.experts.89.up_proj", "model.layers.24.mlp.experts.90.up_proj", "model.layers.24.mlp.experts.91.up_proj", "model.layers.24.mlp.experts.92.up_proj", "model.layers.24.mlp.experts.93.up_proj", "model.layers.24.mlp.experts.94.up_proj", "model.layers.24.mlp.experts.95.up_proj", "model.layers.24.mlp.experts.96.up_proj", "model.layers.24.mlp.experts.97.up_proj", "model.layers.24.mlp.experts.98.up_proj", "model.layers.24.mlp.experts.99.up_proj", "model.layers.24.mlp.experts.100.up_proj", "model.layers.24.mlp.experts.101.up_proj", "model.layers.24.mlp.experts.102.up_proj", "model.layers.24.mlp.experts.103.up_proj", "model.layers.24.mlp.experts.104.up_proj", "model.layers.24.mlp.experts.105.up_proj", "model.layers.24.mlp.experts.106.up_proj", "model.layers.24.mlp.experts.107.up_proj", "model.layers.24.mlp.experts.108.up_proj", "model.layers.24.mlp.experts.109.up_proj", "model.layers.24.mlp.experts.110.up_proj", "model.layers.24.mlp.experts.111.up_proj", "model.layers.24.mlp.experts.112.up_proj", "model.layers.24.mlp.experts.113.up_proj", "model.layers.24.mlp.experts.114.up_proj", "model.layers.24.mlp.experts.115.up_proj", "model.layers.24.mlp.experts.116.up_proj", "model.layers.24.mlp.experts.117.up_proj", "model.layers.24.mlp.experts.118.up_proj", "model.layers.24.mlp.experts.119.up_proj", "model.layers.24.mlp.experts.120.up_proj", "model.layers.24.mlp.experts.121.up_proj", "model.layers.24.mlp.experts.122.up_proj", "model.layers.24.mlp.experts.123.up_proj", "model.layers.24.mlp.experts.124.up_proj", "model.layers.24.mlp.experts.125.up_proj", "model.layers.24.mlp.experts.126.up_proj", "model.layers.24.mlp.experts.127.up_proj", "model.layers.24.mlp.experts.128.up_proj", "model.layers.24.mlp.experts.129.up_proj", "model.layers.24.mlp.experts.130.up_proj", "model.layers.24.mlp.experts.131.up_proj", "model.layers.24.mlp.experts.132.up_proj", "model.layers.24.mlp.experts.133.up_proj", "model.layers.24.mlp.experts.134.up_proj", "model.layers.24.mlp.experts.135.up_proj", "model.layers.24.mlp.experts.136.up_proj", "model.layers.24.mlp.experts.137.up_proj", "model.layers.24.mlp.experts.138.up_proj", "model.layers.24.mlp.experts.139.up_proj", "model.layers.24.mlp.experts.140.up_proj", "model.layers.24.mlp.experts.141.up_proj", "model.layers.24.mlp.experts.142.up_proj", "model.layers.24.mlp.experts.143.up_proj", "model.layers.24.mlp.experts.144.up_proj", "model.layers.24.mlp.experts.145.up_proj", "model.layers.24.mlp.experts.146.up_proj", "model.layers.24.mlp.experts.147.up_proj", "model.layers.24.mlp.experts.148.up_proj", "model.layers.24.mlp.experts.149.up_proj", "model.layers.24.mlp.experts.150.up_proj", "model.layers.24.mlp.experts.151.up_proj", "model.layers.24.mlp.experts.152.up_proj", "model.layers.24.mlp.experts.153.up_proj", "model.layers.24.mlp.experts.154.up_proj", "model.layers.24.mlp.experts.155.up_proj", "model.layers.24.mlp.experts.156.up_proj", "model.layers.24.mlp.experts.157.up_proj", "model.layers.24.mlp.experts.158.up_proj", "model.layers.24.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0010521221905946815, "dbits": 2516582400 }, { "dkld": -0.001059680711477995, "dbits": 5033164800 }, { "dkld": -0.0010482370853424072, "dbits": 7549747200 }, { "dkld": -0.0010239809751510648, "dbits": 12582912000 } ] }, { "idx": 146, "layers": [ "model.layers.24.mlp.experts.0.down_proj", "model.layers.24.mlp.experts.1.down_proj", "model.layers.24.mlp.experts.2.down_proj", "model.layers.24.mlp.experts.3.down_proj", "model.layers.24.mlp.experts.4.down_proj", "model.layers.24.mlp.experts.5.down_proj", "model.layers.24.mlp.experts.6.down_proj", "model.layers.24.mlp.experts.7.down_proj", "model.layers.24.mlp.experts.8.down_proj", "model.layers.24.mlp.experts.9.down_proj", "model.layers.24.mlp.experts.10.down_proj", "model.layers.24.mlp.experts.11.down_proj", "model.layers.24.mlp.experts.12.down_proj", "model.layers.24.mlp.experts.13.down_proj", "model.layers.24.mlp.experts.14.down_proj", "model.layers.24.mlp.experts.15.down_proj", "model.layers.24.mlp.experts.16.down_proj", "model.layers.24.mlp.experts.17.down_proj", "model.layers.24.mlp.experts.18.down_proj", "model.layers.24.mlp.experts.19.down_proj", "model.layers.24.mlp.experts.20.down_proj", "model.layers.24.mlp.experts.21.down_proj", "model.layers.24.mlp.experts.22.down_proj", "model.layers.24.mlp.experts.23.down_proj", "model.layers.24.mlp.experts.24.down_proj", "model.layers.24.mlp.experts.25.down_proj", "model.layers.24.mlp.experts.26.down_proj", "model.layers.24.mlp.experts.27.down_proj", "model.layers.24.mlp.experts.28.down_proj", "model.layers.24.mlp.experts.29.down_proj", "model.layers.24.mlp.experts.30.down_proj", "model.layers.24.mlp.experts.31.down_proj", "model.layers.24.mlp.experts.32.down_proj", "model.layers.24.mlp.experts.33.down_proj", "model.layers.24.mlp.experts.34.down_proj", "model.layers.24.mlp.experts.35.down_proj", "model.layers.24.mlp.experts.36.down_proj", "model.layers.24.mlp.experts.37.down_proj", "model.layers.24.mlp.experts.38.down_proj", "model.layers.24.mlp.experts.39.down_proj", "model.layers.24.mlp.experts.40.down_proj", "model.layers.24.mlp.experts.41.down_proj", "model.layers.24.mlp.experts.42.down_proj", "model.layers.24.mlp.experts.43.down_proj", "model.layers.24.mlp.experts.44.down_proj", "model.layers.24.mlp.experts.45.down_proj", "model.layers.24.mlp.experts.46.down_proj", "model.layers.24.mlp.experts.47.down_proj", "model.layers.24.mlp.experts.48.down_proj", "model.layers.24.mlp.experts.49.down_proj", "model.layers.24.mlp.experts.50.down_proj", "model.layers.24.mlp.experts.51.down_proj", "model.layers.24.mlp.experts.52.down_proj", "model.layers.24.mlp.experts.53.down_proj", "model.layers.24.mlp.experts.54.down_proj", "model.layers.24.mlp.experts.55.down_proj", "model.layers.24.mlp.experts.56.down_proj", "model.layers.24.mlp.experts.57.down_proj", "model.layers.24.mlp.experts.58.down_proj", "model.layers.24.mlp.experts.59.down_proj", "model.layers.24.mlp.experts.60.down_proj", "model.layers.24.mlp.experts.61.down_proj", "model.layers.24.mlp.experts.62.down_proj", "model.layers.24.mlp.experts.63.down_proj", "model.layers.24.mlp.experts.64.down_proj", "model.layers.24.mlp.experts.65.down_proj", "model.layers.24.mlp.experts.66.down_proj", "model.layers.24.mlp.experts.67.down_proj", "model.layers.24.mlp.experts.68.down_proj", "model.layers.24.mlp.experts.69.down_proj", "model.layers.24.mlp.experts.70.down_proj", "model.layers.24.mlp.experts.71.down_proj", "model.layers.24.mlp.experts.72.down_proj", "model.layers.24.mlp.experts.73.down_proj", "model.layers.24.mlp.experts.74.down_proj", "model.layers.24.mlp.experts.75.down_proj", "model.layers.24.mlp.experts.76.down_proj", "model.layers.24.mlp.experts.77.down_proj", "model.layers.24.mlp.experts.78.down_proj", "model.layers.24.mlp.experts.79.down_proj", "model.layers.24.mlp.experts.80.down_proj", "model.layers.24.mlp.experts.81.down_proj", "model.layers.24.mlp.experts.82.down_proj", "model.layers.24.mlp.experts.83.down_proj", "model.layers.24.mlp.experts.84.down_proj", "model.layers.24.mlp.experts.85.down_proj", "model.layers.24.mlp.experts.86.down_proj", "model.layers.24.mlp.experts.87.down_proj", "model.layers.24.mlp.experts.88.down_proj", "model.layers.24.mlp.experts.89.down_proj", "model.layers.24.mlp.experts.90.down_proj", "model.layers.24.mlp.experts.91.down_proj", "model.layers.24.mlp.experts.92.down_proj", "model.layers.24.mlp.experts.93.down_proj", "model.layers.24.mlp.experts.94.down_proj", "model.layers.24.mlp.experts.95.down_proj", "model.layers.24.mlp.experts.96.down_proj", "model.layers.24.mlp.experts.97.down_proj", "model.layers.24.mlp.experts.98.down_proj", "model.layers.24.mlp.experts.99.down_proj", "model.layers.24.mlp.experts.100.down_proj", "model.layers.24.mlp.experts.101.down_proj", "model.layers.24.mlp.experts.102.down_proj", "model.layers.24.mlp.experts.103.down_proj", "model.layers.24.mlp.experts.104.down_proj", "model.layers.24.mlp.experts.105.down_proj", "model.layers.24.mlp.experts.106.down_proj", "model.layers.24.mlp.experts.107.down_proj", "model.layers.24.mlp.experts.108.down_proj", "model.layers.24.mlp.experts.109.down_proj", "model.layers.24.mlp.experts.110.down_proj", "model.layers.24.mlp.experts.111.down_proj", "model.layers.24.mlp.experts.112.down_proj", "model.layers.24.mlp.experts.113.down_proj", "model.layers.24.mlp.experts.114.down_proj", "model.layers.24.mlp.experts.115.down_proj", "model.layers.24.mlp.experts.116.down_proj", "model.layers.24.mlp.experts.117.down_proj", "model.layers.24.mlp.experts.118.down_proj", "model.layers.24.mlp.experts.119.down_proj", "model.layers.24.mlp.experts.120.down_proj", "model.layers.24.mlp.experts.121.down_proj", "model.layers.24.mlp.experts.122.down_proj", "model.layers.24.mlp.experts.123.down_proj", "model.layers.24.mlp.experts.124.down_proj", "model.layers.24.mlp.experts.125.down_proj", "model.layers.24.mlp.experts.126.down_proj", "model.layers.24.mlp.experts.127.down_proj", "model.layers.24.mlp.experts.128.down_proj", "model.layers.24.mlp.experts.129.down_proj", "model.layers.24.mlp.experts.130.down_proj", "model.layers.24.mlp.experts.131.down_proj", "model.layers.24.mlp.experts.132.down_proj", "model.layers.24.mlp.experts.133.down_proj", "model.layers.24.mlp.experts.134.down_proj", "model.layers.24.mlp.experts.135.down_proj", "model.layers.24.mlp.experts.136.down_proj", "model.layers.24.mlp.experts.137.down_proj", "model.layers.24.mlp.experts.138.down_proj", "model.layers.24.mlp.experts.139.down_proj", "model.layers.24.mlp.experts.140.down_proj", "model.layers.24.mlp.experts.141.down_proj", "model.layers.24.mlp.experts.142.down_proj", "model.layers.24.mlp.experts.143.down_proj", "model.layers.24.mlp.experts.144.down_proj", "model.layers.24.mlp.experts.145.down_proj", "model.layers.24.mlp.experts.146.down_proj", "model.layers.24.mlp.experts.147.down_proj", "model.layers.24.mlp.experts.148.down_proj", "model.layers.24.mlp.experts.149.down_proj", "model.layers.24.mlp.experts.150.down_proj", "model.layers.24.mlp.experts.151.down_proj", "model.layers.24.mlp.experts.152.down_proj", "model.layers.24.mlp.experts.153.down_proj", "model.layers.24.mlp.experts.154.down_proj", "model.layers.24.mlp.experts.155.down_proj", "model.layers.24.mlp.experts.156.down_proj", "model.layers.24.mlp.experts.157.down_proj", "model.layers.24.mlp.experts.158.down_proj", "model.layers.24.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003643111325800502, "dbits": 1258291200 }, { "dkld": -0.0004667428322136513, "dbits": 2516582400 }, { "dkld": -0.00045739952474832535, "dbits": 3774873600 }, { "dkld": -0.0005707568489015102, "dbits": 6291456000 } ] }, { "idx": 147, "layers": [ "model.layers.25.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0006501811556517967, "dbits": 62914560 }, { "dkld": 0.000636821705847973, "dbits": 125829120 }, { "dkld": 0.0007082451134920037, "dbits": 188743680 }, { "dkld": 0.0007074345834553186, "dbits": 314572800 } ] }, { "idx": 148, "layers": [ "model.layers.25.self_attn.k_proj", "model.layers.25.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00040982356294988476, "dbits": 10485760 }, { "dkld": 0.00035473657771945, "dbits": 20971520 }, { "dkld": 0.00044207852333782993, "dbits": 31457280 }, { "dkld": 0.0003607153892516979, "dbits": 52428800 } ] }, { "idx": 149, "layers": [ "model.layers.25.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0008909258991479929, "dbits": 62914560 }, { "dkld": -0.0008623466826975373, "dbits": 125829120 }, { "dkld": -0.000742810498923066, "dbits": 188743680 }, { "dkld": -0.0008400168269872693, "dbits": 314572800 } ] }, { "idx": 150, "layers": [ "model.layers.25.mlp.shared_experts.gate_proj", "model.layers.25.mlp.shared_experts.up_proj", "model.layers.25.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0011522994376719081, "dbits": 23592960 }, { "dkld": -0.0008393471129238689, "dbits": 47185920 }, { "dkld": -0.000933502335101366, "dbits": 70778880 }, { "dkld": -0.0008351070806384142, "dbits": 117964800 } ] }, { "idx": 151, "layers": [ "model.layers.25.mlp.experts.0.gate_proj", "model.layers.25.mlp.experts.1.gate_proj", "model.layers.25.mlp.experts.2.gate_proj", "model.layers.25.mlp.experts.3.gate_proj", "model.layers.25.mlp.experts.4.gate_proj", "model.layers.25.mlp.experts.5.gate_proj", "model.layers.25.mlp.experts.6.gate_proj", "model.layers.25.mlp.experts.7.gate_proj", "model.layers.25.mlp.experts.8.gate_proj", "model.layers.25.mlp.experts.9.gate_proj", "model.layers.25.mlp.experts.10.gate_proj", "model.layers.25.mlp.experts.11.gate_proj", "model.layers.25.mlp.experts.12.gate_proj", "model.layers.25.mlp.experts.13.gate_proj", "model.layers.25.mlp.experts.14.gate_proj", "model.layers.25.mlp.experts.15.gate_proj", "model.layers.25.mlp.experts.16.gate_proj", "model.layers.25.mlp.experts.17.gate_proj", "model.layers.25.mlp.experts.18.gate_proj", "model.layers.25.mlp.experts.19.gate_proj", "model.layers.25.mlp.experts.20.gate_proj", "model.layers.25.mlp.experts.21.gate_proj", "model.layers.25.mlp.experts.22.gate_proj", "model.layers.25.mlp.experts.23.gate_proj", "model.layers.25.mlp.experts.24.gate_proj", "model.layers.25.mlp.experts.25.gate_proj", "model.layers.25.mlp.experts.26.gate_proj", "model.layers.25.mlp.experts.27.gate_proj", "model.layers.25.mlp.experts.28.gate_proj", "model.layers.25.mlp.experts.29.gate_proj", "model.layers.25.mlp.experts.30.gate_proj", "model.layers.25.mlp.experts.31.gate_proj", "model.layers.25.mlp.experts.32.gate_proj", "model.layers.25.mlp.experts.33.gate_proj", "model.layers.25.mlp.experts.34.gate_proj", "model.layers.25.mlp.experts.35.gate_proj", "model.layers.25.mlp.experts.36.gate_proj", "model.layers.25.mlp.experts.37.gate_proj", "model.layers.25.mlp.experts.38.gate_proj", "model.layers.25.mlp.experts.39.gate_proj", "model.layers.25.mlp.experts.40.gate_proj", "model.layers.25.mlp.experts.41.gate_proj", "model.layers.25.mlp.experts.42.gate_proj", "model.layers.25.mlp.experts.43.gate_proj", "model.layers.25.mlp.experts.44.gate_proj", "model.layers.25.mlp.experts.45.gate_proj", "model.layers.25.mlp.experts.46.gate_proj", "model.layers.25.mlp.experts.47.gate_proj", "model.layers.25.mlp.experts.48.gate_proj", "model.layers.25.mlp.experts.49.gate_proj", "model.layers.25.mlp.experts.50.gate_proj", "model.layers.25.mlp.experts.51.gate_proj", "model.layers.25.mlp.experts.52.gate_proj", "model.layers.25.mlp.experts.53.gate_proj", "model.layers.25.mlp.experts.54.gate_proj", "model.layers.25.mlp.experts.55.gate_proj", "model.layers.25.mlp.experts.56.gate_proj", "model.layers.25.mlp.experts.57.gate_proj", "model.layers.25.mlp.experts.58.gate_proj", "model.layers.25.mlp.experts.59.gate_proj", "model.layers.25.mlp.experts.60.gate_proj", "model.layers.25.mlp.experts.61.gate_proj", "model.layers.25.mlp.experts.62.gate_proj", "model.layers.25.mlp.experts.63.gate_proj", "model.layers.25.mlp.experts.64.gate_proj", "model.layers.25.mlp.experts.65.gate_proj", "model.layers.25.mlp.experts.66.gate_proj", "model.layers.25.mlp.experts.67.gate_proj", "model.layers.25.mlp.experts.68.gate_proj", "model.layers.25.mlp.experts.69.gate_proj", "model.layers.25.mlp.experts.70.gate_proj", "model.layers.25.mlp.experts.71.gate_proj", "model.layers.25.mlp.experts.72.gate_proj", "model.layers.25.mlp.experts.73.gate_proj", "model.layers.25.mlp.experts.74.gate_proj", "model.layers.25.mlp.experts.75.gate_proj", "model.layers.25.mlp.experts.76.gate_proj", "model.layers.25.mlp.experts.77.gate_proj", "model.layers.25.mlp.experts.78.gate_proj", "model.layers.25.mlp.experts.79.gate_proj", "model.layers.25.mlp.experts.80.gate_proj", "model.layers.25.mlp.experts.81.gate_proj", "model.layers.25.mlp.experts.82.gate_proj", "model.layers.25.mlp.experts.83.gate_proj", "model.layers.25.mlp.experts.84.gate_proj", "model.layers.25.mlp.experts.85.gate_proj", "model.layers.25.mlp.experts.86.gate_proj", "model.layers.25.mlp.experts.87.gate_proj", "model.layers.25.mlp.experts.88.gate_proj", "model.layers.25.mlp.experts.89.gate_proj", "model.layers.25.mlp.experts.90.gate_proj", "model.layers.25.mlp.experts.91.gate_proj", "model.layers.25.mlp.experts.92.gate_proj", "model.layers.25.mlp.experts.93.gate_proj", "model.layers.25.mlp.experts.94.gate_proj", "model.layers.25.mlp.experts.95.gate_proj", "model.layers.25.mlp.experts.96.gate_proj", "model.layers.25.mlp.experts.97.gate_proj", "model.layers.25.mlp.experts.98.gate_proj", "model.layers.25.mlp.experts.99.gate_proj", "model.layers.25.mlp.experts.100.gate_proj", "model.layers.25.mlp.experts.101.gate_proj", "model.layers.25.mlp.experts.102.gate_proj", "model.layers.25.mlp.experts.103.gate_proj", "model.layers.25.mlp.experts.104.gate_proj", "model.layers.25.mlp.experts.105.gate_proj", "model.layers.25.mlp.experts.106.gate_proj", "model.layers.25.mlp.experts.107.gate_proj", "model.layers.25.mlp.experts.108.gate_proj", "model.layers.25.mlp.experts.109.gate_proj", "model.layers.25.mlp.experts.110.gate_proj", "model.layers.25.mlp.experts.111.gate_proj", "model.layers.25.mlp.experts.112.gate_proj", "model.layers.25.mlp.experts.113.gate_proj", "model.layers.25.mlp.experts.114.gate_proj", "model.layers.25.mlp.experts.115.gate_proj", "model.layers.25.mlp.experts.116.gate_proj", "model.layers.25.mlp.experts.117.gate_proj", "model.layers.25.mlp.experts.118.gate_proj", "model.layers.25.mlp.experts.119.gate_proj", "model.layers.25.mlp.experts.120.gate_proj", "model.layers.25.mlp.experts.121.gate_proj", "model.layers.25.mlp.experts.122.gate_proj", "model.layers.25.mlp.experts.123.gate_proj", "model.layers.25.mlp.experts.124.gate_proj", "model.layers.25.mlp.experts.125.gate_proj", "model.layers.25.mlp.experts.126.gate_proj", "model.layers.25.mlp.experts.127.gate_proj", "model.layers.25.mlp.experts.128.gate_proj", "model.layers.25.mlp.experts.129.gate_proj", "model.layers.25.mlp.experts.130.gate_proj", "model.layers.25.mlp.experts.131.gate_proj", "model.layers.25.mlp.experts.132.gate_proj", "model.layers.25.mlp.experts.133.gate_proj", "model.layers.25.mlp.experts.134.gate_proj", "model.layers.25.mlp.experts.135.gate_proj", "model.layers.25.mlp.experts.136.gate_proj", "model.layers.25.mlp.experts.137.gate_proj", "model.layers.25.mlp.experts.138.gate_proj", "model.layers.25.mlp.experts.139.gate_proj", "model.layers.25.mlp.experts.140.gate_proj", "model.layers.25.mlp.experts.141.gate_proj", "model.layers.25.mlp.experts.142.gate_proj", "model.layers.25.mlp.experts.143.gate_proj", "model.layers.25.mlp.experts.144.gate_proj", "model.layers.25.mlp.experts.145.gate_proj", "model.layers.25.mlp.experts.146.gate_proj", "model.layers.25.mlp.experts.147.gate_proj", "model.layers.25.mlp.experts.148.gate_proj", "model.layers.25.mlp.experts.149.gate_proj", "model.layers.25.mlp.experts.150.gate_proj", "model.layers.25.mlp.experts.151.gate_proj", "model.layers.25.mlp.experts.152.gate_proj", "model.layers.25.mlp.experts.153.gate_proj", "model.layers.25.mlp.experts.154.gate_proj", "model.layers.25.mlp.experts.155.gate_proj", "model.layers.25.mlp.experts.156.gate_proj", "model.layers.25.mlp.experts.157.gate_proj", "model.layers.25.mlp.experts.158.gate_proj", "model.layers.25.mlp.experts.159.gate_proj", "model.layers.25.mlp.experts.0.up_proj", "model.layers.25.mlp.experts.1.up_proj", "model.layers.25.mlp.experts.2.up_proj", "model.layers.25.mlp.experts.3.up_proj", "model.layers.25.mlp.experts.4.up_proj", "model.layers.25.mlp.experts.5.up_proj", "model.layers.25.mlp.experts.6.up_proj", "model.layers.25.mlp.experts.7.up_proj", "model.layers.25.mlp.experts.8.up_proj", "model.layers.25.mlp.experts.9.up_proj", "model.layers.25.mlp.experts.10.up_proj", "model.layers.25.mlp.experts.11.up_proj", "model.layers.25.mlp.experts.12.up_proj", "model.layers.25.mlp.experts.13.up_proj", "model.layers.25.mlp.experts.14.up_proj", "model.layers.25.mlp.experts.15.up_proj", "model.layers.25.mlp.experts.16.up_proj", "model.layers.25.mlp.experts.17.up_proj", "model.layers.25.mlp.experts.18.up_proj", "model.layers.25.mlp.experts.19.up_proj", "model.layers.25.mlp.experts.20.up_proj", "model.layers.25.mlp.experts.21.up_proj", "model.layers.25.mlp.experts.22.up_proj", "model.layers.25.mlp.experts.23.up_proj", "model.layers.25.mlp.experts.24.up_proj", "model.layers.25.mlp.experts.25.up_proj", "model.layers.25.mlp.experts.26.up_proj", "model.layers.25.mlp.experts.27.up_proj", "model.layers.25.mlp.experts.28.up_proj", "model.layers.25.mlp.experts.29.up_proj", "model.layers.25.mlp.experts.30.up_proj", "model.layers.25.mlp.experts.31.up_proj", "model.layers.25.mlp.experts.32.up_proj", "model.layers.25.mlp.experts.33.up_proj", "model.layers.25.mlp.experts.34.up_proj", "model.layers.25.mlp.experts.35.up_proj", "model.layers.25.mlp.experts.36.up_proj", "model.layers.25.mlp.experts.37.up_proj", "model.layers.25.mlp.experts.38.up_proj", "model.layers.25.mlp.experts.39.up_proj", "model.layers.25.mlp.experts.40.up_proj", "model.layers.25.mlp.experts.41.up_proj", "model.layers.25.mlp.experts.42.up_proj", "model.layers.25.mlp.experts.43.up_proj", "model.layers.25.mlp.experts.44.up_proj", "model.layers.25.mlp.experts.45.up_proj", "model.layers.25.mlp.experts.46.up_proj", "model.layers.25.mlp.experts.47.up_proj", "model.layers.25.mlp.experts.48.up_proj", "model.layers.25.mlp.experts.49.up_proj", "model.layers.25.mlp.experts.50.up_proj", "model.layers.25.mlp.experts.51.up_proj", "model.layers.25.mlp.experts.52.up_proj", "model.layers.25.mlp.experts.53.up_proj", "model.layers.25.mlp.experts.54.up_proj", "model.layers.25.mlp.experts.55.up_proj", "model.layers.25.mlp.experts.56.up_proj", "model.layers.25.mlp.experts.57.up_proj", "model.layers.25.mlp.experts.58.up_proj", "model.layers.25.mlp.experts.59.up_proj", "model.layers.25.mlp.experts.60.up_proj", "model.layers.25.mlp.experts.61.up_proj", "model.layers.25.mlp.experts.62.up_proj", "model.layers.25.mlp.experts.63.up_proj", "model.layers.25.mlp.experts.64.up_proj", "model.layers.25.mlp.experts.65.up_proj", "model.layers.25.mlp.experts.66.up_proj", "model.layers.25.mlp.experts.67.up_proj", "model.layers.25.mlp.experts.68.up_proj", "model.layers.25.mlp.experts.69.up_proj", "model.layers.25.mlp.experts.70.up_proj", "model.layers.25.mlp.experts.71.up_proj", "model.layers.25.mlp.experts.72.up_proj", "model.layers.25.mlp.experts.73.up_proj", "model.layers.25.mlp.experts.74.up_proj", "model.layers.25.mlp.experts.75.up_proj", "model.layers.25.mlp.experts.76.up_proj", "model.layers.25.mlp.experts.77.up_proj", "model.layers.25.mlp.experts.78.up_proj", "model.layers.25.mlp.experts.79.up_proj", "model.layers.25.mlp.experts.80.up_proj", "model.layers.25.mlp.experts.81.up_proj", "model.layers.25.mlp.experts.82.up_proj", "model.layers.25.mlp.experts.83.up_proj", "model.layers.25.mlp.experts.84.up_proj", "model.layers.25.mlp.experts.85.up_proj", "model.layers.25.mlp.experts.86.up_proj", "model.layers.25.mlp.experts.87.up_proj", "model.layers.25.mlp.experts.88.up_proj", "model.layers.25.mlp.experts.89.up_proj", "model.layers.25.mlp.experts.90.up_proj", "model.layers.25.mlp.experts.91.up_proj", "model.layers.25.mlp.experts.92.up_proj", "model.layers.25.mlp.experts.93.up_proj", "model.layers.25.mlp.experts.94.up_proj", "model.layers.25.mlp.experts.95.up_proj", "model.layers.25.mlp.experts.96.up_proj", "model.layers.25.mlp.experts.97.up_proj", "model.layers.25.mlp.experts.98.up_proj", "model.layers.25.mlp.experts.99.up_proj", "model.layers.25.mlp.experts.100.up_proj", "model.layers.25.mlp.experts.101.up_proj", "model.layers.25.mlp.experts.102.up_proj", "model.layers.25.mlp.experts.103.up_proj", "model.layers.25.mlp.experts.104.up_proj", "model.layers.25.mlp.experts.105.up_proj", "model.layers.25.mlp.experts.106.up_proj", "model.layers.25.mlp.experts.107.up_proj", "model.layers.25.mlp.experts.108.up_proj", "model.layers.25.mlp.experts.109.up_proj", "model.layers.25.mlp.experts.110.up_proj", "model.layers.25.mlp.experts.111.up_proj", "model.layers.25.mlp.experts.112.up_proj", "model.layers.25.mlp.experts.113.up_proj", "model.layers.25.mlp.experts.114.up_proj", "model.layers.25.mlp.experts.115.up_proj", "model.layers.25.mlp.experts.116.up_proj", "model.layers.25.mlp.experts.117.up_proj", "model.layers.25.mlp.experts.118.up_proj", "model.layers.25.mlp.experts.119.up_proj", "model.layers.25.mlp.experts.120.up_proj", "model.layers.25.mlp.experts.121.up_proj", "model.layers.25.mlp.experts.122.up_proj", "model.layers.25.mlp.experts.123.up_proj", "model.layers.25.mlp.experts.124.up_proj", "model.layers.25.mlp.experts.125.up_proj", "model.layers.25.mlp.experts.126.up_proj", "model.layers.25.mlp.experts.127.up_proj", "model.layers.25.mlp.experts.128.up_proj", "model.layers.25.mlp.experts.129.up_proj", "model.layers.25.mlp.experts.130.up_proj", "model.layers.25.mlp.experts.131.up_proj", "model.layers.25.mlp.experts.132.up_proj", "model.layers.25.mlp.experts.133.up_proj", "model.layers.25.mlp.experts.134.up_proj", "model.layers.25.mlp.experts.135.up_proj", "model.layers.25.mlp.experts.136.up_proj", "model.layers.25.mlp.experts.137.up_proj", "model.layers.25.mlp.experts.138.up_proj", "model.layers.25.mlp.experts.139.up_proj", "model.layers.25.mlp.experts.140.up_proj", "model.layers.25.mlp.experts.141.up_proj", "model.layers.25.mlp.experts.142.up_proj", "model.layers.25.mlp.experts.143.up_proj", "model.layers.25.mlp.experts.144.up_proj", "model.layers.25.mlp.experts.145.up_proj", "model.layers.25.mlp.experts.146.up_proj", "model.layers.25.mlp.experts.147.up_proj", "model.layers.25.mlp.experts.148.up_proj", "model.layers.25.mlp.experts.149.up_proj", "model.layers.25.mlp.experts.150.up_proj", "model.layers.25.mlp.experts.151.up_proj", "model.layers.25.mlp.experts.152.up_proj", "model.layers.25.mlp.experts.153.up_proj", "model.layers.25.mlp.experts.154.up_proj", "model.layers.25.mlp.experts.155.up_proj", "model.layers.25.mlp.experts.156.up_proj", "model.layers.25.mlp.experts.157.up_proj", "model.layers.25.mlp.experts.158.up_proj", "model.layers.25.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00041105309501290876, "dbits": 2516582400 }, { "dkld": -0.0006616489961743383, "dbits": 5033164800 }, { "dkld": -0.000689419545233258, "dbits": 7549747200 }, { "dkld": -0.0007310884073376683, "dbits": 12582912000 } ] }, { "idx": 152, "layers": [ "model.layers.25.mlp.experts.0.down_proj", "model.layers.25.mlp.experts.1.down_proj", "model.layers.25.mlp.experts.2.down_proj", "model.layers.25.mlp.experts.3.down_proj", "model.layers.25.mlp.experts.4.down_proj", "model.layers.25.mlp.experts.5.down_proj", "model.layers.25.mlp.experts.6.down_proj", "model.layers.25.mlp.experts.7.down_proj", "model.layers.25.mlp.experts.8.down_proj", "model.layers.25.mlp.experts.9.down_proj", "model.layers.25.mlp.experts.10.down_proj", "model.layers.25.mlp.experts.11.down_proj", "model.layers.25.mlp.experts.12.down_proj", "model.layers.25.mlp.experts.13.down_proj", "model.layers.25.mlp.experts.14.down_proj", "model.layers.25.mlp.experts.15.down_proj", "model.layers.25.mlp.experts.16.down_proj", "model.layers.25.mlp.experts.17.down_proj", "model.layers.25.mlp.experts.18.down_proj", "model.layers.25.mlp.experts.19.down_proj", "model.layers.25.mlp.experts.20.down_proj", "model.layers.25.mlp.experts.21.down_proj", "model.layers.25.mlp.experts.22.down_proj", "model.layers.25.mlp.experts.23.down_proj", "model.layers.25.mlp.experts.24.down_proj", "model.layers.25.mlp.experts.25.down_proj", "model.layers.25.mlp.experts.26.down_proj", "model.layers.25.mlp.experts.27.down_proj", "model.layers.25.mlp.experts.28.down_proj", "model.layers.25.mlp.experts.29.down_proj", "model.layers.25.mlp.experts.30.down_proj", "model.layers.25.mlp.experts.31.down_proj", "model.layers.25.mlp.experts.32.down_proj", "model.layers.25.mlp.experts.33.down_proj", "model.layers.25.mlp.experts.34.down_proj", "model.layers.25.mlp.experts.35.down_proj", "model.layers.25.mlp.experts.36.down_proj", "model.layers.25.mlp.experts.37.down_proj", "model.layers.25.mlp.experts.38.down_proj", "model.layers.25.mlp.experts.39.down_proj", "model.layers.25.mlp.experts.40.down_proj", "model.layers.25.mlp.experts.41.down_proj", "model.layers.25.mlp.experts.42.down_proj", "model.layers.25.mlp.experts.43.down_proj", "model.layers.25.mlp.experts.44.down_proj", "model.layers.25.mlp.experts.45.down_proj", "model.layers.25.mlp.experts.46.down_proj", "model.layers.25.mlp.experts.47.down_proj", "model.layers.25.mlp.experts.48.down_proj", "model.layers.25.mlp.experts.49.down_proj", "model.layers.25.mlp.experts.50.down_proj", "model.layers.25.mlp.experts.51.down_proj", "model.layers.25.mlp.experts.52.down_proj", "model.layers.25.mlp.experts.53.down_proj", "model.layers.25.mlp.experts.54.down_proj", "model.layers.25.mlp.experts.55.down_proj", "model.layers.25.mlp.experts.56.down_proj", "model.layers.25.mlp.experts.57.down_proj", "model.layers.25.mlp.experts.58.down_proj", "model.layers.25.mlp.experts.59.down_proj", "model.layers.25.mlp.experts.60.down_proj", "model.layers.25.mlp.experts.61.down_proj", "model.layers.25.mlp.experts.62.down_proj", "model.layers.25.mlp.experts.63.down_proj", "model.layers.25.mlp.experts.64.down_proj", "model.layers.25.mlp.experts.65.down_proj", "model.layers.25.mlp.experts.66.down_proj", "model.layers.25.mlp.experts.67.down_proj", "model.layers.25.mlp.experts.68.down_proj", "model.layers.25.mlp.experts.69.down_proj", "model.layers.25.mlp.experts.70.down_proj", "model.layers.25.mlp.experts.71.down_proj", "model.layers.25.mlp.experts.72.down_proj", "model.layers.25.mlp.experts.73.down_proj", "model.layers.25.mlp.experts.74.down_proj", "model.layers.25.mlp.experts.75.down_proj", "model.layers.25.mlp.experts.76.down_proj", "model.layers.25.mlp.experts.77.down_proj", "model.layers.25.mlp.experts.78.down_proj", "model.layers.25.mlp.experts.79.down_proj", "model.layers.25.mlp.experts.80.down_proj", "model.layers.25.mlp.experts.81.down_proj", "model.layers.25.mlp.experts.82.down_proj", "model.layers.25.mlp.experts.83.down_proj", "model.layers.25.mlp.experts.84.down_proj", "model.layers.25.mlp.experts.85.down_proj", "model.layers.25.mlp.experts.86.down_proj", "model.layers.25.mlp.experts.87.down_proj", "model.layers.25.mlp.experts.88.down_proj", "model.layers.25.mlp.experts.89.down_proj", "model.layers.25.mlp.experts.90.down_proj", "model.layers.25.mlp.experts.91.down_proj", "model.layers.25.mlp.experts.92.down_proj", "model.layers.25.mlp.experts.93.down_proj", "model.layers.25.mlp.experts.94.down_proj", "model.layers.25.mlp.experts.95.down_proj", "model.layers.25.mlp.experts.96.down_proj", "model.layers.25.mlp.experts.97.down_proj", "model.layers.25.mlp.experts.98.down_proj", "model.layers.25.mlp.experts.99.down_proj", "model.layers.25.mlp.experts.100.down_proj", "model.layers.25.mlp.experts.101.down_proj", "model.layers.25.mlp.experts.102.down_proj", "model.layers.25.mlp.experts.103.down_proj", "model.layers.25.mlp.experts.104.down_proj", "model.layers.25.mlp.experts.105.down_proj", "model.layers.25.mlp.experts.106.down_proj", "model.layers.25.mlp.experts.107.down_proj", "model.layers.25.mlp.experts.108.down_proj", "model.layers.25.mlp.experts.109.down_proj", "model.layers.25.mlp.experts.110.down_proj", "model.layers.25.mlp.experts.111.down_proj", "model.layers.25.mlp.experts.112.down_proj", "model.layers.25.mlp.experts.113.down_proj", "model.layers.25.mlp.experts.114.down_proj", "model.layers.25.mlp.experts.115.down_proj", "model.layers.25.mlp.experts.116.down_proj", "model.layers.25.mlp.experts.117.down_proj", "model.layers.25.mlp.experts.118.down_proj", "model.layers.25.mlp.experts.119.down_proj", "model.layers.25.mlp.experts.120.down_proj", "model.layers.25.mlp.experts.121.down_proj", "model.layers.25.mlp.experts.122.down_proj", "model.layers.25.mlp.experts.123.down_proj", "model.layers.25.mlp.experts.124.down_proj", "model.layers.25.mlp.experts.125.down_proj", "model.layers.25.mlp.experts.126.down_proj", "model.layers.25.mlp.experts.127.down_proj", "model.layers.25.mlp.experts.128.down_proj", "model.layers.25.mlp.experts.129.down_proj", "model.layers.25.mlp.experts.130.down_proj", "model.layers.25.mlp.experts.131.down_proj", "model.layers.25.mlp.experts.132.down_proj", "model.layers.25.mlp.experts.133.down_proj", "model.layers.25.mlp.experts.134.down_proj", "model.layers.25.mlp.experts.135.down_proj", "model.layers.25.mlp.experts.136.down_proj", "model.layers.25.mlp.experts.137.down_proj", "model.layers.25.mlp.experts.138.down_proj", "model.layers.25.mlp.experts.139.down_proj", "model.layers.25.mlp.experts.140.down_proj", "model.layers.25.mlp.experts.141.down_proj", "model.layers.25.mlp.experts.142.down_proj", "model.layers.25.mlp.experts.143.down_proj", "model.layers.25.mlp.experts.144.down_proj", "model.layers.25.mlp.experts.145.down_proj", "model.layers.25.mlp.experts.146.down_proj", "model.layers.25.mlp.experts.147.down_proj", "model.layers.25.mlp.experts.148.down_proj", "model.layers.25.mlp.experts.149.down_proj", "model.layers.25.mlp.experts.150.down_proj", "model.layers.25.mlp.experts.151.down_proj", "model.layers.25.mlp.experts.152.down_proj", "model.layers.25.mlp.experts.153.down_proj", "model.layers.25.mlp.experts.154.down_proj", "model.layers.25.mlp.experts.155.down_proj", "model.layers.25.mlp.experts.156.down_proj", "model.layers.25.mlp.experts.157.down_proj", "model.layers.25.mlp.experts.158.down_proj", "model.layers.25.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00043016932904721417, "dbits": 1258291200 }, { "dkld": -0.0006006477400660598, "dbits": 2516582400 }, { "dkld": -0.0005361311137676295, "dbits": 3774873600 }, { "dkld": -0.0004986763000488365, "dbits": 6291456000 } ] }, { "idx": 153, "layers": [ "model.layers.26.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0003519527614116724, "dbits": 62914560 }, { "dkld": -0.00032717734575271606, "dbits": 125829120 }, { "dkld": -0.0004467945545911789, "dbits": 188743680 }, { "dkld": -0.0004113997332751834, "dbits": 314572800 } ] }, { "idx": 154, "layers": [ "model.layers.26.self_attn.k_proj", "model.layers.26.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00017543574795125805, "dbits": 10485760 }, { "dkld": -0.0001689003780484255, "dbits": 20971520 }, { "dkld": -0.00028838654980063716, "dbits": 31457280 }, { "dkld": -0.00026114797219634056, "dbits": 52428800 } ] }, { "idx": 155, "layers": [ "model.layers.26.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003456516191363418, "dbits": 62914560 }, { "dkld": -0.0010552492924034707, "dbits": 125829120 }, { "dkld": -0.001042845007032167, "dbits": 188743680 }, { "dkld": -0.0010266723111271858, "dbits": 314572800 } ] }, { "idx": 156, "layers": [ "model.layers.26.mlp.shared_experts.gate_proj", "model.layers.26.mlp.shared_experts.up_proj", "model.layers.26.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0009037891402840503, "dbits": 23592960 }, { "dkld": 0.0007297818548977347, "dbits": 47185920 }, { "dkld": 0.0006472039967775345, "dbits": 70778880 }, { "dkld": 0.0006820941343903542, "dbits": 117964800 } ] }, { "idx": 157, "layers": [ "model.layers.26.mlp.experts.0.gate_proj", "model.layers.26.mlp.experts.1.gate_proj", "model.layers.26.mlp.experts.2.gate_proj", "model.layers.26.mlp.experts.3.gate_proj", "model.layers.26.mlp.experts.4.gate_proj", "model.layers.26.mlp.experts.5.gate_proj", "model.layers.26.mlp.experts.6.gate_proj", "model.layers.26.mlp.experts.7.gate_proj", "model.layers.26.mlp.experts.8.gate_proj", "model.layers.26.mlp.experts.9.gate_proj", "model.layers.26.mlp.experts.10.gate_proj", "model.layers.26.mlp.experts.11.gate_proj", "model.layers.26.mlp.experts.12.gate_proj", "model.layers.26.mlp.experts.13.gate_proj", "model.layers.26.mlp.experts.14.gate_proj", "model.layers.26.mlp.experts.15.gate_proj", "model.layers.26.mlp.experts.16.gate_proj", "model.layers.26.mlp.experts.17.gate_proj", "model.layers.26.mlp.experts.18.gate_proj", "model.layers.26.mlp.experts.19.gate_proj", "model.layers.26.mlp.experts.20.gate_proj", "model.layers.26.mlp.experts.21.gate_proj", "model.layers.26.mlp.experts.22.gate_proj", "model.layers.26.mlp.experts.23.gate_proj", "model.layers.26.mlp.experts.24.gate_proj", "model.layers.26.mlp.experts.25.gate_proj", "model.layers.26.mlp.experts.26.gate_proj", "model.layers.26.mlp.experts.27.gate_proj", "model.layers.26.mlp.experts.28.gate_proj", "model.layers.26.mlp.experts.29.gate_proj", "model.layers.26.mlp.experts.30.gate_proj", "model.layers.26.mlp.experts.31.gate_proj", "model.layers.26.mlp.experts.32.gate_proj", "model.layers.26.mlp.experts.33.gate_proj", "model.layers.26.mlp.experts.34.gate_proj", "model.layers.26.mlp.experts.35.gate_proj", "model.layers.26.mlp.experts.36.gate_proj", "model.layers.26.mlp.experts.37.gate_proj", "model.layers.26.mlp.experts.38.gate_proj", "model.layers.26.mlp.experts.39.gate_proj", "model.layers.26.mlp.experts.40.gate_proj", "model.layers.26.mlp.experts.41.gate_proj", "model.layers.26.mlp.experts.42.gate_proj", "model.layers.26.mlp.experts.43.gate_proj", "model.layers.26.mlp.experts.44.gate_proj", "model.layers.26.mlp.experts.45.gate_proj", "model.layers.26.mlp.experts.46.gate_proj", "model.layers.26.mlp.experts.47.gate_proj", "model.layers.26.mlp.experts.48.gate_proj", "model.layers.26.mlp.experts.49.gate_proj", "model.layers.26.mlp.experts.50.gate_proj", "model.layers.26.mlp.experts.51.gate_proj", "model.layers.26.mlp.experts.52.gate_proj", "model.layers.26.mlp.experts.53.gate_proj", "model.layers.26.mlp.experts.54.gate_proj", "model.layers.26.mlp.experts.55.gate_proj", "model.layers.26.mlp.experts.56.gate_proj", "model.layers.26.mlp.experts.57.gate_proj", "model.layers.26.mlp.experts.58.gate_proj", "model.layers.26.mlp.experts.59.gate_proj", "model.layers.26.mlp.experts.60.gate_proj", "model.layers.26.mlp.experts.61.gate_proj", "model.layers.26.mlp.experts.62.gate_proj", "model.layers.26.mlp.experts.63.gate_proj", "model.layers.26.mlp.experts.64.gate_proj", "model.layers.26.mlp.experts.65.gate_proj", "model.layers.26.mlp.experts.66.gate_proj", "model.layers.26.mlp.experts.67.gate_proj", "model.layers.26.mlp.experts.68.gate_proj", "model.layers.26.mlp.experts.69.gate_proj", "model.layers.26.mlp.experts.70.gate_proj", "model.layers.26.mlp.experts.71.gate_proj", "model.layers.26.mlp.experts.72.gate_proj", "model.layers.26.mlp.experts.73.gate_proj", "model.layers.26.mlp.experts.74.gate_proj", "model.layers.26.mlp.experts.75.gate_proj", "model.layers.26.mlp.experts.76.gate_proj", "model.layers.26.mlp.experts.77.gate_proj", "model.layers.26.mlp.experts.78.gate_proj", "model.layers.26.mlp.experts.79.gate_proj", "model.layers.26.mlp.experts.80.gate_proj", "model.layers.26.mlp.experts.81.gate_proj", "model.layers.26.mlp.experts.82.gate_proj", "model.layers.26.mlp.experts.83.gate_proj", "model.layers.26.mlp.experts.84.gate_proj", "model.layers.26.mlp.experts.85.gate_proj", "model.layers.26.mlp.experts.86.gate_proj", "model.layers.26.mlp.experts.87.gate_proj", "model.layers.26.mlp.experts.88.gate_proj", "model.layers.26.mlp.experts.89.gate_proj", "model.layers.26.mlp.experts.90.gate_proj", "model.layers.26.mlp.experts.91.gate_proj", "model.layers.26.mlp.experts.92.gate_proj", "model.layers.26.mlp.experts.93.gate_proj", "model.layers.26.mlp.experts.94.gate_proj", "model.layers.26.mlp.experts.95.gate_proj", "model.layers.26.mlp.experts.96.gate_proj", "model.layers.26.mlp.experts.97.gate_proj", "model.layers.26.mlp.experts.98.gate_proj", "model.layers.26.mlp.experts.99.gate_proj", "model.layers.26.mlp.experts.100.gate_proj", "model.layers.26.mlp.experts.101.gate_proj", "model.layers.26.mlp.experts.102.gate_proj", "model.layers.26.mlp.experts.103.gate_proj", "model.layers.26.mlp.experts.104.gate_proj", "model.layers.26.mlp.experts.105.gate_proj", "model.layers.26.mlp.experts.106.gate_proj", "model.layers.26.mlp.experts.107.gate_proj", "model.layers.26.mlp.experts.108.gate_proj", "model.layers.26.mlp.experts.109.gate_proj", "model.layers.26.mlp.experts.110.gate_proj", "model.layers.26.mlp.experts.111.gate_proj", "model.layers.26.mlp.experts.112.gate_proj", "model.layers.26.mlp.experts.113.gate_proj", "model.layers.26.mlp.experts.114.gate_proj", "model.layers.26.mlp.experts.115.gate_proj", "model.layers.26.mlp.experts.116.gate_proj", "model.layers.26.mlp.experts.117.gate_proj", "model.layers.26.mlp.experts.118.gate_proj", "model.layers.26.mlp.experts.119.gate_proj", "model.layers.26.mlp.experts.120.gate_proj", "model.layers.26.mlp.experts.121.gate_proj", "model.layers.26.mlp.experts.122.gate_proj", "model.layers.26.mlp.experts.123.gate_proj", "model.layers.26.mlp.experts.124.gate_proj", "model.layers.26.mlp.experts.125.gate_proj", "model.layers.26.mlp.experts.126.gate_proj", "model.layers.26.mlp.experts.127.gate_proj", "model.layers.26.mlp.experts.128.gate_proj", "model.layers.26.mlp.experts.129.gate_proj", "model.layers.26.mlp.experts.130.gate_proj", "model.layers.26.mlp.experts.131.gate_proj", "model.layers.26.mlp.experts.132.gate_proj", "model.layers.26.mlp.experts.133.gate_proj", "model.layers.26.mlp.experts.134.gate_proj", "model.layers.26.mlp.experts.135.gate_proj", "model.layers.26.mlp.experts.136.gate_proj", "model.layers.26.mlp.experts.137.gate_proj", "model.layers.26.mlp.experts.138.gate_proj", "model.layers.26.mlp.experts.139.gate_proj", "model.layers.26.mlp.experts.140.gate_proj", "model.layers.26.mlp.experts.141.gate_proj", "model.layers.26.mlp.experts.142.gate_proj", "model.layers.26.mlp.experts.143.gate_proj", "model.layers.26.mlp.experts.144.gate_proj", "model.layers.26.mlp.experts.145.gate_proj", "model.layers.26.mlp.experts.146.gate_proj", "model.layers.26.mlp.experts.147.gate_proj", "model.layers.26.mlp.experts.148.gate_proj", "model.layers.26.mlp.experts.149.gate_proj", "model.layers.26.mlp.experts.150.gate_proj", "model.layers.26.mlp.experts.151.gate_proj", "model.layers.26.mlp.experts.152.gate_proj", "model.layers.26.mlp.experts.153.gate_proj", "model.layers.26.mlp.experts.154.gate_proj", "model.layers.26.mlp.experts.155.gate_proj", "model.layers.26.mlp.experts.156.gate_proj", "model.layers.26.mlp.experts.157.gate_proj", "model.layers.26.mlp.experts.158.gate_proj", "model.layers.26.mlp.experts.159.gate_proj", "model.layers.26.mlp.experts.0.up_proj", "model.layers.26.mlp.experts.1.up_proj", "model.layers.26.mlp.experts.2.up_proj", "model.layers.26.mlp.experts.3.up_proj", "model.layers.26.mlp.experts.4.up_proj", "model.layers.26.mlp.experts.5.up_proj", "model.layers.26.mlp.experts.6.up_proj", "model.layers.26.mlp.experts.7.up_proj", "model.layers.26.mlp.experts.8.up_proj", "model.layers.26.mlp.experts.9.up_proj", "model.layers.26.mlp.experts.10.up_proj", "model.layers.26.mlp.experts.11.up_proj", "model.layers.26.mlp.experts.12.up_proj", "model.layers.26.mlp.experts.13.up_proj", "model.layers.26.mlp.experts.14.up_proj", "model.layers.26.mlp.experts.15.up_proj", "model.layers.26.mlp.experts.16.up_proj", "model.layers.26.mlp.experts.17.up_proj", "model.layers.26.mlp.experts.18.up_proj", "model.layers.26.mlp.experts.19.up_proj", "model.layers.26.mlp.experts.20.up_proj", "model.layers.26.mlp.experts.21.up_proj", "model.layers.26.mlp.experts.22.up_proj", "model.layers.26.mlp.experts.23.up_proj", "model.layers.26.mlp.experts.24.up_proj", "model.layers.26.mlp.experts.25.up_proj", "model.layers.26.mlp.experts.26.up_proj", "model.layers.26.mlp.experts.27.up_proj", "model.layers.26.mlp.experts.28.up_proj", "model.layers.26.mlp.experts.29.up_proj", "model.layers.26.mlp.experts.30.up_proj", "model.layers.26.mlp.experts.31.up_proj", "model.layers.26.mlp.experts.32.up_proj", "model.layers.26.mlp.experts.33.up_proj", "model.layers.26.mlp.experts.34.up_proj", "model.layers.26.mlp.experts.35.up_proj", "model.layers.26.mlp.experts.36.up_proj", "model.layers.26.mlp.experts.37.up_proj", "model.layers.26.mlp.experts.38.up_proj", "model.layers.26.mlp.experts.39.up_proj", "model.layers.26.mlp.experts.40.up_proj", "model.layers.26.mlp.experts.41.up_proj", "model.layers.26.mlp.experts.42.up_proj", "model.layers.26.mlp.experts.43.up_proj", "model.layers.26.mlp.experts.44.up_proj", "model.layers.26.mlp.experts.45.up_proj", "model.layers.26.mlp.experts.46.up_proj", "model.layers.26.mlp.experts.47.up_proj", "model.layers.26.mlp.experts.48.up_proj", "model.layers.26.mlp.experts.49.up_proj", "model.layers.26.mlp.experts.50.up_proj", "model.layers.26.mlp.experts.51.up_proj", "model.layers.26.mlp.experts.52.up_proj", "model.layers.26.mlp.experts.53.up_proj", "model.layers.26.mlp.experts.54.up_proj", "model.layers.26.mlp.experts.55.up_proj", "model.layers.26.mlp.experts.56.up_proj", "model.layers.26.mlp.experts.57.up_proj", "model.layers.26.mlp.experts.58.up_proj", "model.layers.26.mlp.experts.59.up_proj", "model.layers.26.mlp.experts.60.up_proj", "model.layers.26.mlp.experts.61.up_proj", "model.layers.26.mlp.experts.62.up_proj", "model.layers.26.mlp.experts.63.up_proj", "model.layers.26.mlp.experts.64.up_proj", "model.layers.26.mlp.experts.65.up_proj", "model.layers.26.mlp.experts.66.up_proj", "model.layers.26.mlp.experts.67.up_proj", "model.layers.26.mlp.experts.68.up_proj", "model.layers.26.mlp.experts.69.up_proj", "model.layers.26.mlp.experts.70.up_proj", "model.layers.26.mlp.experts.71.up_proj", "model.layers.26.mlp.experts.72.up_proj", "model.layers.26.mlp.experts.73.up_proj", "model.layers.26.mlp.experts.74.up_proj", "model.layers.26.mlp.experts.75.up_proj", "model.layers.26.mlp.experts.76.up_proj", "model.layers.26.mlp.experts.77.up_proj", "model.layers.26.mlp.experts.78.up_proj", "model.layers.26.mlp.experts.79.up_proj", "model.layers.26.mlp.experts.80.up_proj", "model.layers.26.mlp.experts.81.up_proj", "model.layers.26.mlp.experts.82.up_proj", "model.layers.26.mlp.experts.83.up_proj", "model.layers.26.mlp.experts.84.up_proj", "model.layers.26.mlp.experts.85.up_proj", "model.layers.26.mlp.experts.86.up_proj", "model.layers.26.mlp.experts.87.up_proj", "model.layers.26.mlp.experts.88.up_proj", "model.layers.26.mlp.experts.89.up_proj", "model.layers.26.mlp.experts.90.up_proj", "model.layers.26.mlp.experts.91.up_proj", "model.layers.26.mlp.experts.92.up_proj", "model.layers.26.mlp.experts.93.up_proj", "model.layers.26.mlp.experts.94.up_proj", "model.layers.26.mlp.experts.95.up_proj", "model.layers.26.mlp.experts.96.up_proj", "model.layers.26.mlp.experts.97.up_proj", "model.layers.26.mlp.experts.98.up_proj", "model.layers.26.mlp.experts.99.up_proj", "model.layers.26.mlp.experts.100.up_proj", "model.layers.26.mlp.experts.101.up_proj", "model.layers.26.mlp.experts.102.up_proj", "model.layers.26.mlp.experts.103.up_proj", "model.layers.26.mlp.experts.104.up_proj", "model.layers.26.mlp.experts.105.up_proj", "model.layers.26.mlp.experts.106.up_proj", "model.layers.26.mlp.experts.107.up_proj", "model.layers.26.mlp.experts.108.up_proj", "model.layers.26.mlp.experts.109.up_proj", "model.layers.26.mlp.experts.110.up_proj", "model.layers.26.mlp.experts.111.up_proj", "model.layers.26.mlp.experts.112.up_proj", "model.layers.26.mlp.experts.113.up_proj", "model.layers.26.mlp.experts.114.up_proj", "model.layers.26.mlp.experts.115.up_proj", "model.layers.26.mlp.experts.116.up_proj", "model.layers.26.mlp.experts.117.up_proj", "model.layers.26.mlp.experts.118.up_proj", "model.layers.26.mlp.experts.119.up_proj", "model.layers.26.mlp.experts.120.up_proj", "model.layers.26.mlp.experts.121.up_proj", "model.layers.26.mlp.experts.122.up_proj", "model.layers.26.mlp.experts.123.up_proj", "model.layers.26.mlp.experts.124.up_proj", "model.layers.26.mlp.experts.125.up_proj", "model.layers.26.mlp.experts.126.up_proj", "model.layers.26.mlp.experts.127.up_proj", "model.layers.26.mlp.experts.128.up_proj", "model.layers.26.mlp.experts.129.up_proj", "model.layers.26.mlp.experts.130.up_proj", "model.layers.26.mlp.experts.131.up_proj", "model.layers.26.mlp.experts.132.up_proj", "model.layers.26.mlp.experts.133.up_proj", "model.layers.26.mlp.experts.134.up_proj", "model.layers.26.mlp.experts.135.up_proj", "model.layers.26.mlp.experts.136.up_proj", "model.layers.26.mlp.experts.137.up_proj", "model.layers.26.mlp.experts.138.up_proj", "model.layers.26.mlp.experts.139.up_proj", "model.layers.26.mlp.experts.140.up_proj", "model.layers.26.mlp.experts.141.up_proj", "model.layers.26.mlp.experts.142.up_proj", "model.layers.26.mlp.experts.143.up_proj", "model.layers.26.mlp.experts.144.up_proj", "model.layers.26.mlp.experts.145.up_proj", "model.layers.26.mlp.experts.146.up_proj", "model.layers.26.mlp.experts.147.up_proj", "model.layers.26.mlp.experts.148.up_proj", "model.layers.26.mlp.experts.149.up_proj", "model.layers.26.mlp.experts.150.up_proj", "model.layers.26.mlp.experts.151.up_proj", "model.layers.26.mlp.experts.152.up_proj", "model.layers.26.mlp.experts.153.up_proj", "model.layers.26.mlp.experts.154.up_proj", "model.layers.26.mlp.experts.155.up_proj", "model.layers.26.mlp.experts.156.up_proj", "model.layers.26.mlp.experts.157.up_proj", "model.layers.26.mlp.experts.158.up_proj", "model.layers.26.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0007053770124912373, "dbits": 2516582400 }, { "dkld": -0.0008468346670270077, "dbits": 5033164800 }, { "dkld": -0.0007683902047574548, "dbits": 7549747200 }, { "dkld": -0.0008749038912355955, "dbits": 12582912000 } ] }, { "idx": 158, "layers": [ "model.layers.26.mlp.experts.0.down_proj", "model.layers.26.mlp.experts.1.down_proj", "model.layers.26.mlp.experts.2.down_proj", "model.layers.26.mlp.experts.3.down_proj", "model.layers.26.mlp.experts.4.down_proj", "model.layers.26.mlp.experts.5.down_proj", "model.layers.26.mlp.experts.6.down_proj", "model.layers.26.mlp.experts.7.down_proj", "model.layers.26.mlp.experts.8.down_proj", "model.layers.26.mlp.experts.9.down_proj", "model.layers.26.mlp.experts.10.down_proj", "model.layers.26.mlp.experts.11.down_proj", "model.layers.26.mlp.experts.12.down_proj", "model.layers.26.mlp.experts.13.down_proj", "model.layers.26.mlp.experts.14.down_proj", "model.layers.26.mlp.experts.15.down_proj", "model.layers.26.mlp.experts.16.down_proj", "model.layers.26.mlp.experts.17.down_proj", "model.layers.26.mlp.experts.18.down_proj", "model.layers.26.mlp.experts.19.down_proj", "model.layers.26.mlp.experts.20.down_proj", "model.layers.26.mlp.experts.21.down_proj", "model.layers.26.mlp.experts.22.down_proj", "model.layers.26.mlp.experts.23.down_proj", "model.layers.26.mlp.experts.24.down_proj", "model.layers.26.mlp.experts.25.down_proj", "model.layers.26.mlp.experts.26.down_proj", "model.layers.26.mlp.experts.27.down_proj", "model.layers.26.mlp.experts.28.down_proj", "model.layers.26.mlp.experts.29.down_proj", "model.layers.26.mlp.experts.30.down_proj", "model.layers.26.mlp.experts.31.down_proj", "model.layers.26.mlp.experts.32.down_proj", "model.layers.26.mlp.experts.33.down_proj", "model.layers.26.mlp.experts.34.down_proj", "model.layers.26.mlp.experts.35.down_proj", "model.layers.26.mlp.experts.36.down_proj", "model.layers.26.mlp.experts.37.down_proj", "model.layers.26.mlp.experts.38.down_proj", "model.layers.26.mlp.experts.39.down_proj", "model.layers.26.mlp.experts.40.down_proj", "model.layers.26.mlp.experts.41.down_proj", "model.layers.26.mlp.experts.42.down_proj", "model.layers.26.mlp.experts.43.down_proj", "model.layers.26.mlp.experts.44.down_proj", "model.layers.26.mlp.experts.45.down_proj", "model.layers.26.mlp.experts.46.down_proj", "model.layers.26.mlp.experts.47.down_proj", "model.layers.26.mlp.experts.48.down_proj", "model.layers.26.mlp.experts.49.down_proj", "model.layers.26.mlp.experts.50.down_proj", "model.layers.26.mlp.experts.51.down_proj", "model.layers.26.mlp.experts.52.down_proj", "model.layers.26.mlp.experts.53.down_proj", "model.layers.26.mlp.experts.54.down_proj", "model.layers.26.mlp.experts.55.down_proj", "model.layers.26.mlp.experts.56.down_proj", "model.layers.26.mlp.experts.57.down_proj", "model.layers.26.mlp.experts.58.down_proj", "model.layers.26.mlp.experts.59.down_proj", "model.layers.26.mlp.experts.60.down_proj", "model.layers.26.mlp.experts.61.down_proj", "model.layers.26.mlp.experts.62.down_proj", "model.layers.26.mlp.experts.63.down_proj", "model.layers.26.mlp.experts.64.down_proj", "model.layers.26.mlp.experts.65.down_proj", "model.layers.26.mlp.experts.66.down_proj", "model.layers.26.mlp.experts.67.down_proj", "model.layers.26.mlp.experts.68.down_proj", "model.layers.26.mlp.experts.69.down_proj", "model.layers.26.mlp.experts.70.down_proj", "model.layers.26.mlp.experts.71.down_proj", "model.layers.26.mlp.experts.72.down_proj", "model.layers.26.mlp.experts.73.down_proj", "model.layers.26.mlp.experts.74.down_proj", "model.layers.26.mlp.experts.75.down_proj", "model.layers.26.mlp.experts.76.down_proj", "model.layers.26.mlp.experts.77.down_proj", "model.layers.26.mlp.experts.78.down_proj", "model.layers.26.mlp.experts.79.down_proj", "model.layers.26.mlp.experts.80.down_proj", "model.layers.26.mlp.experts.81.down_proj", "model.layers.26.mlp.experts.82.down_proj", "model.layers.26.mlp.experts.83.down_proj", "model.layers.26.mlp.experts.84.down_proj", "model.layers.26.mlp.experts.85.down_proj", "model.layers.26.mlp.experts.86.down_proj", "model.layers.26.mlp.experts.87.down_proj", "model.layers.26.mlp.experts.88.down_proj", "model.layers.26.mlp.experts.89.down_proj", "model.layers.26.mlp.experts.90.down_proj", "model.layers.26.mlp.experts.91.down_proj", "model.layers.26.mlp.experts.92.down_proj", "model.layers.26.mlp.experts.93.down_proj", "model.layers.26.mlp.experts.94.down_proj", "model.layers.26.mlp.experts.95.down_proj", "model.layers.26.mlp.experts.96.down_proj", "model.layers.26.mlp.experts.97.down_proj", "model.layers.26.mlp.experts.98.down_proj", "model.layers.26.mlp.experts.99.down_proj", "model.layers.26.mlp.experts.100.down_proj", "model.layers.26.mlp.experts.101.down_proj", "model.layers.26.mlp.experts.102.down_proj", "model.layers.26.mlp.experts.103.down_proj", "model.layers.26.mlp.experts.104.down_proj", "model.layers.26.mlp.experts.105.down_proj", "model.layers.26.mlp.experts.106.down_proj", "model.layers.26.mlp.experts.107.down_proj", "model.layers.26.mlp.experts.108.down_proj", "model.layers.26.mlp.experts.109.down_proj", "model.layers.26.mlp.experts.110.down_proj", "model.layers.26.mlp.experts.111.down_proj", "model.layers.26.mlp.experts.112.down_proj", "model.layers.26.mlp.experts.113.down_proj", "model.layers.26.mlp.experts.114.down_proj", "model.layers.26.mlp.experts.115.down_proj", "model.layers.26.mlp.experts.116.down_proj", "model.layers.26.mlp.experts.117.down_proj", "model.layers.26.mlp.experts.118.down_proj", "model.layers.26.mlp.experts.119.down_proj", "model.layers.26.mlp.experts.120.down_proj", "model.layers.26.mlp.experts.121.down_proj", "model.layers.26.mlp.experts.122.down_proj", "model.layers.26.mlp.experts.123.down_proj", "model.layers.26.mlp.experts.124.down_proj", "model.layers.26.mlp.experts.125.down_proj", "model.layers.26.mlp.experts.126.down_proj", "model.layers.26.mlp.experts.127.down_proj", "model.layers.26.mlp.experts.128.down_proj", "model.layers.26.mlp.experts.129.down_proj", "model.layers.26.mlp.experts.130.down_proj", "model.layers.26.mlp.experts.131.down_proj", "model.layers.26.mlp.experts.132.down_proj", "model.layers.26.mlp.experts.133.down_proj", "model.layers.26.mlp.experts.134.down_proj", "model.layers.26.mlp.experts.135.down_proj", "model.layers.26.mlp.experts.136.down_proj", "model.layers.26.mlp.experts.137.down_proj", "model.layers.26.mlp.experts.138.down_proj", "model.layers.26.mlp.experts.139.down_proj", "model.layers.26.mlp.experts.140.down_proj", "model.layers.26.mlp.experts.141.down_proj", "model.layers.26.mlp.experts.142.down_proj", "model.layers.26.mlp.experts.143.down_proj", "model.layers.26.mlp.experts.144.down_proj", "model.layers.26.mlp.experts.145.down_proj", "model.layers.26.mlp.experts.146.down_proj", "model.layers.26.mlp.experts.147.down_proj", "model.layers.26.mlp.experts.148.down_proj", "model.layers.26.mlp.experts.149.down_proj", "model.layers.26.mlp.experts.150.down_proj", "model.layers.26.mlp.experts.151.down_proj", "model.layers.26.mlp.experts.152.down_proj", "model.layers.26.mlp.experts.153.down_proj", "model.layers.26.mlp.experts.154.down_proj", "model.layers.26.mlp.experts.155.down_proj", "model.layers.26.mlp.experts.156.down_proj", "model.layers.26.mlp.experts.157.down_proj", "model.layers.26.mlp.experts.158.down_proj", "model.layers.26.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003503325395286111, "dbits": 1258291200 }, { "dkld": -0.0004368121735751629, "dbits": 2516582400 }, { "dkld": -0.0004261141642928207, "dbits": 3774873600 }, { "dkld": -0.00042823068797588626, "dbits": 6291456000 } ] }, { "idx": 159, "layers": [ "model.layers.27.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002796765416860636, "dbits": 62914560 }, { "dkld": -0.00023849438875914175, "dbits": 125829120 }, { "dkld": -0.00021733110770583153, "dbits": 188743680 }, { "dkld": -0.0003507232293486623, "dbits": 314572800 } ] }, { "idx": 160, "layers": [ "model.layers.27.self_attn.k_proj", "model.layers.27.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007261058315634755, "dbits": 10485760 }, { "dkld": -0.0009191951714456192, "dbits": 20971520 }, { "dkld": -0.0009517907164990902, "dbits": 31457280 }, { "dkld": -0.0009040528908371981, "dbits": 52428800 } ] }, { "idx": 161, "layers": [ "model.layers.27.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0011459964327514283, "dbits": 62914560 }, { "dkld": -0.001098944246768957, "dbits": 125829120 }, { "dkld": -0.0014526454731822042, "dbits": 188743680 }, { "dkld": -0.0013433237560093458, "dbits": 314572800 } ] }, { "idx": 162, "layers": [ "model.layers.27.mlp.shared_experts.gate_proj", "model.layers.27.mlp.shared_experts.up_proj", "model.layers.27.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00105355065315961, "dbits": 23592960 }, { "dkld": 0.0013151211664080564, "dbits": 47185920 }, { "dkld": 0.0012652440927922726, "dbits": 70778880 }, { "dkld": 0.001278596930205822, "dbits": 117964800 } ] }, { "idx": 163, "layers": [ "model.layers.27.mlp.experts.0.gate_proj", "model.layers.27.mlp.experts.1.gate_proj", "model.layers.27.mlp.experts.2.gate_proj", "model.layers.27.mlp.experts.3.gate_proj", "model.layers.27.mlp.experts.4.gate_proj", "model.layers.27.mlp.experts.5.gate_proj", "model.layers.27.mlp.experts.6.gate_proj", "model.layers.27.mlp.experts.7.gate_proj", "model.layers.27.mlp.experts.8.gate_proj", "model.layers.27.mlp.experts.9.gate_proj", "model.layers.27.mlp.experts.10.gate_proj", "model.layers.27.mlp.experts.11.gate_proj", "model.layers.27.mlp.experts.12.gate_proj", "model.layers.27.mlp.experts.13.gate_proj", "model.layers.27.mlp.experts.14.gate_proj", "model.layers.27.mlp.experts.15.gate_proj", "model.layers.27.mlp.experts.16.gate_proj", "model.layers.27.mlp.experts.17.gate_proj", "model.layers.27.mlp.experts.18.gate_proj", "model.layers.27.mlp.experts.19.gate_proj", "model.layers.27.mlp.experts.20.gate_proj", "model.layers.27.mlp.experts.21.gate_proj", "model.layers.27.mlp.experts.22.gate_proj", "model.layers.27.mlp.experts.23.gate_proj", "model.layers.27.mlp.experts.24.gate_proj", "model.layers.27.mlp.experts.25.gate_proj", "model.layers.27.mlp.experts.26.gate_proj", "model.layers.27.mlp.experts.27.gate_proj", "model.layers.27.mlp.experts.28.gate_proj", "model.layers.27.mlp.experts.29.gate_proj", "model.layers.27.mlp.experts.30.gate_proj", "model.layers.27.mlp.experts.31.gate_proj", "model.layers.27.mlp.experts.32.gate_proj", "model.layers.27.mlp.experts.33.gate_proj", "model.layers.27.mlp.experts.34.gate_proj", "model.layers.27.mlp.experts.35.gate_proj", "model.layers.27.mlp.experts.36.gate_proj", "model.layers.27.mlp.experts.37.gate_proj", "model.layers.27.mlp.experts.38.gate_proj", "model.layers.27.mlp.experts.39.gate_proj", "model.layers.27.mlp.experts.40.gate_proj", "model.layers.27.mlp.experts.41.gate_proj", "model.layers.27.mlp.experts.42.gate_proj", "model.layers.27.mlp.experts.43.gate_proj", "model.layers.27.mlp.experts.44.gate_proj", "model.layers.27.mlp.experts.45.gate_proj", "model.layers.27.mlp.experts.46.gate_proj", "model.layers.27.mlp.experts.47.gate_proj", "model.layers.27.mlp.experts.48.gate_proj", "model.layers.27.mlp.experts.49.gate_proj", "model.layers.27.mlp.experts.50.gate_proj", "model.layers.27.mlp.experts.51.gate_proj", "model.layers.27.mlp.experts.52.gate_proj", "model.layers.27.mlp.experts.53.gate_proj", "model.layers.27.mlp.experts.54.gate_proj", "model.layers.27.mlp.experts.55.gate_proj", "model.layers.27.mlp.experts.56.gate_proj", "model.layers.27.mlp.experts.57.gate_proj", "model.layers.27.mlp.experts.58.gate_proj", "model.layers.27.mlp.experts.59.gate_proj", "model.layers.27.mlp.experts.60.gate_proj", "model.layers.27.mlp.experts.61.gate_proj", "model.layers.27.mlp.experts.62.gate_proj", "model.layers.27.mlp.experts.63.gate_proj", "model.layers.27.mlp.experts.64.gate_proj", "model.layers.27.mlp.experts.65.gate_proj", "model.layers.27.mlp.experts.66.gate_proj", "model.layers.27.mlp.experts.67.gate_proj", "model.layers.27.mlp.experts.68.gate_proj", "model.layers.27.mlp.experts.69.gate_proj", "model.layers.27.mlp.experts.70.gate_proj", "model.layers.27.mlp.experts.71.gate_proj", "model.layers.27.mlp.experts.72.gate_proj", "model.layers.27.mlp.experts.73.gate_proj", "model.layers.27.mlp.experts.74.gate_proj", "model.layers.27.mlp.experts.75.gate_proj", "model.layers.27.mlp.experts.76.gate_proj", "model.layers.27.mlp.experts.77.gate_proj", "model.layers.27.mlp.experts.78.gate_proj", "model.layers.27.mlp.experts.79.gate_proj", "model.layers.27.mlp.experts.80.gate_proj", "model.layers.27.mlp.experts.81.gate_proj", "model.layers.27.mlp.experts.82.gate_proj", "model.layers.27.mlp.experts.83.gate_proj", "model.layers.27.mlp.experts.84.gate_proj", "model.layers.27.mlp.experts.85.gate_proj", "model.layers.27.mlp.experts.86.gate_proj", "model.layers.27.mlp.experts.87.gate_proj", "model.layers.27.mlp.experts.88.gate_proj", "model.layers.27.mlp.experts.89.gate_proj", "model.layers.27.mlp.experts.90.gate_proj", "model.layers.27.mlp.experts.91.gate_proj", "model.layers.27.mlp.experts.92.gate_proj", "model.layers.27.mlp.experts.93.gate_proj", "model.layers.27.mlp.experts.94.gate_proj", "model.layers.27.mlp.experts.95.gate_proj", "model.layers.27.mlp.experts.96.gate_proj", "model.layers.27.mlp.experts.97.gate_proj", "model.layers.27.mlp.experts.98.gate_proj", "model.layers.27.mlp.experts.99.gate_proj", "model.layers.27.mlp.experts.100.gate_proj", "model.layers.27.mlp.experts.101.gate_proj", "model.layers.27.mlp.experts.102.gate_proj", "model.layers.27.mlp.experts.103.gate_proj", "model.layers.27.mlp.experts.104.gate_proj", "model.layers.27.mlp.experts.105.gate_proj", "model.layers.27.mlp.experts.106.gate_proj", "model.layers.27.mlp.experts.107.gate_proj", "model.layers.27.mlp.experts.108.gate_proj", "model.layers.27.mlp.experts.109.gate_proj", "model.layers.27.mlp.experts.110.gate_proj", "model.layers.27.mlp.experts.111.gate_proj", "model.layers.27.mlp.experts.112.gate_proj", "model.layers.27.mlp.experts.113.gate_proj", "model.layers.27.mlp.experts.114.gate_proj", "model.layers.27.mlp.experts.115.gate_proj", "model.layers.27.mlp.experts.116.gate_proj", "model.layers.27.mlp.experts.117.gate_proj", "model.layers.27.mlp.experts.118.gate_proj", "model.layers.27.mlp.experts.119.gate_proj", "model.layers.27.mlp.experts.120.gate_proj", "model.layers.27.mlp.experts.121.gate_proj", "model.layers.27.mlp.experts.122.gate_proj", "model.layers.27.mlp.experts.123.gate_proj", "model.layers.27.mlp.experts.124.gate_proj", "model.layers.27.mlp.experts.125.gate_proj", "model.layers.27.mlp.experts.126.gate_proj", "model.layers.27.mlp.experts.127.gate_proj", "model.layers.27.mlp.experts.128.gate_proj", "model.layers.27.mlp.experts.129.gate_proj", "model.layers.27.mlp.experts.130.gate_proj", "model.layers.27.mlp.experts.131.gate_proj", "model.layers.27.mlp.experts.132.gate_proj", "model.layers.27.mlp.experts.133.gate_proj", "model.layers.27.mlp.experts.134.gate_proj", "model.layers.27.mlp.experts.135.gate_proj", "model.layers.27.mlp.experts.136.gate_proj", "model.layers.27.mlp.experts.137.gate_proj", "model.layers.27.mlp.experts.138.gate_proj", "model.layers.27.mlp.experts.139.gate_proj", "model.layers.27.mlp.experts.140.gate_proj", "model.layers.27.mlp.experts.141.gate_proj", "model.layers.27.mlp.experts.142.gate_proj", "model.layers.27.mlp.experts.143.gate_proj", "model.layers.27.mlp.experts.144.gate_proj", "model.layers.27.mlp.experts.145.gate_proj", "model.layers.27.mlp.experts.146.gate_proj", "model.layers.27.mlp.experts.147.gate_proj", "model.layers.27.mlp.experts.148.gate_proj", "model.layers.27.mlp.experts.149.gate_proj", "model.layers.27.mlp.experts.150.gate_proj", "model.layers.27.mlp.experts.151.gate_proj", "model.layers.27.mlp.experts.152.gate_proj", "model.layers.27.mlp.experts.153.gate_proj", "model.layers.27.mlp.experts.154.gate_proj", "model.layers.27.mlp.experts.155.gate_proj", "model.layers.27.mlp.experts.156.gate_proj", "model.layers.27.mlp.experts.157.gate_proj", "model.layers.27.mlp.experts.158.gate_proj", "model.layers.27.mlp.experts.159.gate_proj", "model.layers.27.mlp.experts.0.up_proj", "model.layers.27.mlp.experts.1.up_proj", "model.layers.27.mlp.experts.2.up_proj", "model.layers.27.mlp.experts.3.up_proj", "model.layers.27.mlp.experts.4.up_proj", "model.layers.27.mlp.experts.5.up_proj", "model.layers.27.mlp.experts.6.up_proj", "model.layers.27.mlp.experts.7.up_proj", "model.layers.27.mlp.experts.8.up_proj", "model.layers.27.mlp.experts.9.up_proj", "model.layers.27.mlp.experts.10.up_proj", "model.layers.27.mlp.experts.11.up_proj", "model.layers.27.mlp.experts.12.up_proj", "model.layers.27.mlp.experts.13.up_proj", "model.layers.27.mlp.experts.14.up_proj", "model.layers.27.mlp.experts.15.up_proj", "model.layers.27.mlp.experts.16.up_proj", "model.layers.27.mlp.experts.17.up_proj", "model.layers.27.mlp.experts.18.up_proj", "model.layers.27.mlp.experts.19.up_proj", "model.layers.27.mlp.experts.20.up_proj", "model.layers.27.mlp.experts.21.up_proj", "model.layers.27.mlp.experts.22.up_proj", "model.layers.27.mlp.experts.23.up_proj", "model.layers.27.mlp.experts.24.up_proj", "model.layers.27.mlp.experts.25.up_proj", "model.layers.27.mlp.experts.26.up_proj", "model.layers.27.mlp.experts.27.up_proj", "model.layers.27.mlp.experts.28.up_proj", "model.layers.27.mlp.experts.29.up_proj", "model.layers.27.mlp.experts.30.up_proj", "model.layers.27.mlp.experts.31.up_proj", "model.layers.27.mlp.experts.32.up_proj", "model.layers.27.mlp.experts.33.up_proj", "model.layers.27.mlp.experts.34.up_proj", "model.layers.27.mlp.experts.35.up_proj", "model.layers.27.mlp.experts.36.up_proj", "model.layers.27.mlp.experts.37.up_proj", "model.layers.27.mlp.experts.38.up_proj", "model.layers.27.mlp.experts.39.up_proj", "model.layers.27.mlp.experts.40.up_proj", "model.layers.27.mlp.experts.41.up_proj", "model.layers.27.mlp.experts.42.up_proj", "model.layers.27.mlp.experts.43.up_proj", "model.layers.27.mlp.experts.44.up_proj", "model.layers.27.mlp.experts.45.up_proj", "model.layers.27.mlp.experts.46.up_proj", "model.layers.27.mlp.experts.47.up_proj", "model.layers.27.mlp.experts.48.up_proj", "model.layers.27.mlp.experts.49.up_proj", "model.layers.27.mlp.experts.50.up_proj", "model.layers.27.mlp.experts.51.up_proj", "model.layers.27.mlp.experts.52.up_proj", "model.layers.27.mlp.experts.53.up_proj", "model.layers.27.mlp.experts.54.up_proj", "model.layers.27.mlp.experts.55.up_proj", "model.layers.27.mlp.experts.56.up_proj", "model.layers.27.mlp.experts.57.up_proj", "model.layers.27.mlp.experts.58.up_proj", "model.layers.27.mlp.experts.59.up_proj", "model.layers.27.mlp.experts.60.up_proj", "model.layers.27.mlp.experts.61.up_proj", "model.layers.27.mlp.experts.62.up_proj", "model.layers.27.mlp.experts.63.up_proj", "model.layers.27.mlp.experts.64.up_proj", "model.layers.27.mlp.experts.65.up_proj", "model.layers.27.mlp.experts.66.up_proj", "model.layers.27.mlp.experts.67.up_proj", "model.layers.27.mlp.experts.68.up_proj", "model.layers.27.mlp.experts.69.up_proj", "model.layers.27.mlp.experts.70.up_proj", "model.layers.27.mlp.experts.71.up_proj", "model.layers.27.mlp.experts.72.up_proj", "model.layers.27.mlp.experts.73.up_proj", "model.layers.27.mlp.experts.74.up_proj", "model.layers.27.mlp.experts.75.up_proj", "model.layers.27.mlp.experts.76.up_proj", "model.layers.27.mlp.experts.77.up_proj", "model.layers.27.mlp.experts.78.up_proj", "model.layers.27.mlp.experts.79.up_proj", "model.layers.27.mlp.experts.80.up_proj", "model.layers.27.mlp.experts.81.up_proj", "model.layers.27.mlp.experts.82.up_proj", "model.layers.27.mlp.experts.83.up_proj", "model.layers.27.mlp.experts.84.up_proj", "model.layers.27.mlp.experts.85.up_proj", "model.layers.27.mlp.experts.86.up_proj", "model.layers.27.mlp.experts.87.up_proj", "model.layers.27.mlp.experts.88.up_proj", "model.layers.27.mlp.experts.89.up_proj", "model.layers.27.mlp.experts.90.up_proj", "model.layers.27.mlp.experts.91.up_proj", "model.layers.27.mlp.experts.92.up_proj", "model.layers.27.mlp.experts.93.up_proj", "model.layers.27.mlp.experts.94.up_proj", "model.layers.27.mlp.experts.95.up_proj", "model.layers.27.mlp.experts.96.up_proj", "model.layers.27.mlp.experts.97.up_proj", "model.layers.27.mlp.experts.98.up_proj", "model.layers.27.mlp.experts.99.up_proj", "model.layers.27.mlp.experts.100.up_proj", "model.layers.27.mlp.experts.101.up_proj", "model.layers.27.mlp.experts.102.up_proj", "model.layers.27.mlp.experts.103.up_proj", "model.layers.27.mlp.experts.104.up_proj", "model.layers.27.mlp.experts.105.up_proj", "model.layers.27.mlp.experts.106.up_proj", "model.layers.27.mlp.experts.107.up_proj", "model.layers.27.mlp.experts.108.up_proj", "model.layers.27.mlp.experts.109.up_proj", "model.layers.27.mlp.experts.110.up_proj", "model.layers.27.mlp.experts.111.up_proj", "model.layers.27.mlp.experts.112.up_proj", "model.layers.27.mlp.experts.113.up_proj", "model.layers.27.mlp.experts.114.up_proj", "model.layers.27.mlp.experts.115.up_proj", "model.layers.27.mlp.experts.116.up_proj", "model.layers.27.mlp.experts.117.up_proj", "model.layers.27.mlp.experts.118.up_proj", "model.layers.27.mlp.experts.119.up_proj", "model.layers.27.mlp.experts.120.up_proj", "model.layers.27.mlp.experts.121.up_proj", "model.layers.27.mlp.experts.122.up_proj", "model.layers.27.mlp.experts.123.up_proj", "model.layers.27.mlp.experts.124.up_proj", "model.layers.27.mlp.experts.125.up_proj", "model.layers.27.mlp.experts.126.up_proj", "model.layers.27.mlp.experts.127.up_proj", "model.layers.27.mlp.experts.128.up_proj", "model.layers.27.mlp.experts.129.up_proj", "model.layers.27.mlp.experts.130.up_proj", "model.layers.27.mlp.experts.131.up_proj", "model.layers.27.mlp.experts.132.up_proj", "model.layers.27.mlp.experts.133.up_proj", "model.layers.27.mlp.experts.134.up_proj", "model.layers.27.mlp.experts.135.up_proj", "model.layers.27.mlp.experts.136.up_proj", "model.layers.27.mlp.experts.137.up_proj", "model.layers.27.mlp.experts.138.up_proj", "model.layers.27.mlp.experts.139.up_proj", "model.layers.27.mlp.experts.140.up_proj", "model.layers.27.mlp.experts.141.up_proj", "model.layers.27.mlp.experts.142.up_proj", "model.layers.27.mlp.experts.143.up_proj", "model.layers.27.mlp.experts.144.up_proj", "model.layers.27.mlp.experts.145.up_proj", "model.layers.27.mlp.experts.146.up_proj", "model.layers.27.mlp.experts.147.up_proj", "model.layers.27.mlp.experts.148.up_proj", "model.layers.27.mlp.experts.149.up_proj", "model.layers.27.mlp.experts.150.up_proj", "model.layers.27.mlp.experts.151.up_proj", "model.layers.27.mlp.experts.152.up_proj", "model.layers.27.mlp.experts.153.up_proj", "model.layers.27.mlp.experts.154.up_proj", "model.layers.27.mlp.experts.155.up_proj", "model.layers.27.mlp.experts.156.up_proj", "model.layers.27.mlp.experts.157.up_proj", "model.layers.27.mlp.experts.158.up_proj", "model.layers.27.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0005524426698684803, "dbits": 2516582400 }, { "dkld": -0.0005594408139586449, "dbits": 5033164800 }, { "dkld": -0.0005457811057567596, "dbits": 7549747200 }, { "dkld": -0.0005378453060984612, "dbits": 12582912000 } ] }, { "idx": 164, "layers": [ "model.layers.27.mlp.experts.0.down_proj", "model.layers.27.mlp.experts.1.down_proj", "model.layers.27.mlp.experts.2.down_proj", "model.layers.27.mlp.experts.3.down_proj", "model.layers.27.mlp.experts.4.down_proj", "model.layers.27.mlp.experts.5.down_proj", "model.layers.27.mlp.experts.6.down_proj", "model.layers.27.mlp.experts.7.down_proj", "model.layers.27.mlp.experts.8.down_proj", "model.layers.27.mlp.experts.9.down_proj", "model.layers.27.mlp.experts.10.down_proj", "model.layers.27.mlp.experts.11.down_proj", "model.layers.27.mlp.experts.12.down_proj", "model.layers.27.mlp.experts.13.down_proj", "model.layers.27.mlp.experts.14.down_proj", "model.layers.27.mlp.experts.15.down_proj", "model.layers.27.mlp.experts.16.down_proj", "model.layers.27.mlp.experts.17.down_proj", "model.layers.27.mlp.experts.18.down_proj", "model.layers.27.mlp.experts.19.down_proj", "model.layers.27.mlp.experts.20.down_proj", "model.layers.27.mlp.experts.21.down_proj", "model.layers.27.mlp.experts.22.down_proj", "model.layers.27.mlp.experts.23.down_proj", "model.layers.27.mlp.experts.24.down_proj", "model.layers.27.mlp.experts.25.down_proj", "model.layers.27.mlp.experts.26.down_proj", "model.layers.27.mlp.experts.27.down_proj", "model.layers.27.mlp.experts.28.down_proj", "model.layers.27.mlp.experts.29.down_proj", "model.layers.27.mlp.experts.30.down_proj", "model.layers.27.mlp.experts.31.down_proj", "model.layers.27.mlp.experts.32.down_proj", "model.layers.27.mlp.experts.33.down_proj", "model.layers.27.mlp.experts.34.down_proj", "model.layers.27.mlp.experts.35.down_proj", "model.layers.27.mlp.experts.36.down_proj", "model.layers.27.mlp.experts.37.down_proj", "model.layers.27.mlp.experts.38.down_proj", "model.layers.27.mlp.experts.39.down_proj", "model.layers.27.mlp.experts.40.down_proj", "model.layers.27.mlp.experts.41.down_proj", "model.layers.27.mlp.experts.42.down_proj", "model.layers.27.mlp.experts.43.down_proj", "model.layers.27.mlp.experts.44.down_proj", "model.layers.27.mlp.experts.45.down_proj", "model.layers.27.mlp.experts.46.down_proj", "model.layers.27.mlp.experts.47.down_proj", "model.layers.27.mlp.experts.48.down_proj", "model.layers.27.mlp.experts.49.down_proj", "model.layers.27.mlp.experts.50.down_proj", "model.layers.27.mlp.experts.51.down_proj", "model.layers.27.mlp.experts.52.down_proj", "model.layers.27.mlp.experts.53.down_proj", "model.layers.27.mlp.experts.54.down_proj", "model.layers.27.mlp.experts.55.down_proj", "model.layers.27.mlp.experts.56.down_proj", "model.layers.27.mlp.experts.57.down_proj", "model.layers.27.mlp.experts.58.down_proj", "model.layers.27.mlp.experts.59.down_proj", "model.layers.27.mlp.experts.60.down_proj", "model.layers.27.mlp.experts.61.down_proj", "model.layers.27.mlp.experts.62.down_proj", "model.layers.27.mlp.experts.63.down_proj", "model.layers.27.mlp.experts.64.down_proj", "model.layers.27.mlp.experts.65.down_proj", "model.layers.27.mlp.experts.66.down_proj", "model.layers.27.mlp.experts.67.down_proj", "model.layers.27.mlp.experts.68.down_proj", "model.layers.27.mlp.experts.69.down_proj", "model.layers.27.mlp.experts.70.down_proj", "model.layers.27.mlp.experts.71.down_proj", "model.layers.27.mlp.experts.72.down_proj", "model.layers.27.mlp.experts.73.down_proj", "model.layers.27.mlp.experts.74.down_proj", "model.layers.27.mlp.experts.75.down_proj", "model.layers.27.mlp.experts.76.down_proj", "model.layers.27.mlp.experts.77.down_proj", "model.layers.27.mlp.experts.78.down_proj", "model.layers.27.mlp.experts.79.down_proj", "model.layers.27.mlp.experts.80.down_proj", "model.layers.27.mlp.experts.81.down_proj", "model.layers.27.mlp.experts.82.down_proj", "model.layers.27.mlp.experts.83.down_proj", "model.layers.27.mlp.experts.84.down_proj", "model.layers.27.mlp.experts.85.down_proj", "model.layers.27.mlp.experts.86.down_proj", "model.layers.27.mlp.experts.87.down_proj", "model.layers.27.mlp.experts.88.down_proj", "model.layers.27.mlp.experts.89.down_proj", "model.layers.27.mlp.experts.90.down_proj", "model.layers.27.mlp.experts.91.down_proj", "model.layers.27.mlp.experts.92.down_proj", "model.layers.27.mlp.experts.93.down_proj", "model.layers.27.mlp.experts.94.down_proj", "model.layers.27.mlp.experts.95.down_proj", "model.layers.27.mlp.experts.96.down_proj", "model.layers.27.mlp.experts.97.down_proj", "model.layers.27.mlp.experts.98.down_proj", "model.layers.27.mlp.experts.99.down_proj", "model.layers.27.mlp.experts.100.down_proj", "model.layers.27.mlp.experts.101.down_proj", "model.layers.27.mlp.experts.102.down_proj", "model.layers.27.mlp.experts.103.down_proj", "model.layers.27.mlp.experts.104.down_proj", "model.layers.27.mlp.experts.105.down_proj", "model.layers.27.mlp.experts.106.down_proj", "model.layers.27.mlp.experts.107.down_proj", "model.layers.27.mlp.experts.108.down_proj", "model.layers.27.mlp.experts.109.down_proj", "model.layers.27.mlp.experts.110.down_proj", "model.layers.27.mlp.experts.111.down_proj", "model.layers.27.mlp.experts.112.down_proj", "model.layers.27.mlp.experts.113.down_proj", "model.layers.27.mlp.experts.114.down_proj", "model.layers.27.mlp.experts.115.down_proj", "model.layers.27.mlp.experts.116.down_proj", "model.layers.27.mlp.experts.117.down_proj", "model.layers.27.mlp.experts.118.down_proj", "model.layers.27.mlp.experts.119.down_proj", "model.layers.27.mlp.experts.120.down_proj", "model.layers.27.mlp.experts.121.down_proj", "model.layers.27.mlp.experts.122.down_proj", "model.layers.27.mlp.experts.123.down_proj", "model.layers.27.mlp.experts.124.down_proj", "model.layers.27.mlp.experts.125.down_proj", "model.layers.27.mlp.experts.126.down_proj", "model.layers.27.mlp.experts.127.down_proj", "model.layers.27.mlp.experts.128.down_proj", "model.layers.27.mlp.experts.129.down_proj", "model.layers.27.mlp.experts.130.down_proj", "model.layers.27.mlp.experts.131.down_proj", "model.layers.27.mlp.experts.132.down_proj", "model.layers.27.mlp.experts.133.down_proj", "model.layers.27.mlp.experts.134.down_proj", "model.layers.27.mlp.experts.135.down_proj", "model.layers.27.mlp.experts.136.down_proj", "model.layers.27.mlp.experts.137.down_proj", "model.layers.27.mlp.experts.138.down_proj", "model.layers.27.mlp.experts.139.down_proj", "model.layers.27.mlp.experts.140.down_proj", "model.layers.27.mlp.experts.141.down_proj", "model.layers.27.mlp.experts.142.down_proj", "model.layers.27.mlp.experts.143.down_proj", "model.layers.27.mlp.experts.144.down_proj", "model.layers.27.mlp.experts.145.down_proj", "model.layers.27.mlp.experts.146.down_proj", "model.layers.27.mlp.experts.147.down_proj", "model.layers.27.mlp.experts.148.down_proj", "model.layers.27.mlp.experts.149.down_proj", "model.layers.27.mlp.experts.150.down_proj", "model.layers.27.mlp.experts.151.down_proj", "model.layers.27.mlp.experts.152.down_proj", "model.layers.27.mlp.experts.153.down_proj", "model.layers.27.mlp.experts.154.down_proj", "model.layers.27.mlp.experts.155.down_proj", "model.layers.27.mlp.experts.156.down_proj", "model.layers.27.mlp.experts.157.down_proj", "model.layers.27.mlp.experts.158.down_proj", "model.layers.27.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003717657178640421, "dbits": 1258291200 }, { "dkld": -0.0005179615691304207, "dbits": 2516582400 }, { "dkld": -0.0005550665780901937, "dbits": 3774873600 }, { "dkld": -0.0005133016034960802, "dbits": 6291456000 } ] }, { "idx": 165, "layers": [ "model.layers.28.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00010506678372622369, "dbits": 62914560 }, { "dkld": -0.00024088909849524776, "dbits": 125829120 }, { "dkld": -0.0002318659797310857, "dbits": 188743680 }, { "dkld": -0.00027934284880758165, "dbits": 314572800 } ] }, { "idx": 166, "layers": [ "model.layers.28.self_attn.k_proj", "model.layers.28.self_attn.v_proj" ], "candidates": [ { "dkld": 0.000800279900431633, "dbits": 10485760 }, { "dkld": 0.00048210881650447013, "dbits": 20971520 }, { "dkld": 0.0003096344880759716, "dbits": 31457280 }, { "dkld": 0.0003769686445593834, "dbits": 52428800 } ] }, { "idx": 167, "layers": [ "model.layers.28.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0006548946723342053, "dbits": 62914560 }, { "dkld": -0.0009450862184166992, "dbits": 125829120 }, { "dkld": -0.0010559773072600365, "dbits": 188743680 }, { "dkld": -0.001042729988694202, "dbits": 314572800 } ] }, { "idx": 168, "layers": [ "model.layers.28.mlp.shared_experts.gate_proj", "model.layers.28.mlp.shared_experts.up_proj", "model.layers.28.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00037573818117379865, "dbits": 23592960 }, { "dkld": 0.0002589661628007861, "dbits": 47185920 }, { "dkld": 3.351625055074414e-05, "dbits": 70778880 }, { "dkld": 0.00022198744118212543, "dbits": 117964800 } ] }, { "idx": 169, "layers": [ "model.layers.28.mlp.experts.0.gate_proj", "model.layers.28.mlp.experts.1.gate_proj", "model.layers.28.mlp.experts.2.gate_proj", "model.layers.28.mlp.experts.3.gate_proj", "model.layers.28.mlp.experts.4.gate_proj", "model.layers.28.mlp.experts.5.gate_proj", "model.layers.28.mlp.experts.6.gate_proj", "model.layers.28.mlp.experts.7.gate_proj", "model.layers.28.mlp.experts.8.gate_proj", "model.layers.28.mlp.experts.9.gate_proj", "model.layers.28.mlp.experts.10.gate_proj", "model.layers.28.mlp.experts.11.gate_proj", "model.layers.28.mlp.experts.12.gate_proj", "model.layers.28.mlp.experts.13.gate_proj", "model.layers.28.mlp.experts.14.gate_proj", "model.layers.28.mlp.experts.15.gate_proj", "model.layers.28.mlp.experts.16.gate_proj", "model.layers.28.mlp.experts.17.gate_proj", "model.layers.28.mlp.experts.18.gate_proj", "model.layers.28.mlp.experts.19.gate_proj", "model.layers.28.mlp.experts.20.gate_proj", "model.layers.28.mlp.experts.21.gate_proj", "model.layers.28.mlp.experts.22.gate_proj", "model.layers.28.mlp.experts.23.gate_proj", "model.layers.28.mlp.experts.24.gate_proj", "model.layers.28.mlp.experts.25.gate_proj", "model.layers.28.mlp.experts.26.gate_proj", "model.layers.28.mlp.experts.27.gate_proj", "model.layers.28.mlp.experts.28.gate_proj", "model.layers.28.mlp.experts.29.gate_proj", "model.layers.28.mlp.experts.30.gate_proj", "model.layers.28.mlp.experts.31.gate_proj", "model.layers.28.mlp.experts.32.gate_proj", "model.layers.28.mlp.experts.33.gate_proj", "model.layers.28.mlp.experts.34.gate_proj", "model.layers.28.mlp.experts.35.gate_proj", "model.layers.28.mlp.experts.36.gate_proj", "model.layers.28.mlp.experts.37.gate_proj", "model.layers.28.mlp.experts.38.gate_proj", "model.layers.28.mlp.experts.39.gate_proj", "model.layers.28.mlp.experts.40.gate_proj", "model.layers.28.mlp.experts.41.gate_proj", "model.layers.28.mlp.experts.42.gate_proj", "model.layers.28.mlp.experts.43.gate_proj", "model.layers.28.mlp.experts.44.gate_proj", "model.layers.28.mlp.experts.45.gate_proj", "model.layers.28.mlp.experts.46.gate_proj", "model.layers.28.mlp.experts.47.gate_proj", "model.layers.28.mlp.experts.48.gate_proj", "model.layers.28.mlp.experts.49.gate_proj", "model.layers.28.mlp.experts.50.gate_proj", "model.layers.28.mlp.experts.51.gate_proj", "model.layers.28.mlp.experts.52.gate_proj", "model.layers.28.mlp.experts.53.gate_proj", "model.layers.28.mlp.experts.54.gate_proj", "model.layers.28.mlp.experts.55.gate_proj", "model.layers.28.mlp.experts.56.gate_proj", "model.layers.28.mlp.experts.57.gate_proj", "model.layers.28.mlp.experts.58.gate_proj", "model.layers.28.mlp.experts.59.gate_proj", "model.layers.28.mlp.experts.60.gate_proj", "model.layers.28.mlp.experts.61.gate_proj", "model.layers.28.mlp.experts.62.gate_proj", "model.layers.28.mlp.experts.63.gate_proj", "model.layers.28.mlp.experts.64.gate_proj", "model.layers.28.mlp.experts.65.gate_proj", "model.layers.28.mlp.experts.66.gate_proj", "model.layers.28.mlp.experts.67.gate_proj", "model.layers.28.mlp.experts.68.gate_proj", "model.layers.28.mlp.experts.69.gate_proj", "model.layers.28.mlp.experts.70.gate_proj", "model.layers.28.mlp.experts.71.gate_proj", "model.layers.28.mlp.experts.72.gate_proj", "model.layers.28.mlp.experts.73.gate_proj", "model.layers.28.mlp.experts.74.gate_proj", "model.layers.28.mlp.experts.75.gate_proj", "model.layers.28.mlp.experts.76.gate_proj", "model.layers.28.mlp.experts.77.gate_proj", "model.layers.28.mlp.experts.78.gate_proj", "model.layers.28.mlp.experts.79.gate_proj", "model.layers.28.mlp.experts.80.gate_proj", "model.layers.28.mlp.experts.81.gate_proj", "model.layers.28.mlp.experts.82.gate_proj", "model.layers.28.mlp.experts.83.gate_proj", "model.layers.28.mlp.experts.84.gate_proj", "model.layers.28.mlp.experts.85.gate_proj", "model.layers.28.mlp.experts.86.gate_proj", "model.layers.28.mlp.experts.87.gate_proj", "model.layers.28.mlp.experts.88.gate_proj", "model.layers.28.mlp.experts.89.gate_proj", "model.layers.28.mlp.experts.90.gate_proj", "model.layers.28.mlp.experts.91.gate_proj", "model.layers.28.mlp.experts.92.gate_proj", "model.layers.28.mlp.experts.93.gate_proj", "model.layers.28.mlp.experts.94.gate_proj", "model.layers.28.mlp.experts.95.gate_proj", "model.layers.28.mlp.experts.96.gate_proj", "model.layers.28.mlp.experts.97.gate_proj", "model.layers.28.mlp.experts.98.gate_proj", "model.layers.28.mlp.experts.99.gate_proj", "model.layers.28.mlp.experts.100.gate_proj", "model.layers.28.mlp.experts.101.gate_proj", "model.layers.28.mlp.experts.102.gate_proj", "model.layers.28.mlp.experts.103.gate_proj", "model.layers.28.mlp.experts.104.gate_proj", "model.layers.28.mlp.experts.105.gate_proj", "model.layers.28.mlp.experts.106.gate_proj", "model.layers.28.mlp.experts.107.gate_proj", "model.layers.28.mlp.experts.108.gate_proj", "model.layers.28.mlp.experts.109.gate_proj", "model.layers.28.mlp.experts.110.gate_proj", "model.layers.28.mlp.experts.111.gate_proj", "model.layers.28.mlp.experts.112.gate_proj", "model.layers.28.mlp.experts.113.gate_proj", "model.layers.28.mlp.experts.114.gate_proj", "model.layers.28.mlp.experts.115.gate_proj", "model.layers.28.mlp.experts.116.gate_proj", "model.layers.28.mlp.experts.117.gate_proj", "model.layers.28.mlp.experts.118.gate_proj", "model.layers.28.mlp.experts.119.gate_proj", "model.layers.28.mlp.experts.120.gate_proj", "model.layers.28.mlp.experts.121.gate_proj", "model.layers.28.mlp.experts.122.gate_proj", "model.layers.28.mlp.experts.123.gate_proj", "model.layers.28.mlp.experts.124.gate_proj", "model.layers.28.mlp.experts.125.gate_proj", "model.layers.28.mlp.experts.126.gate_proj", "model.layers.28.mlp.experts.127.gate_proj", "model.layers.28.mlp.experts.128.gate_proj", "model.layers.28.mlp.experts.129.gate_proj", "model.layers.28.mlp.experts.130.gate_proj", "model.layers.28.mlp.experts.131.gate_proj", "model.layers.28.mlp.experts.132.gate_proj", "model.layers.28.mlp.experts.133.gate_proj", "model.layers.28.mlp.experts.134.gate_proj", "model.layers.28.mlp.experts.135.gate_proj", "model.layers.28.mlp.experts.136.gate_proj", "model.layers.28.mlp.experts.137.gate_proj", "model.layers.28.mlp.experts.138.gate_proj", "model.layers.28.mlp.experts.139.gate_proj", "model.layers.28.mlp.experts.140.gate_proj", "model.layers.28.mlp.experts.141.gate_proj", "model.layers.28.mlp.experts.142.gate_proj", "model.layers.28.mlp.experts.143.gate_proj", "model.layers.28.mlp.experts.144.gate_proj", "model.layers.28.mlp.experts.145.gate_proj", "model.layers.28.mlp.experts.146.gate_proj", "model.layers.28.mlp.experts.147.gate_proj", "model.layers.28.mlp.experts.148.gate_proj", "model.layers.28.mlp.experts.149.gate_proj", "model.layers.28.mlp.experts.150.gate_proj", "model.layers.28.mlp.experts.151.gate_proj", "model.layers.28.mlp.experts.152.gate_proj", "model.layers.28.mlp.experts.153.gate_proj", "model.layers.28.mlp.experts.154.gate_proj", "model.layers.28.mlp.experts.155.gate_proj", "model.layers.28.mlp.experts.156.gate_proj", "model.layers.28.mlp.experts.157.gate_proj", "model.layers.28.mlp.experts.158.gate_proj", "model.layers.28.mlp.experts.159.gate_proj", "model.layers.28.mlp.experts.0.up_proj", "model.layers.28.mlp.experts.1.up_proj", "model.layers.28.mlp.experts.2.up_proj", "model.layers.28.mlp.experts.3.up_proj", "model.layers.28.mlp.experts.4.up_proj", "model.layers.28.mlp.experts.5.up_proj", "model.layers.28.mlp.experts.6.up_proj", "model.layers.28.mlp.experts.7.up_proj", "model.layers.28.mlp.experts.8.up_proj", "model.layers.28.mlp.experts.9.up_proj", "model.layers.28.mlp.experts.10.up_proj", "model.layers.28.mlp.experts.11.up_proj", "model.layers.28.mlp.experts.12.up_proj", "model.layers.28.mlp.experts.13.up_proj", "model.layers.28.mlp.experts.14.up_proj", "model.layers.28.mlp.experts.15.up_proj", "model.layers.28.mlp.experts.16.up_proj", "model.layers.28.mlp.experts.17.up_proj", "model.layers.28.mlp.experts.18.up_proj", "model.layers.28.mlp.experts.19.up_proj", "model.layers.28.mlp.experts.20.up_proj", "model.layers.28.mlp.experts.21.up_proj", "model.layers.28.mlp.experts.22.up_proj", "model.layers.28.mlp.experts.23.up_proj", "model.layers.28.mlp.experts.24.up_proj", "model.layers.28.mlp.experts.25.up_proj", "model.layers.28.mlp.experts.26.up_proj", "model.layers.28.mlp.experts.27.up_proj", "model.layers.28.mlp.experts.28.up_proj", "model.layers.28.mlp.experts.29.up_proj", "model.layers.28.mlp.experts.30.up_proj", "model.layers.28.mlp.experts.31.up_proj", "model.layers.28.mlp.experts.32.up_proj", "model.layers.28.mlp.experts.33.up_proj", "model.layers.28.mlp.experts.34.up_proj", "model.layers.28.mlp.experts.35.up_proj", "model.layers.28.mlp.experts.36.up_proj", "model.layers.28.mlp.experts.37.up_proj", "model.layers.28.mlp.experts.38.up_proj", "model.layers.28.mlp.experts.39.up_proj", "model.layers.28.mlp.experts.40.up_proj", "model.layers.28.mlp.experts.41.up_proj", "model.layers.28.mlp.experts.42.up_proj", "model.layers.28.mlp.experts.43.up_proj", "model.layers.28.mlp.experts.44.up_proj", "model.layers.28.mlp.experts.45.up_proj", "model.layers.28.mlp.experts.46.up_proj", "model.layers.28.mlp.experts.47.up_proj", "model.layers.28.mlp.experts.48.up_proj", "model.layers.28.mlp.experts.49.up_proj", "model.layers.28.mlp.experts.50.up_proj", "model.layers.28.mlp.experts.51.up_proj", "model.layers.28.mlp.experts.52.up_proj", "model.layers.28.mlp.experts.53.up_proj", "model.layers.28.mlp.experts.54.up_proj", "model.layers.28.mlp.experts.55.up_proj", "model.layers.28.mlp.experts.56.up_proj", "model.layers.28.mlp.experts.57.up_proj", "model.layers.28.mlp.experts.58.up_proj", "model.layers.28.mlp.experts.59.up_proj", "model.layers.28.mlp.experts.60.up_proj", "model.layers.28.mlp.experts.61.up_proj", "model.layers.28.mlp.experts.62.up_proj", "model.layers.28.mlp.experts.63.up_proj", "model.layers.28.mlp.experts.64.up_proj", "model.layers.28.mlp.experts.65.up_proj", "model.layers.28.mlp.experts.66.up_proj", "model.layers.28.mlp.experts.67.up_proj", "model.layers.28.mlp.experts.68.up_proj", "model.layers.28.mlp.experts.69.up_proj", "model.layers.28.mlp.experts.70.up_proj", "model.layers.28.mlp.experts.71.up_proj", "model.layers.28.mlp.experts.72.up_proj", "model.layers.28.mlp.experts.73.up_proj", "model.layers.28.mlp.experts.74.up_proj", "model.layers.28.mlp.experts.75.up_proj", "model.layers.28.mlp.experts.76.up_proj", "model.layers.28.mlp.experts.77.up_proj", "model.layers.28.mlp.experts.78.up_proj", "model.layers.28.mlp.experts.79.up_proj", "model.layers.28.mlp.experts.80.up_proj", "model.layers.28.mlp.experts.81.up_proj", "model.layers.28.mlp.experts.82.up_proj", "model.layers.28.mlp.experts.83.up_proj", "model.layers.28.mlp.experts.84.up_proj", "model.layers.28.mlp.experts.85.up_proj", "model.layers.28.mlp.experts.86.up_proj", "model.layers.28.mlp.experts.87.up_proj", "model.layers.28.mlp.experts.88.up_proj", "model.layers.28.mlp.experts.89.up_proj", "model.layers.28.mlp.experts.90.up_proj", "model.layers.28.mlp.experts.91.up_proj", "model.layers.28.mlp.experts.92.up_proj", "model.layers.28.mlp.experts.93.up_proj", "model.layers.28.mlp.experts.94.up_proj", "model.layers.28.mlp.experts.95.up_proj", "model.layers.28.mlp.experts.96.up_proj", "model.layers.28.mlp.experts.97.up_proj", "model.layers.28.mlp.experts.98.up_proj", "model.layers.28.mlp.experts.99.up_proj", "model.layers.28.mlp.experts.100.up_proj", "model.layers.28.mlp.experts.101.up_proj", "model.layers.28.mlp.experts.102.up_proj", "model.layers.28.mlp.experts.103.up_proj", "model.layers.28.mlp.experts.104.up_proj", "model.layers.28.mlp.experts.105.up_proj", "model.layers.28.mlp.experts.106.up_proj", "model.layers.28.mlp.experts.107.up_proj", "model.layers.28.mlp.experts.108.up_proj", "model.layers.28.mlp.experts.109.up_proj", "model.layers.28.mlp.experts.110.up_proj", "model.layers.28.mlp.experts.111.up_proj", "model.layers.28.mlp.experts.112.up_proj", "model.layers.28.mlp.experts.113.up_proj", "model.layers.28.mlp.experts.114.up_proj", "model.layers.28.mlp.experts.115.up_proj", "model.layers.28.mlp.experts.116.up_proj", "model.layers.28.mlp.experts.117.up_proj", "model.layers.28.mlp.experts.118.up_proj", "model.layers.28.mlp.experts.119.up_proj", "model.layers.28.mlp.experts.120.up_proj", "model.layers.28.mlp.experts.121.up_proj", "model.layers.28.mlp.experts.122.up_proj", "model.layers.28.mlp.experts.123.up_proj", "model.layers.28.mlp.experts.124.up_proj", "model.layers.28.mlp.experts.125.up_proj", "model.layers.28.mlp.experts.126.up_proj", "model.layers.28.mlp.experts.127.up_proj", "model.layers.28.mlp.experts.128.up_proj", "model.layers.28.mlp.experts.129.up_proj", "model.layers.28.mlp.experts.130.up_proj", "model.layers.28.mlp.experts.131.up_proj", "model.layers.28.mlp.experts.132.up_proj", "model.layers.28.mlp.experts.133.up_proj", "model.layers.28.mlp.experts.134.up_proj", "model.layers.28.mlp.experts.135.up_proj", "model.layers.28.mlp.experts.136.up_proj", "model.layers.28.mlp.experts.137.up_proj", "model.layers.28.mlp.experts.138.up_proj", "model.layers.28.mlp.experts.139.up_proj", "model.layers.28.mlp.experts.140.up_proj", "model.layers.28.mlp.experts.141.up_proj", "model.layers.28.mlp.experts.142.up_proj", "model.layers.28.mlp.experts.143.up_proj", "model.layers.28.mlp.experts.144.up_proj", "model.layers.28.mlp.experts.145.up_proj", "model.layers.28.mlp.experts.146.up_proj", "model.layers.28.mlp.experts.147.up_proj", "model.layers.28.mlp.experts.148.up_proj", "model.layers.28.mlp.experts.149.up_proj", "model.layers.28.mlp.experts.150.up_proj", "model.layers.28.mlp.experts.151.up_proj", "model.layers.28.mlp.experts.152.up_proj", "model.layers.28.mlp.experts.153.up_proj", "model.layers.28.mlp.experts.154.up_proj", "model.layers.28.mlp.experts.155.up_proj", "model.layers.28.mlp.experts.156.up_proj", "model.layers.28.mlp.experts.157.up_proj", "model.layers.28.mlp.experts.158.up_proj", "model.layers.28.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0008727868087589769, "dbits": 2516582400 }, { "dkld": -0.0008292215876281317, "dbits": 5033164800 }, { "dkld": -0.0009351437911391258, "dbits": 7549747200 }, { "dkld": -0.0009454556740820519, "dbits": 12582912000 } ] }, { "idx": 170, "layers": [ "model.layers.28.mlp.experts.0.down_proj", "model.layers.28.mlp.experts.1.down_proj", "model.layers.28.mlp.experts.2.down_proj", "model.layers.28.mlp.experts.3.down_proj", "model.layers.28.mlp.experts.4.down_proj", "model.layers.28.mlp.experts.5.down_proj", "model.layers.28.mlp.experts.6.down_proj", "model.layers.28.mlp.experts.7.down_proj", "model.layers.28.mlp.experts.8.down_proj", "model.layers.28.mlp.experts.9.down_proj", "model.layers.28.mlp.experts.10.down_proj", "model.layers.28.mlp.experts.11.down_proj", "model.layers.28.mlp.experts.12.down_proj", "model.layers.28.mlp.experts.13.down_proj", "model.layers.28.mlp.experts.14.down_proj", "model.layers.28.mlp.experts.15.down_proj", "model.layers.28.mlp.experts.16.down_proj", "model.layers.28.mlp.experts.17.down_proj", "model.layers.28.mlp.experts.18.down_proj", "model.layers.28.mlp.experts.19.down_proj", "model.layers.28.mlp.experts.20.down_proj", "model.layers.28.mlp.experts.21.down_proj", "model.layers.28.mlp.experts.22.down_proj", "model.layers.28.mlp.experts.23.down_proj", "model.layers.28.mlp.experts.24.down_proj", "model.layers.28.mlp.experts.25.down_proj", "model.layers.28.mlp.experts.26.down_proj", "model.layers.28.mlp.experts.27.down_proj", "model.layers.28.mlp.experts.28.down_proj", "model.layers.28.mlp.experts.29.down_proj", "model.layers.28.mlp.experts.30.down_proj", "model.layers.28.mlp.experts.31.down_proj", "model.layers.28.mlp.experts.32.down_proj", "model.layers.28.mlp.experts.33.down_proj", "model.layers.28.mlp.experts.34.down_proj", "model.layers.28.mlp.experts.35.down_proj", "model.layers.28.mlp.experts.36.down_proj", "model.layers.28.mlp.experts.37.down_proj", "model.layers.28.mlp.experts.38.down_proj", "model.layers.28.mlp.experts.39.down_proj", "model.layers.28.mlp.experts.40.down_proj", "model.layers.28.mlp.experts.41.down_proj", "model.layers.28.mlp.experts.42.down_proj", "model.layers.28.mlp.experts.43.down_proj", "model.layers.28.mlp.experts.44.down_proj", "model.layers.28.mlp.experts.45.down_proj", "model.layers.28.mlp.experts.46.down_proj", "model.layers.28.mlp.experts.47.down_proj", "model.layers.28.mlp.experts.48.down_proj", "model.layers.28.mlp.experts.49.down_proj", "model.layers.28.mlp.experts.50.down_proj", "model.layers.28.mlp.experts.51.down_proj", "model.layers.28.mlp.experts.52.down_proj", "model.layers.28.mlp.experts.53.down_proj", "model.layers.28.mlp.experts.54.down_proj", "model.layers.28.mlp.experts.55.down_proj", "model.layers.28.mlp.experts.56.down_proj", "model.layers.28.mlp.experts.57.down_proj", "model.layers.28.mlp.experts.58.down_proj", "model.layers.28.mlp.experts.59.down_proj", "model.layers.28.mlp.experts.60.down_proj", "model.layers.28.mlp.experts.61.down_proj", "model.layers.28.mlp.experts.62.down_proj", "model.layers.28.mlp.experts.63.down_proj", "model.layers.28.mlp.experts.64.down_proj", "model.layers.28.mlp.experts.65.down_proj", "model.layers.28.mlp.experts.66.down_proj", "model.layers.28.mlp.experts.67.down_proj", "model.layers.28.mlp.experts.68.down_proj", "model.layers.28.mlp.experts.69.down_proj", "model.layers.28.mlp.experts.70.down_proj", "model.layers.28.mlp.experts.71.down_proj", "model.layers.28.mlp.experts.72.down_proj", "model.layers.28.mlp.experts.73.down_proj", "model.layers.28.mlp.experts.74.down_proj", "model.layers.28.mlp.experts.75.down_proj", "model.layers.28.mlp.experts.76.down_proj", "model.layers.28.mlp.experts.77.down_proj", "model.layers.28.mlp.experts.78.down_proj", "model.layers.28.mlp.experts.79.down_proj", "model.layers.28.mlp.experts.80.down_proj", "model.layers.28.mlp.experts.81.down_proj", "model.layers.28.mlp.experts.82.down_proj", "model.layers.28.mlp.experts.83.down_proj", "model.layers.28.mlp.experts.84.down_proj", "model.layers.28.mlp.experts.85.down_proj", "model.layers.28.mlp.experts.86.down_proj", "model.layers.28.mlp.experts.87.down_proj", "model.layers.28.mlp.experts.88.down_proj", "model.layers.28.mlp.experts.89.down_proj", "model.layers.28.mlp.experts.90.down_proj", "model.layers.28.mlp.experts.91.down_proj", "model.layers.28.mlp.experts.92.down_proj", "model.layers.28.mlp.experts.93.down_proj", "model.layers.28.mlp.experts.94.down_proj", "model.layers.28.mlp.experts.95.down_proj", "model.layers.28.mlp.experts.96.down_proj", "model.layers.28.mlp.experts.97.down_proj", "model.layers.28.mlp.experts.98.down_proj", "model.layers.28.mlp.experts.99.down_proj", "model.layers.28.mlp.experts.100.down_proj", "model.layers.28.mlp.experts.101.down_proj", "model.layers.28.mlp.experts.102.down_proj", "model.layers.28.mlp.experts.103.down_proj", "model.layers.28.mlp.experts.104.down_proj", "model.layers.28.mlp.experts.105.down_proj", "model.layers.28.mlp.experts.106.down_proj", "model.layers.28.mlp.experts.107.down_proj", "model.layers.28.mlp.experts.108.down_proj", "model.layers.28.mlp.experts.109.down_proj", "model.layers.28.mlp.experts.110.down_proj", "model.layers.28.mlp.experts.111.down_proj", "model.layers.28.mlp.experts.112.down_proj", "model.layers.28.mlp.experts.113.down_proj", "model.layers.28.mlp.experts.114.down_proj", "model.layers.28.mlp.experts.115.down_proj", "model.layers.28.mlp.experts.116.down_proj", "model.layers.28.mlp.experts.117.down_proj", "model.layers.28.mlp.experts.118.down_proj", "model.layers.28.mlp.experts.119.down_proj", "model.layers.28.mlp.experts.120.down_proj", "model.layers.28.mlp.experts.121.down_proj", "model.layers.28.mlp.experts.122.down_proj", "model.layers.28.mlp.experts.123.down_proj", "model.layers.28.mlp.experts.124.down_proj", "model.layers.28.mlp.experts.125.down_proj", "model.layers.28.mlp.experts.126.down_proj", "model.layers.28.mlp.experts.127.down_proj", "model.layers.28.mlp.experts.128.down_proj", "model.layers.28.mlp.experts.129.down_proj", "model.layers.28.mlp.experts.130.down_proj", "model.layers.28.mlp.experts.131.down_proj", "model.layers.28.mlp.experts.132.down_proj", "model.layers.28.mlp.experts.133.down_proj", "model.layers.28.mlp.experts.134.down_proj", "model.layers.28.mlp.experts.135.down_proj", "model.layers.28.mlp.experts.136.down_proj", "model.layers.28.mlp.experts.137.down_proj", "model.layers.28.mlp.experts.138.down_proj", "model.layers.28.mlp.experts.139.down_proj", "model.layers.28.mlp.experts.140.down_proj", "model.layers.28.mlp.experts.141.down_proj", "model.layers.28.mlp.experts.142.down_proj", "model.layers.28.mlp.experts.143.down_proj", "model.layers.28.mlp.experts.144.down_proj", "model.layers.28.mlp.experts.145.down_proj", "model.layers.28.mlp.experts.146.down_proj", "model.layers.28.mlp.experts.147.down_proj", "model.layers.28.mlp.experts.148.down_proj", "model.layers.28.mlp.experts.149.down_proj", "model.layers.28.mlp.experts.150.down_proj", "model.layers.28.mlp.experts.151.down_proj", "model.layers.28.mlp.experts.152.down_proj", "model.layers.28.mlp.experts.153.down_proj", "model.layers.28.mlp.experts.154.down_proj", "model.layers.28.mlp.experts.155.down_proj", "model.layers.28.mlp.experts.156.down_proj", "model.layers.28.mlp.experts.157.down_proj", "model.layers.28.mlp.experts.158.down_proj", "model.layers.28.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003283786587417181, "dbits": 1258291200 }, { "dkld": -0.00047243032604456225, "dbits": 2516582400 }, { "dkld": -0.0004228253848850838, "dbits": 3774873600 }, { "dkld": -0.00047037154436112283, "dbits": 6291456000 } ] }, { "idx": 171, "layers": [ "model.layers.29.self_attn.q_proj" ], "candidates": [ { "dkld": -6.368421018124182e-05, "dbits": 62914560 }, { "dkld": 3.175847232340934e-05, "dbits": 125829120 }, { "dkld": 5.1161088049409e-05, "dbits": 188743680 }, { "dkld": 0.0001946140080690273, "dbits": 314572800 } ] }, { "idx": 172, "layers": [ "model.layers.29.self_attn.k_proj", "model.layers.29.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0005481723695993507, "dbits": 10485760 }, { "dkld": -8.270544931293644e-05, "dbits": 20971520 }, { "dkld": -0.000482110120356094, "dbits": 31457280 }, { "dkld": -0.0004218490794301033, "dbits": 52428800 } ] }, { "idx": 173, "layers": [ "model.layers.29.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00039868270978332676, "dbits": 62914560 }, { "dkld": -0.0006228185258805863, "dbits": 125829120 }, { "dkld": -0.0005558955483138672, "dbits": 188743680 }, { "dkld": -0.000570737477391961, "dbits": 314572800 } ] }, { "idx": 174, "layers": [ "model.layers.29.mlp.shared_experts.gate_proj", "model.layers.29.mlp.shared_experts.up_proj", "model.layers.29.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0009144671261310661, "dbits": 23592960 }, { "dkld": -0.0007649328559637125, "dbits": 47185920 }, { "dkld": -0.0006620204076170921, "dbits": 70778880 }, { "dkld": -0.0007229341194033706, "dbits": 117964800 } ] }, { "idx": 175, "layers": [ "model.layers.29.mlp.experts.0.gate_proj", "model.layers.29.mlp.experts.1.gate_proj", "model.layers.29.mlp.experts.2.gate_proj", "model.layers.29.mlp.experts.3.gate_proj", "model.layers.29.mlp.experts.4.gate_proj", "model.layers.29.mlp.experts.5.gate_proj", "model.layers.29.mlp.experts.6.gate_proj", "model.layers.29.mlp.experts.7.gate_proj", "model.layers.29.mlp.experts.8.gate_proj", "model.layers.29.mlp.experts.9.gate_proj", "model.layers.29.mlp.experts.10.gate_proj", "model.layers.29.mlp.experts.11.gate_proj", "model.layers.29.mlp.experts.12.gate_proj", "model.layers.29.mlp.experts.13.gate_proj", "model.layers.29.mlp.experts.14.gate_proj", "model.layers.29.mlp.experts.15.gate_proj", "model.layers.29.mlp.experts.16.gate_proj", "model.layers.29.mlp.experts.17.gate_proj", "model.layers.29.mlp.experts.18.gate_proj", "model.layers.29.mlp.experts.19.gate_proj", "model.layers.29.mlp.experts.20.gate_proj", "model.layers.29.mlp.experts.21.gate_proj", "model.layers.29.mlp.experts.22.gate_proj", "model.layers.29.mlp.experts.23.gate_proj", "model.layers.29.mlp.experts.24.gate_proj", "model.layers.29.mlp.experts.25.gate_proj", "model.layers.29.mlp.experts.26.gate_proj", "model.layers.29.mlp.experts.27.gate_proj", "model.layers.29.mlp.experts.28.gate_proj", "model.layers.29.mlp.experts.29.gate_proj", "model.layers.29.mlp.experts.30.gate_proj", "model.layers.29.mlp.experts.31.gate_proj", "model.layers.29.mlp.experts.32.gate_proj", "model.layers.29.mlp.experts.33.gate_proj", "model.layers.29.mlp.experts.34.gate_proj", "model.layers.29.mlp.experts.35.gate_proj", "model.layers.29.mlp.experts.36.gate_proj", "model.layers.29.mlp.experts.37.gate_proj", "model.layers.29.mlp.experts.38.gate_proj", "model.layers.29.mlp.experts.39.gate_proj", "model.layers.29.mlp.experts.40.gate_proj", "model.layers.29.mlp.experts.41.gate_proj", "model.layers.29.mlp.experts.42.gate_proj", "model.layers.29.mlp.experts.43.gate_proj", "model.layers.29.mlp.experts.44.gate_proj", "model.layers.29.mlp.experts.45.gate_proj", "model.layers.29.mlp.experts.46.gate_proj", "model.layers.29.mlp.experts.47.gate_proj", "model.layers.29.mlp.experts.48.gate_proj", "model.layers.29.mlp.experts.49.gate_proj", "model.layers.29.mlp.experts.50.gate_proj", "model.layers.29.mlp.experts.51.gate_proj", "model.layers.29.mlp.experts.52.gate_proj", "model.layers.29.mlp.experts.53.gate_proj", "model.layers.29.mlp.experts.54.gate_proj", "model.layers.29.mlp.experts.55.gate_proj", "model.layers.29.mlp.experts.56.gate_proj", "model.layers.29.mlp.experts.57.gate_proj", "model.layers.29.mlp.experts.58.gate_proj", "model.layers.29.mlp.experts.59.gate_proj", "model.layers.29.mlp.experts.60.gate_proj", "model.layers.29.mlp.experts.61.gate_proj", "model.layers.29.mlp.experts.62.gate_proj", "model.layers.29.mlp.experts.63.gate_proj", "model.layers.29.mlp.experts.64.gate_proj", "model.layers.29.mlp.experts.65.gate_proj", "model.layers.29.mlp.experts.66.gate_proj", "model.layers.29.mlp.experts.67.gate_proj", "model.layers.29.mlp.experts.68.gate_proj", "model.layers.29.mlp.experts.69.gate_proj", "model.layers.29.mlp.experts.70.gate_proj", "model.layers.29.mlp.experts.71.gate_proj", "model.layers.29.mlp.experts.72.gate_proj", "model.layers.29.mlp.experts.73.gate_proj", "model.layers.29.mlp.experts.74.gate_proj", "model.layers.29.mlp.experts.75.gate_proj", "model.layers.29.mlp.experts.76.gate_proj", "model.layers.29.mlp.experts.77.gate_proj", "model.layers.29.mlp.experts.78.gate_proj", "model.layers.29.mlp.experts.79.gate_proj", "model.layers.29.mlp.experts.80.gate_proj", "model.layers.29.mlp.experts.81.gate_proj", "model.layers.29.mlp.experts.82.gate_proj", "model.layers.29.mlp.experts.83.gate_proj", "model.layers.29.mlp.experts.84.gate_proj", "model.layers.29.mlp.experts.85.gate_proj", "model.layers.29.mlp.experts.86.gate_proj", "model.layers.29.mlp.experts.87.gate_proj", "model.layers.29.mlp.experts.88.gate_proj", "model.layers.29.mlp.experts.89.gate_proj", "model.layers.29.mlp.experts.90.gate_proj", "model.layers.29.mlp.experts.91.gate_proj", "model.layers.29.mlp.experts.92.gate_proj", "model.layers.29.mlp.experts.93.gate_proj", "model.layers.29.mlp.experts.94.gate_proj", "model.layers.29.mlp.experts.95.gate_proj", "model.layers.29.mlp.experts.96.gate_proj", "model.layers.29.mlp.experts.97.gate_proj", "model.layers.29.mlp.experts.98.gate_proj", "model.layers.29.mlp.experts.99.gate_proj", "model.layers.29.mlp.experts.100.gate_proj", "model.layers.29.mlp.experts.101.gate_proj", "model.layers.29.mlp.experts.102.gate_proj", "model.layers.29.mlp.experts.103.gate_proj", "model.layers.29.mlp.experts.104.gate_proj", "model.layers.29.mlp.experts.105.gate_proj", "model.layers.29.mlp.experts.106.gate_proj", "model.layers.29.mlp.experts.107.gate_proj", "model.layers.29.mlp.experts.108.gate_proj", "model.layers.29.mlp.experts.109.gate_proj", "model.layers.29.mlp.experts.110.gate_proj", "model.layers.29.mlp.experts.111.gate_proj", "model.layers.29.mlp.experts.112.gate_proj", "model.layers.29.mlp.experts.113.gate_proj", "model.layers.29.mlp.experts.114.gate_proj", "model.layers.29.mlp.experts.115.gate_proj", "model.layers.29.mlp.experts.116.gate_proj", "model.layers.29.mlp.experts.117.gate_proj", "model.layers.29.mlp.experts.118.gate_proj", "model.layers.29.mlp.experts.119.gate_proj", "model.layers.29.mlp.experts.120.gate_proj", "model.layers.29.mlp.experts.121.gate_proj", "model.layers.29.mlp.experts.122.gate_proj", "model.layers.29.mlp.experts.123.gate_proj", "model.layers.29.mlp.experts.124.gate_proj", "model.layers.29.mlp.experts.125.gate_proj", "model.layers.29.mlp.experts.126.gate_proj", "model.layers.29.mlp.experts.127.gate_proj", "model.layers.29.mlp.experts.128.gate_proj", "model.layers.29.mlp.experts.129.gate_proj", "model.layers.29.mlp.experts.130.gate_proj", "model.layers.29.mlp.experts.131.gate_proj", "model.layers.29.mlp.experts.132.gate_proj", "model.layers.29.mlp.experts.133.gate_proj", "model.layers.29.mlp.experts.134.gate_proj", "model.layers.29.mlp.experts.135.gate_proj", "model.layers.29.mlp.experts.136.gate_proj", "model.layers.29.mlp.experts.137.gate_proj", "model.layers.29.mlp.experts.138.gate_proj", "model.layers.29.mlp.experts.139.gate_proj", "model.layers.29.mlp.experts.140.gate_proj", "model.layers.29.mlp.experts.141.gate_proj", "model.layers.29.mlp.experts.142.gate_proj", "model.layers.29.mlp.experts.143.gate_proj", "model.layers.29.mlp.experts.144.gate_proj", "model.layers.29.mlp.experts.145.gate_proj", "model.layers.29.mlp.experts.146.gate_proj", "model.layers.29.mlp.experts.147.gate_proj", "model.layers.29.mlp.experts.148.gate_proj", "model.layers.29.mlp.experts.149.gate_proj", "model.layers.29.mlp.experts.150.gate_proj", "model.layers.29.mlp.experts.151.gate_proj", "model.layers.29.mlp.experts.152.gate_proj", "model.layers.29.mlp.experts.153.gate_proj", "model.layers.29.mlp.experts.154.gate_proj", "model.layers.29.mlp.experts.155.gate_proj", "model.layers.29.mlp.experts.156.gate_proj", "model.layers.29.mlp.experts.157.gate_proj", "model.layers.29.mlp.experts.158.gate_proj", "model.layers.29.mlp.experts.159.gate_proj", "model.layers.29.mlp.experts.0.up_proj", "model.layers.29.mlp.experts.1.up_proj", "model.layers.29.mlp.experts.2.up_proj", "model.layers.29.mlp.experts.3.up_proj", "model.layers.29.mlp.experts.4.up_proj", "model.layers.29.mlp.experts.5.up_proj", "model.layers.29.mlp.experts.6.up_proj", "model.layers.29.mlp.experts.7.up_proj", "model.layers.29.mlp.experts.8.up_proj", "model.layers.29.mlp.experts.9.up_proj", "model.layers.29.mlp.experts.10.up_proj", "model.layers.29.mlp.experts.11.up_proj", "model.layers.29.mlp.experts.12.up_proj", "model.layers.29.mlp.experts.13.up_proj", "model.layers.29.mlp.experts.14.up_proj", "model.layers.29.mlp.experts.15.up_proj", "model.layers.29.mlp.experts.16.up_proj", "model.layers.29.mlp.experts.17.up_proj", "model.layers.29.mlp.experts.18.up_proj", "model.layers.29.mlp.experts.19.up_proj", "model.layers.29.mlp.experts.20.up_proj", "model.layers.29.mlp.experts.21.up_proj", "model.layers.29.mlp.experts.22.up_proj", "model.layers.29.mlp.experts.23.up_proj", "model.layers.29.mlp.experts.24.up_proj", "model.layers.29.mlp.experts.25.up_proj", "model.layers.29.mlp.experts.26.up_proj", "model.layers.29.mlp.experts.27.up_proj", "model.layers.29.mlp.experts.28.up_proj", "model.layers.29.mlp.experts.29.up_proj", "model.layers.29.mlp.experts.30.up_proj", "model.layers.29.mlp.experts.31.up_proj", "model.layers.29.mlp.experts.32.up_proj", "model.layers.29.mlp.experts.33.up_proj", "model.layers.29.mlp.experts.34.up_proj", "model.layers.29.mlp.experts.35.up_proj", "model.layers.29.mlp.experts.36.up_proj", "model.layers.29.mlp.experts.37.up_proj", "model.layers.29.mlp.experts.38.up_proj", "model.layers.29.mlp.experts.39.up_proj", "model.layers.29.mlp.experts.40.up_proj", "model.layers.29.mlp.experts.41.up_proj", "model.layers.29.mlp.experts.42.up_proj", "model.layers.29.mlp.experts.43.up_proj", "model.layers.29.mlp.experts.44.up_proj", "model.layers.29.mlp.experts.45.up_proj", "model.layers.29.mlp.experts.46.up_proj", "model.layers.29.mlp.experts.47.up_proj", "model.layers.29.mlp.experts.48.up_proj", "model.layers.29.mlp.experts.49.up_proj", "model.layers.29.mlp.experts.50.up_proj", "model.layers.29.mlp.experts.51.up_proj", "model.layers.29.mlp.experts.52.up_proj", "model.layers.29.mlp.experts.53.up_proj", "model.layers.29.mlp.experts.54.up_proj", "model.layers.29.mlp.experts.55.up_proj", "model.layers.29.mlp.experts.56.up_proj", "model.layers.29.mlp.experts.57.up_proj", "model.layers.29.mlp.experts.58.up_proj", "model.layers.29.mlp.experts.59.up_proj", "model.layers.29.mlp.experts.60.up_proj", "model.layers.29.mlp.experts.61.up_proj", "model.layers.29.mlp.experts.62.up_proj", "model.layers.29.mlp.experts.63.up_proj", "model.layers.29.mlp.experts.64.up_proj", "model.layers.29.mlp.experts.65.up_proj", "model.layers.29.mlp.experts.66.up_proj", "model.layers.29.mlp.experts.67.up_proj", "model.layers.29.mlp.experts.68.up_proj", "model.layers.29.mlp.experts.69.up_proj", "model.layers.29.mlp.experts.70.up_proj", "model.layers.29.mlp.experts.71.up_proj", "model.layers.29.mlp.experts.72.up_proj", "model.layers.29.mlp.experts.73.up_proj", "model.layers.29.mlp.experts.74.up_proj", "model.layers.29.mlp.experts.75.up_proj", "model.layers.29.mlp.experts.76.up_proj", "model.layers.29.mlp.experts.77.up_proj", "model.layers.29.mlp.experts.78.up_proj", "model.layers.29.mlp.experts.79.up_proj", "model.layers.29.mlp.experts.80.up_proj", "model.layers.29.mlp.experts.81.up_proj", "model.layers.29.mlp.experts.82.up_proj", "model.layers.29.mlp.experts.83.up_proj", "model.layers.29.mlp.experts.84.up_proj", "model.layers.29.mlp.experts.85.up_proj", "model.layers.29.mlp.experts.86.up_proj", "model.layers.29.mlp.experts.87.up_proj", "model.layers.29.mlp.experts.88.up_proj", "model.layers.29.mlp.experts.89.up_proj", "model.layers.29.mlp.experts.90.up_proj", "model.layers.29.mlp.experts.91.up_proj", "model.layers.29.mlp.experts.92.up_proj", "model.layers.29.mlp.experts.93.up_proj", "model.layers.29.mlp.experts.94.up_proj", "model.layers.29.mlp.experts.95.up_proj", "model.layers.29.mlp.experts.96.up_proj", "model.layers.29.mlp.experts.97.up_proj", "model.layers.29.mlp.experts.98.up_proj", "model.layers.29.mlp.experts.99.up_proj", "model.layers.29.mlp.experts.100.up_proj", "model.layers.29.mlp.experts.101.up_proj", "model.layers.29.mlp.experts.102.up_proj", "model.layers.29.mlp.experts.103.up_proj", "model.layers.29.mlp.experts.104.up_proj", "model.layers.29.mlp.experts.105.up_proj", "model.layers.29.mlp.experts.106.up_proj", "model.layers.29.mlp.experts.107.up_proj", "model.layers.29.mlp.experts.108.up_proj", "model.layers.29.mlp.experts.109.up_proj", "model.layers.29.mlp.experts.110.up_proj", "model.layers.29.mlp.experts.111.up_proj", "model.layers.29.mlp.experts.112.up_proj", "model.layers.29.mlp.experts.113.up_proj", "model.layers.29.mlp.experts.114.up_proj", "model.layers.29.mlp.experts.115.up_proj", "model.layers.29.mlp.experts.116.up_proj", "model.layers.29.mlp.experts.117.up_proj", "model.layers.29.mlp.experts.118.up_proj", "model.layers.29.mlp.experts.119.up_proj", "model.layers.29.mlp.experts.120.up_proj", "model.layers.29.mlp.experts.121.up_proj", "model.layers.29.mlp.experts.122.up_proj", "model.layers.29.mlp.experts.123.up_proj", "model.layers.29.mlp.experts.124.up_proj", "model.layers.29.mlp.experts.125.up_proj", "model.layers.29.mlp.experts.126.up_proj", "model.layers.29.mlp.experts.127.up_proj", "model.layers.29.mlp.experts.128.up_proj", "model.layers.29.mlp.experts.129.up_proj", "model.layers.29.mlp.experts.130.up_proj", "model.layers.29.mlp.experts.131.up_proj", "model.layers.29.mlp.experts.132.up_proj", "model.layers.29.mlp.experts.133.up_proj", "model.layers.29.mlp.experts.134.up_proj", "model.layers.29.mlp.experts.135.up_proj", "model.layers.29.mlp.experts.136.up_proj", "model.layers.29.mlp.experts.137.up_proj", "model.layers.29.mlp.experts.138.up_proj", "model.layers.29.mlp.experts.139.up_proj", "model.layers.29.mlp.experts.140.up_proj", "model.layers.29.mlp.experts.141.up_proj", "model.layers.29.mlp.experts.142.up_proj", "model.layers.29.mlp.experts.143.up_proj", "model.layers.29.mlp.experts.144.up_proj", "model.layers.29.mlp.experts.145.up_proj", "model.layers.29.mlp.experts.146.up_proj", "model.layers.29.mlp.experts.147.up_proj", "model.layers.29.mlp.experts.148.up_proj", "model.layers.29.mlp.experts.149.up_proj", "model.layers.29.mlp.experts.150.up_proj", "model.layers.29.mlp.experts.151.up_proj", "model.layers.29.mlp.experts.152.up_proj", "model.layers.29.mlp.experts.153.up_proj", "model.layers.29.mlp.experts.154.up_proj", "model.layers.29.mlp.experts.155.up_proj", "model.layers.29.mlp.experts.156.up_proj", "model.layers.29.mlp.experts.157.up_proj", "model.layers.29.mlp.experts.158.up_proj", "model.layers.29.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000694247242063295, "dbits": 2516582400 }, { "dkld": -0.000568846613168722, "dbits": 5033164800 }, { "dkld": -0.0007067480124533232, "dbits": 7549747200 }, { "dkld": -0.0007608024403452929, "dbits": 12582912000 } ] }, { "idx": 176, "layers": [ "model.layers.29.mlp.experts.0.down_proj", "model.layers.29.mlp.experts.1.down_proj", "model.layers.29.mlp.experts.2.down_proj", "model.layers.29.mlp.experts.3.down_proj", "model.layers.29.mlp.experts.4.down_proj", "model.layers.29.mlp.experts.5.down_proj", "model.layers.29.mlp.experts.6.down_proj", "model.layers.29.mlp.experts.7.down_proj", "model.layers.29.mlp.experts.8.down_proj", "model.layers.29.mlp.experts.9.down_proj", "model.layers.29.mlp.experts.10.down_proj", "model.layers.29.mlp.experts.11.down_proj", "model.layers.29.mlp.experts.12.down_proj", "model.layers.29.mlp.experts.13.down_proj", "model.layers.29.mlp.experts.14.down_proj", "model.layers.29.mlp.experts.15.down_proj", "model.layers.29.mlp.experts.16.down_proj", "model.layers.29.mlp.experts.17.down_proj", "model.layers.29.mlp.experts.18.down_proj", "model.layers.29.mlp.experts.19.down_proj", "model.layers.29.mlp.experts.20.down_proj", "model.layers.29.mlp.experts.21.down_proj", "model.layers.29.mlp.experts.22.down_proj", "model.layers.29.mlp.experts.23.down_proj", "model.layers.29.mlp.experts.24.down_proj", "model.layers.29.mlp.experts.25.down_proj", "model.layers.29.mlp.experts.26.down_proj", "model.layers.29.mlp.experts.27.down_proj", "model.layers.29.mlp.experts.28.down_proj", "model.layers.29.mlp.experts.29.down_proj", "model.layers.29.mlp.experts.30.down_proj", "model.layers.29.mlp.experts.31.down_proj", "model.layers.29.mlp.experts.32.down_proj", "model.layers.29.mlp.experts.33.down_proj", "model.layers.29.mlp.experts.34.down_proj", "model.layers.29.mlp.experts.35.down_proj", "model.layers.29.mlp.experts.36.down_proj", "model.layers.29.mlp.experts.37.down_proj", "model.layers.29.mlp.experts.38.down_proj", "model.layers.29.mlp.experts.39.down_proj", "model.layers.29.mlp.experts.40.down_proj", "model.layers.29.mlp.experts.41.down_proj", "model.layers.29.mlp.experts.42.down_proj", "model.layers.29.mlp.experts.43.down_proj", "model.layers.29.mlp.experts.44.down_proj", "model.layers.29.mlp.experts.45.down_proj", "model.layers.29.mlp.experts.46.down_proj", "model.layers.29.mlp.experts.47.down_proj", "model.layers.29.mlp.experts.48.down_proj", "model.layers.29.mlp.experts.49.down_proj", "model.layers.29.mlp.experts.50.down_proj", "model.layers.29.mlp.experts.51.down_proj", "model.layers.29.mlp.experts.52.down_proj", "model.layers.29.mlp.experts.53.down_proj", "model.layers.29.mlp.experts.54.down_proj", "model.layers.29.mlp.experts.55.down_proj", "model.layers.29.mlp.experts.56.down_proj", "model.layers.29.mlp.experts.57.down_proj", "model.layers.29.mlp.experts.58.down_proj", "model.layers.29.mlp.experts.59.down_proj", "model.layers.29.mlp.experts.60.down_proj", "model.layers.29.mlp.experts.61.down_proj", "model.layers.29.mlp.experts.62.down_proj", "model.layers.29.mlp.experts.63.down_proj", "model.layers.29.mlp.experts.64.down_proj", "model.layers.29.mlp.experts.65.down_proj", "model.layers.29.mlp.experts.66.down_proj", "model.layers.29.mlp.experts.67.down_proj", "model.layers.29.mlp.experts.68.down_proj", "model.layers.29.mlp.experts.69.down_proj", "model.layers.29.mlp.experts.70.down_proj", "model.layers.29.mlp.experts.71.down_proj", "model.layers.29.mlp.experts.72.down_proj", "model.layers.29.mlp.experts.73.down_proj", "model.layers.29.mlp.experts.74.down_proj", "model.layers.29.mlp.experts.75.down_proj", "model.layers.29.mlp.experts.76.down_proj", "model.layers.29.mlp.experts.77.down_proj", "model.layers.29.mlp.experts.78.down_proj", "model.layers.29.mlp.experts.79.down_proj", "model.layers.29.mlp.experts.80.down_proj", "model.layers.29.mlp.experts.81.down_proj", "model.layers.29.mlp.experts.82.down_proj", "model.layers.29.mlp.experts.83.down_proj", "model.layers.29.mlp.experts.84.down_proj", "model.layers.29.mlp.experts.85.down_proj", "model.layers.29.mlp.experts.86.down_proj", "model.layers.29.mlp.experts.87.down_proj", "model.layers.29.mlp.experts.88.down_proj", "model.layers.29.mlp.experts.89.down_proj", "model.layers.29.mlp.experts.90.down_proj", "model.layers.29.mlp.experts.91.down_proj", "model.layers.29.mlp.experts.92.down_proj", "model.layers.29.mlp.experts.93.down_proj", "model.layers.29.mlp.experts.94.down_proj", "model.layers.29.mlp.experts.95.down_proj", "model.layers.29.mlp.experts.96.down_proj", "model.layers.29.mlp.experts.97.down_proj", "model.layers.29.mlp.experts.98.down_proj", "model.layers.29.mlp.experts.99.down_proj", "model.layers.29.mlp.experts.100.down_proj", "model.layers.29.mlp.experts.101.down_proj", "model.layers.29.mlp.experts.102.down_proj", "model.layers.29.mlp.experts.103.down_proj", "model.layers.29.mlp.experts.104.down_proj", "model.layers.29.mlp.experts.105.down_proj", "model.layers.29.mlp.experts.106.down_proj", "model.layers.29.mlp.experts.107.down_proj", "model.layers.29.mlp.experts.108.down_proj", "model.layers.29.mlp.experts.109.down_proj", "model.layers.29.mlp.experts.110.down_proj", "model.layers.29.mlp.experts.111.down_proj", "model.layers.29.mlp.experts.112.down_proj", "model.layers.29.mlp.experts.113.down_proj", "model.layers.29.mlp.experts.114.down_proj", "model.layers.29.mlp.experts.115.down_proj", "model.layers.29.mlp.experts.116.down_proj", "model.layers.29.mlp.experts.117.down_proj", "model.layers.29.mlp.experts.118.down_proj", "model.layers.29.mlp.experts.119.down_proj", "model.layers.29.mlp.experts.120.down_proj", "model.layers.29.mlp.experts.121.down_proj", "model.layers.29.mlp.experts.122.down_proj", "model.layers.29.mlp.experts.123.down_proj", "model.layers.29.mlp.experts.124.down_proj", "model.layers.29.mlp.experts.125.down_proj", "model.layers.29.mlp.experts.126.down_proj", "model.layers.29.mlp.experts.127.down_proj", "model.layers.29.mlp.experts.128.down_proj", "model.layers.29.mlp.experts.129.down_proj", "model.layers.29.mlp.experts.130.down_proj", "model.layers.29.mlp.experts.131.down_proj", "model.layers.29.mlp.experts.132.down_proj", "model.layers.29.mlp.experts.133.down_proj", "model.layers.29.mlp.experts.134.down_proj", "model.layers.29.mlp.experts.135.down_proj", "model.layers.29.mlp.experts.136.down_proj", "model.layers.29.mlp.experts.137.down_proj", "model.layers.29.mlp.experts.138.down_proj", "model.layers.29.mlp.experts.139.down_proj", "model.layers.29.mlp.experts.140.down_proj", "model.layers.29.mlp.experts.141.down_proj", "model.layers.29.mlp.experts.142.down_proj", "model.layers.29.mlp.experts.143.down_proj", "model.layers.29.mlp.experts.144.down_proj", "model.layers.29.mlp.experts.145.down_proj", "model.layers.29.mlp.experts.146.down_proj", "model.layers.29.mlp.experts.147.down_proj", "model.layers.29.mlp.experts.148.down_proj", "model.layers.29.mlp.experts.149.down_proj", "model.layers.29.mlp.experts.150.down_proj", "model.layers.29.mlp.experts.151.down_proj", "model.layers.29.mlp.experts.152.down_proj", "model.layers.29.mlp.experts.153.down_proj", "model.layers.29.mlp.experts.154.down_proj", "model.layers.29.mlp.experts.155.down_proj", "model.layers.29.mlp.experts.156.down_proj", "model.layers.29.mlp.experts.157.down_proj", "model.layers.29.mlp.experts.158.down_proj", "model.layers.29.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003728167153894957, "dbits": 1258291200 }, { "dkld": -0.00047338102012872696, "dbits": 2516582400 }, { "dkld": -0.0004275717772543458, "dbits": 3774873600 }, { "dkld": -0.000516415573656559, "dbits": 6291456000 } ] }, { "idx": 177, "layers": [ "model.layers.30.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0005433311685919817, "dbits": 62914560 }, { "dkld": -0.00041630659252406277, "dbits": 125829120 }, { "dkld": -0.0005358221009373748, "dbits": 188743680 }, { "dkld": -0.0005887582898140009, "dbits": 314572800 } ] }, { "idx": 178, "layers": [ "model.layers.30.self_attn.k_proj", "model.layers.30.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0003056082874536431, "dbits": 10485760 }, { "dkld": 3.77310439944184e-05, "dbits": 20971520 }, { "dkld": -0.00011196229606867669, "dbits": 31457280 }, { "dkld": 2.7572736144010346e-06, "dbits": 52428800 } ] }, { "idx": 179, "layers": [ "model.layers.30.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00011827489361165844, "dbits": 62914560 }, { "dkld": 9.907856583594998e-05, "dbits": 125829120 }, { "dkld": 0.0001582257449626867, "dbits": 188743680 }, { "dkld": 0.00015087611973285675, "dbits": 314572800 } ] }, { "idx": 180, "layers": [ "model.layers.30.mlp.shared_experts.gate_proj", "model.layers.30.mlp.shared_experts.up_proj", "model.layers.30.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0011660382151603699, "dbits": 23592960 }, { "dkld": -0.0013656262308359146, "dbits": 47185920 }, { "dkld": -0.0010997468605637578, "dbits": 70778880 }, { "dkld": -0.0012043366208672607, "dbits": 117964800 } ] }, { "idx": 181, "layers": [ "model.layers.30.mlp.experts.0.gate_proj", "model.layers.30.mlp.experts.1.gate_proj", "model.layers.30.mlp.experts.2.gate_proj", "model.layers.30.mlp.experts.3.gate_proj", "model.layers.30.mlp.experts.4.gate_proj", "model.layers.30.mlp.experts.5.gate_proj", "model.layers.30.mlp.experts.6.gate_proj", "model.layers.30.mlp.experts.7.gate_proj", "model.layers.30.mlp.experts.8.gate_proj", "model.layers.30.mlp.experts.9.gate_proj", "model.layers.30.mlp.experts.10.gate_proj", "model.layers.30.mlp.experts.11.gate_proj", "model.layers.30.mlp.experts.12.gate_proj", "model.layers.30.mlp.experts.13.gate_proj", "model.layers.30.mlp.experts.14.gate_proj", "model.layers.30.mlp.experts.15.gate_proj", "model.layers.30.mlp.experts.16.gate_proj", "model.layers.30.mlp.experts.17.gate_proj", "model.layers.30.mlp.experts.18.gate_proj", "model.layers.30.mlp.experts.19.gate_proj", "model.layers.30.mlp.experts.20.gate_proj", "model.layers.30.mlp.experts.21.gate_proj", "model.layers.30.mlp.experts.22.gate_proj", "model.layers.30.mlp.experts.23.gate_proj", "model.layers.30.mlp.experts.24.gate_proj", "model.layers.30.mlp.experts.25.gate_proj", "model.layers.30.mlp.experts.26.gate_proj", "model.layers.30.mlp.experts.27.gate_proj", "model.layers.30.mlp.experts.28.gate_proj", "model.layers.30.mlp.experts.29.gate_proj", "model.layers.30.mlp.experts.30.gate_proj", "model.layers.30.mlp.experts.31.gate_proj", "model.layers.30.mlp.experts.32.gate_proj", "model.layers.30.mlp.experts.33.gate_proj", "model.layers.30.mlp.experts.34.gate_proj", "model.layers.30.mlp.experts.35.gate_proj", "model.layers.30.mlp.experts.36.gate_proj", "model.layers.30.mlp.experts.37.gate_proj", "model.layers.30.mlp.experts.38.gate_proj", "model.layers.30.mlp.experts.39.gate_proj", "model.layers.30.mlp.experts.40.gate_proj", "model.layers.30.mlp.experts.41.gate_proj", "model.layers.30.mlp.experts.42.gate_proj", "model.layers.30.mlp.experts.43.gate_proj", "model.layers.30.mlp.experts.44.gate_proj", "model.layers.30.mlp.experts.45.gate_proj", "model.layers.30.mlp.experts.46.gate_proj", "model.layers.30.mlp.experts.47.gate_proj", "model.layers.30.mlp.experts.48.gate_proj", "model.layers.30.mlp.experts.49.gate_proj", "model.layers.30.mlp.experts.50.gate_proj", "model.layers.30.mlp.experts.51.gate_proj", "model.layers.30.mlp.experts.52.gate_proj", "model.layers.30.mlp.experts.53.gate_proj", "model.layers.30.mlp.experts.54.gate_proj", "model.layers.30.mlp.experts.55.gate_proj", "model.layers.30.mlp.experts.56.gate_proj", "model.layers.30.mlp.experts.57.gate_proj", "model.layers.30.mlp.experts.58.gate_proj", "model.layers.30.mlp.experts.59.gate_proj", "model.layers.30.mlp.experts.60.gate_proj", "model.layers.30.mlp.experts.61.gate_proj", "model.layers.30.mlp.experts.62.gate_proj", "model.layers.30.mlp.experts.63.gate_proj", "model.layers.30.mlp.experts.64.gate_proj", "model.layers.30.mlp.experts.65.gate_proj", "model.layers.30.mlp.experts.66.gate_proj", "model.layers.30.mlp.experts.67.gate_proj", "model.layers.30.mlp.experts.68.gate_proj", "model.layers.30.mlp.experts.69.gate_proj", "model.layers.30.mlp.experts.70.gate_proj", "model.layers.30.mlp.experts.71.gate_proj", "model.layers.30.mlp.experts.72.gate_proj", "model.layers.30.mlp.experts.73.gate_proj", "model.layers.30.mlp.experts.74.gate_proj", "model.layers.30.mlp.experts.75.gate_proj", "model.layers.30.mlp.experts.76.gate_proj", "model.layers.30.mlp.experts.77.gate_proj", "model.layers.30.mlp.experts.78.gate_proj", "model.layers.30.mlp.experts.79.gate_proj", "model.layers.30.mlp.experts.80.gate_proj", "model.layers.30.mlp.experts.81.gate_proj", "model.layers.30.mlp.experts.82.gate_proj", "model.layers.30.mlp.experts.83.gate_proj", "model.layers.30.mlp.experts.84.gate_proj", "model.layers.30.mlp.experts.85.gate_proj", "model.layers.30.mlp.experts.86.gate_proj", "model.layers.30.mlp.experts.87.gate_proj", "model.layers.30.mlp.experts.88.gate_proj", "model.layers.30.mlp.experts.89.gate_proj", "model.layers.30.mlp.experts.90.gate_proj", "model.layers.30.mlp.experts.91.gate_proj", "model.layers.30.mlp.experts.92.gate_proj", "model.layers.30.mlp.experts.93.gate_proj", "model.layers.30.mlp.experts.94.gate_proj", "model.layers.30.mlp.experts.95.gate_proj", "model.layers.30.mlp.experts.96.gate_proj", "model.layers.30.mlp.experts.97.gate_proj", "model.layers.30.mlp.experts.98.gate_proj", "model.layers.30.mlp.experts.99.gate_proj", "model.layers.30.mlp.experts.100.gate_proj", "model.layers.30.mlp.experts.101.gate_proj", "model.layers.30.mlp.experts.102.gate_proj", "model.layers.30.mlp.experts.103.gate_proj", "model.layers.30.mlp.experts.104.gate_proj", "model.layers.30.mlp.experts.105.gate_proj", "model.layers.30.mlp.experts.106.gate_proj", "model.layers.30.mlp.experts.107.gate_proj", "model.layers.30.mlp.experts.108.gate_proj", "model.layers.30.mlp.experts.109.gate_proj", "model.layers.30.mlp.experts.110.gate_proj", "model.layers.30.mlp.experts.111.gate_proj", "model.layers.30.mlp.experts.112.gate_proj", "model.layers.30.mlp.experts.113.gate_proj", "model.layers.30.mlp.experts.114.gate_proj", "model.layers.30.mlp.experts.115.gate_proj", "model.layers.30.mlp.experts.116.gate_proj", "model.layers.30.mlp.experts.117.gate_proj", "model.layers.30.mlp.experts.118.gate_proj", "model.layers.30.mlp.experts.119.gate_proj", "model.layers.30.mlp.experts.120.gate_proj", "model.layers.30.mlp.experts.121.gate_proj", "model.layers.30.mlp.experts.122.gate_proj", "model.layers.30.mlp.experts.123.gate_proj", "model.layers.30.mlp.experts.124.gate_proj", "model.layers.30.mlp.experts.125.gate_proj", "model.layers.30.mlp.experts.126.gate_proj", "model.layers.30.mlp.experts.127.gate_proj", "model.layers.30.mlp.experts.128.gate_proj", "model.layers.30.mlp.experts.129.gate_proj", "model.layers.30.mlp.experts.130.gate_proj", "model.layers.30.mlp.experts.131.gate_proj", "model.layers.30.mlp.experts.132.gate_proj", "model.layers.30.mlp.experts.133.gate_proj", "model.layers.30.mlp.experts.134.gate_proj", "model.layers.30.mlp.experts.135.gate_proj", "model.layers.30.mlp.experts.136.gate_proj", "model.layers.30.mlp.experts.137.gate_proj", "model.layers.30.mlp.experts.138.gate_proj", "model.layers.30.mlp.experts.139.gate_proj", "model.layers.30.mlp.experts.140.gate_proj", "model.layers.30.mlp.experts.141.gate_proj", "model.layers.30.mlp.experts.142.gate_proj", "model.layers.30.mlp.experts.143.gate_proj", "model.layers.30.mlp.experts.144.gate_proj", "model.layers.30.mlp.experts.145.gate_proj", "model.layers.30.mlp.experts.146.gate_proj", "model.layers.30.mlp.experts.147.gate_proj", "model.layers.30.mlp.experts.148.gate_proj", "model.layers.30.mlp.experts.149.gate_proj", "model.layers.30.mlp.experts.150.gate_proj", "model.layers.30.mlp.experts.151.gate_proj", "model.layers.30.mlp.experts.152.gate_proj", "model.layers.30.mlp.experts.153.gate_proj", "model.layers.30.mlp.experts.154.gate_proj", "model.layers.30.mlp.experts.155.gate_proj", "model.layers.30.mlp.experts.156.gate_proj", "model.layers.30.mlp.experts.157.gate_proj", "model.layers.30.mlp.experts.158.gate_proj", "model.layers.30.mlp.experts.159.gate_proj", "model.layers.30.mlp.experts.0.up_proj", "model.layers.30.mlp.experts.1.up_proj", "model.layers.30.mlp.experts.2.up_proj", "model.layers.30.mlp.experts.3.up_proj", "model.layers.30.mlp.experts.4.up_proj", "model.layers.30.mlp.experts.5.up_proj", "model.layers.30.mlp.experts.6.up_proj", "model.layers.30.mlp.experts.7.up_proj", "model.layers.30.mlp.experts.8.up_proj", "model.layers.30.mlp.experts.9.up_proj", "model.layers.30.mlp.experts.10.up_proj", "model.layers.30.mlp.experts.11.up_proj", "model.layers.30.mlp.experts.12.up_proj", "model.layers.30.mlp.experts.13.up_proj", "model.layers.30.mlp.experts.14.up_proj", "model.layers.30.mlp.experts.15.up_proj", "model.layers.30.mlp.experts.16.up_proj", "model.layers.30.mlp.experts.17.up_proj", "model.layers.30.mlp.experts.18.up_proj", "model.layers.30.mlp.experts.19.up_proj", "model.layers.30.mlp.experts.20.up_proj", "model.layers.30.mlp.experts.21.up_proj", "model.layers.30.mlp.experts.22.up_proj", "model.layers.30.mlp.experts.23.up_proj", "model.layers.30.mlp.experts.24.up_proj", "model.layers.30.mlp.experts.25.up_proj", "model.layers.30.mlp.experts.26.up_proj", "model.layers.30.mlp.experts.27.up_proj", "model.layers.30.mlp.experts.28.up_proj", "model.layers.30.mlp.experts.29.up_proj", "model.layers.30.mlp.experts.30.up_proj", "model.layers.30.mlp.experts.31.up_proj", "model.layers.30.mlp.experts.32.up_proj", "model.layers.30.mlp.experts.33.up_proj", "model.layers.30.mlp.experts.34.up_proj", "model.layers.30.mlp.experts.35.up_proj", "model.layers.30.mlp.experts.36.up_proj", "model.layers.30.mlp.experts.37.up_proj", "model.layers.30.mlp.experts.38.up_proj", "model.layers.30.mlp.experts.39.up_proj", "model.layers.30.mlp.experts.40.up_proj", "model.layers.30.mlp.experts.41.up_proj", "model.layers.30.mlp.experts.42.up_proj", "model.layers.30.mlp.experts.43.up_proj", "model.layers.30.mlp.experts.44.up_proj", "model.layers.30.mlp.experts.45.up_proj", "model.layers.30.mlp.experts.46.up_proj", "model.layers.30.mlp.experts.47.up_proj", "model.layers.30.mlp.experts.48.up_proj", "model.layers.30.mlp.experts.49.up_proj", "model.layers.30.mlp.experts.50.up_proj", "model.layers.30.mlp.experts.51.up_proj", "model.layers.30.mlp.experts.52.up_proj", "model.layers.30.mlp.experts.53.up_proj", "model.layers.30.mlp.experts.54.up_proj", "model.layers.30.mlp.experts.55.up_proj", "model.layers.30.mlp.experts.56.up_proj", "model.layers.30.mlp.experts.57.up_proj", "model.layers.30.mlp.experts.58.up_proj", "model.layers.30.mlp.experts.59.up_proj", "model.layers.30.mlp.experts.60.up_proj", "model.layers.30.mlp.experts.61.up_proj", "model.layers.30.mlp.experts.62.up_proj", "model.layers.30.mlp.experts.63.up_proj", "model.layers.30.mlp.experts.64.up_proj", "model.layers.30.mlp.experts.65.up_proj", "model.layers.30.mlp.experts.66.up_proj", "model.layers.30.mlp.experts.67.up_proj", "model.layers.30.mlp.experts.68.up_proj", "model.layers.30.mlp.experts.69.up_proj", "model.layers.30.mlp.experts.70.up_proj", "model.layers.30.mlp.experts.71.up_proj", "model.layers.30.mlp.experts.72.up_proj", "model.layers.30.mlp.experts.73.up_proj", "model.layers.30.mlp.experts.74.up_proj", "model.layers.30.mlp.experts.75.up_proj", "model.layers.30.mlp.experts.76.up_proj", "model.layers.30.mlp.experts.77.up_proj", "model.layers.30.mlp.experts.78.up_proj", "model.layers.30.mlp.experts.79.up_proj", "model.layers.30.mlp.experts.80.up_proj", "model.layers.30.mlp.experts.81.up_proj", "model.layers.30.mlp.experts.82.up_proj", "model.layers.30.mlp.experts.83.up_proj", "model.layers.30.mlp.experts.84.up_proj", "model.layers.30.mlp.experts.85.up_proj", "model.layers.30.mlp.experts.86.up_proj", "model.layers.30.mlp.experts.87.up_proj", "model.layers.30.mlp.experts.88.up_proj", "model.layers.30.mlp.experts.89.up_proj", "model.layers.30.mlp.experts.90.up_proj", "model.layers.30.mlp.experts.91.up_proj", "model.layers.30.mlp.experts.92.up_proj", "model.layers.30.mlp.experts.93.up_proj", "model.layers.30.mlp.experts.94.up_proj", "model.layers.30.mlp.experts.95.up_proj", "model.layers.30.mlp.experts.96.up_proj", "model.layers.30.mlp.experts.97.up_proj", "model.layers.30.mlp.experts.98.up_proj", "model.layers.30.mlp.experts.99.up_proj", "model.layers.30.mlp.experts.100.up_proj", "model.layers.30.mlp.experts.101.up_proj", "model.layers.30.mlp.experts.102.up_proj", "model.layers.30.mlp.experts.103.up_proj", "model.layers.30.mlp.experts.104.up_proj", "model.layers.30.mlp.experts.105.up_proj", "model.layers.30.mlp.experts.106.up_proj", "model.layers.30.mlp.experts.107.up_proj", "model.layers.30.mlp.experts.108.up_proj", "model.layers.30.mlp.experts.109.up_proj", "model.layers.30.mlp.experts.110.up_proj", "model.layers.30.mlp.experts.111.up_proj", "model.layers.30.mlp.experts.112.up_proj", "model.layers.30.mlp.experts.113.up_proj", "model.layers.30.mlp.experts.114.up_proj", "model.layers.30.mlp.experts.115.up_proj", "model.layers.30.mlp.experts.116.up_proj", "model.layers.30.mlp.experts.117.up_proj", "model.layers.30.mlp.experts.118.up_proj", "model.layers.30.mlp.experts.119.up_proj", "model.layers.30.mlp.experts.120.up_proj", "model.layers.30.mlp.experts.121.up_proj", "model.layers.30.mlp.experts.122.up_proj", "model.layers.30.mlp.experts.123.up_proj", "model.layers.30.mlp.experts.124.up_proj", "model.layers.30.mlp.experts.125.up_proj", "model.layers.30.mlp.experts.126.up_proj", "model.layers.30.mlp.experts.127.up_proj", "model.layers.30.mlp.experts.128.up_proj", "model.layers.30.mlp.experts.129.up_proj", "model.layers.30.mlp.experts.130.up_proj", "model.layers.30.mlp.experts.131.up_proj", "model.layers.30.mlp.experts.132.up_proj", "model.layers.30.mlp.experts.133.up_proj", "model.layers.30.mlp.experts.134.up_proj", "model.layers.30.mlp.experts.135.up_proj", "model.layers.30.mlp.experts.136.up_proj", "model.layers.30.mlp.experts.137.up_proj", "model.layers.30.mlp.experts.138.up_proj", "model.layers.30.mlp.experts.139.up_proj", "model.layers.30.mlp.experts.140.up_proj", "model.layers.30.mlp.experts.141.up_proj", "model.layers.30.mlp.experts.142.up_proj", "model.layers.30.mlp.experts.143.up_proj", "model.layers.30.mlp.experts.144.up_proj", "model.layers.30.mlp.experts.145.up_proj", "model.layers.30.mlp.experts.146.up_proj", "model.layers.30.mlp.experts.147.up_proj", "model.layers.30.mlp.experts.148.up_proj", "model.layers.30.mlp.experts.149.up_proj", "model.layers.30.mlp.experts.150.up_proj", "model.layers.30.mlp.experts.151.up_proj", "model.layers.30.mlp.experts.152.up_proj", "model.layers.30.mlp.experts.153.up_proj", "model.layers.30.mlp.experts.154.up_proj", "model.layers.30.mlp.experts.155.up_proj", "model.layers.30.mlp.experts.156.up_proj", "model.layers.30.mlp.experts.157.up_proj", "model.layers.30.mlp.experts.158.up_proj", "model.layers.30.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000642928108572971, "dbits": 2516582400 }, { "dkld": -0.0007235471159219853, "dbits": 5033164800 }, { "dkld": -0.0007674405351281166, "dbits": 7549747200 }, { "dkld": -0.000790007598698142, "dbits": 12582912000 } ] }, { "idx": 182, "layers": [ "model.layers.30.mlp.experts.0.down_proj", "model.layers.30.mlp.experts.1.down_proj", "model.layers.30.mlp.experts.2.down_proj", "model.layers.30.mlp.experts.3.down_proj", "model.layers.30.mlp.experts.4.down_proj", "model.layers.30.mlp.experts.5.down_proj", "model.layers.30.mlp.experts.6.down_proj", "model.layers.30.mlp.experts.7.down_proj", "model.layers.30.mlp.experts.8.down_proj", "model.layers.30.mlp.experts.9.down_proj", "model.layers.30.mlp.experts.10.down_proj", "model.layers.30.mlp.experts.11.down_proj", "model.layers.30.mlp.experts.12.down_proj", "model.layers.30.mlp.experts.13.down_proj", "model.layers.30.mlp.experts.14.down_proj", "model.layers.30.mlp.experts.15.down_proj", "model.layers.30.mlp.experts.16.down_proj", "model.layers.30.mlp.experts.17.down_proj", "model.layers.30.mlp.experts.18.down_proj", "model.layers.30.mlp.experts.19.down_proj", "model.layers.30.mlp.experts.20.down_proj", "model.layers.30.mlp.experts.21.down_proj", "model.layers.30.mlp.experts.22.down_proj", "model.layers.30.mlp.experts.23.down_proj", "model.layers.30.mlp.experts.24.down_proj", "model.layers.30.mlp.experts.25.down_proj", "model.layers.30.mlp.experts.26.down_proj", "model.layers.30.mlp.experts.27.down_proj", "model.layers.30.mlp.experts.28.down_proj", "model.layers.30.mlp.experts.29.down_proj", "model.layers.30.mlp.experts.30.down_proj", "model.layers.30.mlp.experts.31.down_proj", "model.layers.30.mlp.experts.32.down_proj", "model.layers.30.mlp.experts.33.down_proj", "model.layers.30.mlp.experts.34.down_proj", "model.layers.30.mlp.experts.35.down_proj", "model.layers.30.mlp.experts.36.down_proj", "model.layers.30.mlp.experts.37.down_proj", "model.layers.30.mlp.experts.38.down_proj", "model.layers.30.mlp.experts.39.down_proj", "model.layers.30.mlp.experts.40.down_proj", "model.layers.30.mlp.experts.41.down_proj", "model.layers.30.mlp.experts.42.down_proj", "model.layers.30.mlp.experts.43.down_proj", "model.layers.30.mlp.experts.44.down_proj", "model.layers.30.mlp.experts.45.down_proj", "model.layers.30.mlp.experts.46.down_proj", "model.layers.30.mlp.experts.47.down_proj", "model.layers.30.mlp.experts.48.down_proj", "model.layers.30.mlp.experts.49.down_proj", "model.layers.30.mlp.experts.50.down_proj", "model.layers.30.mlp.experts.51.down_proj", "model.layers.30.mlp.experts.52.down_proj", "model.layers.30.mlp.experts.53.down_proj", "model.layers.30.mlp.experts.54.down_proj", "model.layers.30.mlp.experts.55.down_proj", "model.layers.30.mlp.experts.56.down_proj", "model.layers.30.mlp.experts.57.down_proj", "model.layers.30.mlp.experts.58.down_proj", "model.layers.30.mlp.experts.59.down_proj", "model.layers.30.mlp.experts.60.down_proj", "model.layers.30.mlp.experts.61.down_proj", "model.layers.30.mlp.experts.62.down_proj", "model.layers.30.mlp.experts.63.down_proj", "model.layers.30.mlp.experts.64.down_proj", "model.layers.30.mlp.experts.65.down_proj", "model.layers.30.mlp.experts.66.down_proj", "model.layers.30.mlp.experts.67.down_proj", "model.layers.30.mlp.experts.68.down_proj", "model.layers.30.mlp.experts.69.down_proj", "model.layers.30.mlp.experts.70.down_proj", "model.layers.30.mlp.experts.71.down_proj", "model.layers.30.mlp.experts.72.down_proj", "model.layers.30.mlp.experts.73.down_proj", "model.layers.30.mlp.experts.74.down_proj", "model.layers.30.mlp.experts.75.down_proj", "model.layers.30.mlp.experts.76.down_proj", "model.layers.30.mlp.experts.77.down_proj", "model.layers.30.mlp.experts.78.down_proj", "model.layers.30.mlp.experts.79.down_proj", "model.layers.30.mlp.experts.80.down_proj", "model.layers.30.mlp.experts.81.down_proj", "model.layers.30.mlp.experts.82.down_proj", "model.layers.30.mlp.experts.83.down_proj", "model.layers.30.mlp.experts.84.down_proj", "model.layers.30.mlp.experts.85.down_proj", "model.layers.30.mlp.experts.86.down_proj", "model.layers.30.mlp.experts.87.down_proj", "model.layers.30.mlp.experts.88.down_proj", "model.layers.30.mlp.experts.89.down_proj", "model.layers.30.mlp.experts.90.down_proj", "model.layers.30.mlp.experts.91.down_proj", "model.layers.30.mlp.experts.92.down_proj", "model.layers.30.mlp.experts.93.down_proj", "model.layers.30.mlp.experts.94.down_proj", "model.layers.30.mlp.experts.95.down_proj", "model.layers.30.mlp.experts.96.down_proj", "model.layers.30.mlp.experts.97.down_proj", "model.layers.30.mlp.experts.98.down_proj", "model.layers.30.mlp.experts.99.down_proj", "model.layers.30.mlp.experts.100.down_proj", "model.layers.30.mlp.experts.101.down_proj", "model.layers.30.mlp.experts.102.down_proj", "model.layers.30.mlp.experts.103.down_proj", "model.layers.30.mlp.experts.104.down_proj", "model.layers.30.mlp.experts.105.down_proj", "model.layers.30.mlp.experts.106.down_proj", "model.layers.30.mlp.experts.107.down_proj", "model.layers.30.mlp.experts.108.down_proj", "model.layers.30.mlp.experts.109.down_proj", "model.layers.30.mlp.experts.110.down_proj", "model.layers.30.mlp.experts.111.down_proj", "model.layers.30.mlp.experts.112.down_proj", "model.layers.30.mlp.experts.113.down_proj", "model.layers.30.mlp.experts.114.down_proj", "model.layers.30.mlp.experts.115.down_proj", "model.layers.30.mlp.experts.116.down_proj", "model.layers.30.mlp.experts.117.down_proj", "model.layers.30.mlp.experts.118.down_proj", "model.layers.30.mlp.experts.119.down_proj", "model.layers.30.mlp.experts.120.down_proj", "model.layers.30.mlp.experts.121.down_proj", "model.layers.30.mlp.experts.122.down_proj", "model.layers.30.mlp.experts.123.down_proj", "model.layers.30.mlp.experts.124.down_proj", "model.layers.30.mlp.experts.125.down_proj", "model.layers.30.mlp.experts.126.down_proj", "model.layers.30.mlp.experts.127.down_proj", "model.layers.30.mlp.experts.128.down_proj", "model.layers.30.mlp.experts.129.down_proj", "model.layers.30.mlp.experts.130.down_proj", "model.layers.30.mlp.experts.131.down_proj", "model.layers.30.mlp.experts.132.down_proj", "model.layers.30.mlp.experts.133.down_proj", "model.layers.30.mlp.experts.134.down_proj", "model.layers.30.mlp.experts.135.down_proj", "model.layers.30.mlp.experts.136.down_proj", "model.layers.30.mlp.experts.137.down_proj", "model.layers.30.mlp.experts.138.down_proj", "model.layers.30.mlp.experts.139.down_proj", "model.layers.30.mlp.experts.140.down_proj", "model.layers.30.mlp.experts.141.down_proj", "model.layers.30.mlp.experts.142.down_proj", "model.layers.30.mlp.experts.143.down_proj", "model.layers.30.mlp.experts.144.down_proj", "model.layers.30.mlp.experts.145.down_proj", "model.layers.30.mlp.experts.146.down_proj", "model.layers.30.mlp.experts.147.down_proj", "model.layers.30.mlp.experts.148.down_proj", "model.layers.30.mlp.experts.149.down_proj", "model.layers.30.mlp.experts.150.down_proj", "model.layers.30.mlp.experts.151.down_proj", "model.layers.30.mlp.experts.152.down_proj", "model.layers.30.mlp.experts.153.down_proj", "model.layers.30.mlp.experts.154.down_proj", "model.layers.30.mlp.experts.155.down_proj", "model.layers.30.mlp.experts.156.down_proj", "model.layers.30.mlp.experts.157.down_proj", "model.layers.30.mlp.experts.158.down_proj", "model.layers.30.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002809481695294491, "dbits": 1258291200 }, { "dkld": -0.00029432643204928277, "dbits": 2516582400 }, { "dkld": -0.0003479917533695809, "dbits": 3774873600 }, { "dkld": -0.00033504562452435493, "dbits": 6291456000 } ] }, { "idx": 183, "layers": [ "model.layers.31.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0005715782754123211, "dbits": 62914560 }, { "dkld": 0.0008111003786325371, "dbits": 125829120 }, { "dkld": 0.0007810850627720245, "dbits": 188743680 }, { "dkld": 0.0008655022829771042, "dbits": 314572800 } ] }, { "idx": 184, "layers": [ "model.layers.31.self_attn.k_proj", "model.layers.31.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0016425022855400956, "dbits": 10485760 }, { "dkld": 0.0008795233443379402, "dbits": 20971520 }, { "dkld": 0.0013554511591792051, "dbits": 31457280 }, { "dkld": 0.0011470620520412922, "dbits": 52428800 } ] }, { "idx": 185, "layers": [ "model.layers.31.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00011645564809442677, "dbits": 62914560 }, { "dkld": -0.00022012311965227682, "dbits": 125829120 }, { "dkld": 0.00025942102074622, "dbits": 188743680 }, { "dkld": 0.0003153722733259118, "dbits": 314572800 } ] }, { "idx": 186, "layers": [ "model.layers.31.mlp.shared_experts.gate_proj", "model.layers.31.mlp.shared_experts.up_proj", "model.layers.31.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0019934879615902817, "dbits": 23592960 }, { "dkld": 0.0006275212392210933, "dbits": 47185920 }, { "dkld": 0.0004156701266765511, "dbits": 70778880 }, { "dkld": 0.0005083652213215828, "dbits": 117964800 } ] }, { "idx": 187, "layers": [ "model.layers.31.mlp.experts.0.gate_proj", "model.layers.31.mlp.experts.1.gate_proj", "model.layers.31.mlp.experts.2.gate_proj", "model.layers.31.mlp.experts.3.gate_proj", "model.layers.31.mlp.experts.4.gate_proj", "model.layers.31.mlp.experts.5.gate_proj", "model.layers.31.mlp.experts.6.gate_proj", "model.layers.31.mlp.experts.7.gate_proj", "model.layers.31.mlp.experts.8.gate_proj", "model.layers.31.mlp.experts.9.gate_proj", "model.layers.31.mlp.experts.10.gate_proj", "model.layers.31.mlp.experts.11.gate_proj", "model.layers.31.mlp.experts.12.gate_proj", "model.layers.31.mlp.experts.13.gate_proj", "model.layers.31.mlp.experts.14.gate_proj", "model.layers.31.mlp.experts.15.gate_proj", "model.layers.31.mlp.experts.16.gate_proj", "model.layers.31.mlp.experts.17.gate_proj", "model.layers.31.mlp.experts.18.gate_proj", "model.layers.31.mlp.experts.19.gate_proj", "model.layers.31.mlp.experts.20.gate_proj", "model.layers.31.mlp.experts.21.gate_proj", "model.layers.31.mlp.experts.22.gate_proj", "model.layers.31.mlp.experts.23.gate_proj", "model.layers.31.mlp.experts.24.gate_proj", "model.layers.31.mlp.experts.25.gate_proj", "model.layers.31.mlp.experts.26.gate_proj", "model.layers.31.mlp.experts.27.gate_proj", "model.layers.31.mlp.experts.28.gate_proj", "model.layers.31.mlp.experts.29.gate_proj", "model.layers.31.mlp.experts.30.gate_proj", "model.layers.31.mlp.experts.31.gate_proj", "model.layers.31.mlp.experts.32.gate_proj", "model.layers.31.mlp.experts.33.gate_proj", "model.layers.31.mlp.experts.34.gate_proj", "model.layers.31.mlp.experts.35.gate_proj", "model.layers.31.mlp.experts.36.gate_proj", "model.layers.31.mlp.experts.37.gate_proj", "model.layers.31.mlp.experts.38.gate_proj", "model.layers.31.mlp.experts.39.gate_proj", "model.layers.31.mlp.experts.40.gate_proj", "model.layers.31.mlp.experts.41.gate_proj", "model.layers.31.mlp.experts.42.gate_proj", "model.layers.31.mlp.experts.43.gate_proj", "model.layers.31.mlp.experts.44.gate_proj", "model.layers.31.mlp.experts.45.gate_proj", "model.layers.31.mlp.experts.46.gate_proj", "model.layers.31.mlp.experts.47.gate_proj", "model.layers.31.mlp.experts.48.gate_proj", "model.layers.31.mlp.experts.49.gate_proj", "model.layers.31.mlp.experts.50.gate_proj", "model.layers.31.mlp.experts.51.gate_proj", "model.layers.31.mlp.experts.52.gate_proj", "model.layers.31.mlp.experts.53.gate_proj", "model.layers.31.mlp.experts.54.gate_proj", "model.layers.31.mlp.experts.55.gate_proj", "model.layers.31.mlp.experts.56.gate_proj", "model.layers.31.mlp.experts.57.gate_proj", "model.layers.31.mlp.experts.58.gate_proj", "model.layers.31.mlp.experts.59.gate_proj", "model.layers.31.mlp.experts.60.gate_proj", "model.layers.31.mlp.experts.61.gate_proj", "model.layers.31.mlp.experts.62.gate_proj", "model.layers.31.mlp.experts.63.gate_proj", "model.layers.31.mlp.experts.64.gate_proj", "model.layers.31.mlp.experts.65.gate_proj", "model.layers.31.mlp.experts.66.gate_proj", "model.layers.31.mlp.experts.67.gate_proj", "model.layers.31.mlp.experts.68.gate_proj", "model.layers.31.mlp.experts.69.gate_proj", "model.layers.31.mlp.experts.70.gate_proj", "model.layers.31.mlp.experts.71.gate_proj", "model.layers.31.mlp.experts.72.gate_proj", "model.layers.31.mlp.experts.73.gate_proj", "model.layers.31.mlp.experts.74.gate_proj", "model.layers.31.mlp.experts.75.gate_proj", "model.layers.31.mlp.experts.76.gate_proj", "model.layers.31.mlp.experts.77.gate_proj", "model.layers.31.mlp.experts.78.gate_proj", "model.layers.31.mlp.experts.79.gate_proj", "model.layers.31.mlp.experts.80.gate_proj", "model.layers.31.mlp.experts.81.gate_proj", "model.layers.31.mlp.experts.82.gate_proj", "model.layers.31.mlp.experts.83.gate_proj", "model.layers.31.mlp.experts.84.gate_proj", "model.layers.31.mlp.experts.85.gate_proj", "model.layers.31.mlp.experts.86.gate_proj", "model.layers.31.mlp.experts.87.gate_proj", "model.layers.31.mlp.experts.88.gate_proj", "model.layers.31.mlp.experts.89.gate_proj", "model.layers.31.mlp.experts.90.gate_proj", "model.layers.31.mlp.experts.91.gate_proj", "model.layers.31.mlp.experts.92.gate_proj", "model.layers.31.mlp.experts.93.gate_proj", "model.layers.31.mlp.experts.94.gate_proj", "model.layers.31.mlp.experts.95.gate_proj", "model.layers.31.mlp.experts.96.gate_proj", "model.layers.31.mlp.experts.97.gate_proj", "model.layers.31.mlp.experts.98.gate_proj", "model.layers.31.mlp.experts.99.gate_proj", "model.layers.31.mlp.experts.100.gate_proj", "model.layers.31.mlp.experts.101.gate_proj", "model.layers.31.mlp.experts.102.gate_proj", "model.layers.31.mlp.experts.103.gate_proj", "model.layers.31.mlp.experts.104.gate_proj", "model.layers.31.mlp.experts.105.gate_proj", "model.layers.31.mlp.experts.106.gate_proj", "model.layers.31.mlp.experts.107.gate_proj", "model.layers.31.mlp.experts.108.gate_proj", "model.layers.31.mlp.experts.109.gate_proj", "model.layers.31.mlp.experts.110.gate_proj", "model.layers.31.mlp.experts.111.gate_proj", "model.layers.31.mlp.experts.112.gate_proj", "model.layers.31.mlp.experts.113.gate_proj", "model.layers.31.mlp.experts.114.gate_proj", "model.layers.31.mlp.experts.115.gate_proj", "model.layers.31.mlp.experts.116.gate_proj", "model.layers.31.mlp.experts.117.gate_proj", "model.layers.31.mlp.experts.118.gate_proj", "model.layers.31.mlp.experts.119.gate_proj", "model.layers.31.mlp.experts.120.gate_proj", "model.layers.31.mlp.experts.121.gate_proj", "model.layers.31.mlp.experts.122.gate_proj", "model.layers.31.mlp.experts.123.gate_proj", "model.layers.31.mlp.experts.124.gate_proj", "model.layers.31.mlp.experts.125.gate_proj", "model.layers.31.mlp.experts.126.gate_proj", "model.layers.31.mlp.experts.127.gate_proj", "model.layers.31.mlp.experts.128.gate_proj", "model.layers.31.mlp.experts.129.gate_proj", "model.layers.31.mlp.experts.130.gate_proj", "model.layers.31.mlp.experts.131.gate_proj", "model.layers.31.mlp.experts.132.gate_proj", "model.layers.31.mlp.experts.133.gate_proj", "model.layers.31.mlp.experts.134.gate_proj", "model.layers.31.mlp.experts.135.gate_proj", "model.layers.31.mlp.experts.136.gate_proj", "model.layers.31.mlp.experts.137.gate_proj", "model.layers.31.mlp.experts.138.gate_proj", "model.layers.31.mlp.experts.139.gate_proj", "model.layers.31.mlp.experts.140.gate_proj", "model.layers.31.mlp.experts.141.gate_proj", "model.layers.31.mlp.experts.142.gate_proj", "model.layers.31.mlp.experts.143.gate_proj", "model.layers.31.mlp.experts.144.gate_proj", "model.layers.31.mlp.experts.145.gate_proj", "model.layers.31.mlp.experts.146.gate_proj", "model.layers.31.mlp.experts.147.gate_proj", "model.layers.31.mlp.experts.148.gate_proj", "model.layers.31.mlp.experts.149.gate_proj", "model.layers.31.mlp.experts.150.gate_proj", "model.layers.31.mlp.experts.151.gate_proj", "model.layers.31.mlp.experts.152.gate_proj", "model.layers.31.mlp.experts.153.gate_proj", "model.layers.31.mlp.experts.154.gate_proj", "model.layers.31.mlp.experts.155.gate_proj", "model.layers.31.mlp.experts.156.gate_proj", "model.layers.31.mlp.experts.157.gate_proj", "model.layers.31.mlp.experts.158.gate_proj", "model.layers.31.mlp.experts.159.gate_proj", "model.layers.31.mlp.experts.0.up_proj", "model.layers.31.mlp.experts.1.up_proj", "model.layers.31.mlp.experts.2.up_proj", "model.layers.31.mlp.experts.3.up_proj", "model.layers.31.mlp.experts.4.up_proj", "model.layers.31.mlp.experts.5.up_proj", "model.layers.31.mlp.experts.6.up_proj", "model.layers.31.mlp.experts.7.up_proj", "model.layers.31.mlp.experts.8.up_proj", "model.layers.31.mlp.experts.9.up_proj", "model.layers.31.mlp.experts.10.up_proj", "model.layers.31.mlp.experts.11.up_proj", "model.layers.31.mlp.experts.12.up_proj", "model.layers.31.mlp.experts.13.up_proj", "model.layers.31.mlp.experts.14.up_proj", "model.layers.31.mlp.experts.15.up_proj", "model.layers.31.mlp.experts.16.up_proj", "model.layers.31.mlp.experts.17.up_proj", "model.layers.31.mlp.experts.18.up_proj", "model.layers.31.mlp.experts.19.up_proj", "model.layers.31.mlp.experts.20.up_proj", "model.layers.31.mlp.experts.21.up_proj", "model.layers.31.mlp.experts.22.up_proj", "model.layers.31.mlp.experts.23.up_proj", "model.layers.31.mlp.experts.24.up_proj", "model.layers.31.mlp.experts.25.up_proj", "model.layers.31.mlp.experts.26.up_proj", "model.layers.31.mlp.experts.27.up_proj", "model.layers.31.mlp.experts.28.up_proj", "model.layers.31.mlp.experts.29.up_proj", "model.layers.31.mlp.experts.30.up_proj", "model.layers.31.mlp.experts.31.up_proj", "model.layers.31.mlp.experts.32.up_proj", "model.layers.31.mlp.experts.33.up_proj", "model.layers.31.mlp.experts.34.up_proj", "model.layers.31.mlp.experts.35.up_proj", "model.layers.31.mlp.experts.36.up_proj", "model.layers.31.mlp.experts.37.up_proj", "model.layers.31.mlp.experts.38.up_proj", "model.layers.31.mlp.experts.39.up_proj", "model.layers.31.mlp.experts.40.up_proj", "model.layers.31.mlp.experts.41.up_proj", "model.layers.31.mlp.experts.42.up_proj", "model.layers.31.mlp.experts.43.up_proj", "model.layers.31.mlp.experts.44.up_proj", "model.layers.31.mlp.experts.45.up_proj", "model.layers.31.mlp.experts.46.up_proj", "model.layers.31.mlp.experts.47.up_proj", "model.layers.31.mlp.experts.48.up_proj", "model.layers.31.mlp.experts.49.up_proj", "model.layers.31.mlp.experts.50.up_proj", "model.layers.31.mlp.experts.51.up_proj", "model.layers.31.mlp.experts.52.up_proj", "model.layers.31.mlp.experts.53.up_proj", "model.layers.31.mlp.experts.54.up_proj", "model.layers.31.mlp.experts.55.up_proj", "model.layers.31.mlp.experts.56.up_proj", "model.layers.31.mlp.experts.57.up_proj", "model.layers.31.mlp.experts.58.up_proj", "model.layers.31.mlp.experts.59.up_proj", "model.layers.31.mlp.experts.60.up_proj", "model.layers.31.mlp.experts.61.up_proj", "model.layers.31.mlp.experts.62.up_proj", "model.layers.31.mlp.experts.63.up_proj", "model.layers.31.mlp.experts.64.up_proj", "model.layers.31.mlp.experts.65.up_proj", "model.layers.31.mlp.experts.66.up_proj", "model.layers.31.mlp.experts.67.up_proj", "model.layers.31.mlp.experts.68.up_proj", "model.layers.31.mlp.experts.69.up_proj", "model.layers.31.mlp.experts.70.up_proj", "model.layers.31.mlp.experts.71.up_proj", "model.layers.31.mlp.experts.72.up_proj", "model.layers.31.mlp.experts.73.up_proj", "model.layers.31.mlp.experts.74.up_proj", "model.layers.31.mlp.experts.75.up_proj", "model.layers.31.mlp.experts.76.up_proj", "model.layers.31.mlp.experts.77.up_proj", "model.layers.31.mlp.experts.78.up_proj", "model.layers.31.mlp.experts.79.up_proj", "model.layers.31.mlp.experts.80.up_proj", "model.layers.31.mlp.experts.81.up_proj", "model.layers.31.mlp.experts.82.up_proj", "model.layers.31.mlp.experts.83.up_proj", "model.layers.31.mlp.experts.84.up_proj", "model.layers.31.mlp.experts.85.up_proj", "model.layers.31.mlp.experts.86.up_proj", "model.layers.31.mlp.experts.87.up_proj", "model.layers.31.mlp.experts.88.up_proj", "model.layers.31.mlp.experts.89.up_proj", "model.layers.31.mlp.experts.90.up_proj", "model.layers.31.mlp.experts.91.up_proj", "model.layers.31.mlp.experts.92.up_proj", "model.layers.31.mlp.experts.93.up_proj", "model.layers.31.mlp.experts.94.up_proj", "model.layers.31.mlp.experts.95.up_proj", "model.layers.31.mlp.experts.96.up_proj", "model.layers.31.mlp.experts.97.up_proj", "model.layers.31.mlp.experts.98.up_proj", "model.layers.31.mlp.experts.99.up_proj", "model.layers.31.mlp.experts.100.up_proj", "model.layers.31.mlp.experts.101.up_proj", "model.layers.31.mlp.experts.102.up_proj", "model.layers.31.mlp.experts.103.up_proj", "model.layers.31.mlp.experts.104.up_proj", "model.layers.31.mlp.experts.105.up_proj", "model.layers.31.mlp.experts.106.up_proj", "model.layers.31.mlp.experts.107.up_proj", "model.layers.31.mlp.experts.108.up_proj", "model.layers.31.mlp.experts.109.up_proj", "model.layers.31.mlp.experts.110.up_proj", "model.layers.31.mlp.experts.111.up_proj", "model.layers.31.mlp.experts.112.up_proj", "model.layers.31.mlp.experts.113.up_proj", "model.layers.31.mlp.experts.114.up_proj", "model.layers.31.mlp.experts.115.up_proj", "model.layers.31.mlp.experts.116.up_proj", "model.layers.31.mlp.experts.117.up_proj", "model.layers.31.mlp.experts.118.up_proj", "model.layers.31.mlp.experts.119.up_proj", "model.layers.31.mlp.experts.120.up_proj", "model.layers.31.mlp.experts.121.up_proj", "model.layers.31.mlp.experts.122.up_proj", "model.layers.31.mlp.experts.123.up_proj", "model.layers.31.mlp.experts.124.up_proj", "model.layers.31.mlp.experts.125.up_proj", "model.layers.31.mlp.experts.126.up_proj", "model.layers.31.mlp.experts.127.up_proj", "model.layers.31.mlp.experts.128.up_proj", "model.layers.31.mlp.experts.129.up_proj", "model.layers.31.mlp.experts.130.up_proj", "model.layers.31.mlp.experts.131.up_proj", "model.layers.31.mlp.experts.132.up_proj", "model.layers.31.mlp.experts.133.up_proj", "model.layers.31.mlp.experts.134.up_proj", "model.layers.31.mlp.experts.135.up_proj", "model.layers.31.mlp.experts.136.up_proj", "model.layers.31.mlp.experts.137.up_proj", "model.layers.31.mlp.experts.138.up_proj", "model.layers.31.mlp.experts.139.up_proj", "model.layers.31.mlp.experts.140.up_proj", "model.layers.31.mlp.experts.141.up_proj", "model.layers.31.mlp.experts.142.up_proj", "model.layers.31.mlp.experts.143.up_proj", "model.layers.31.mlp.experts.144.up_proj", "model.layers.31.mlp.experts.145.up_proj", "model.layers.31.mlp.experts.146.up_proj", "model.layers.31.mlp.experts.147.up_proj", "model.layers.31.mlp.experts.148.up_proj", "model.layers.31.mlp.experts.149.up_proj", "model.layers.31.mlp.experts.150.up_proj", "model.layers.31.mlp.experts.151.up_proj", "model.layers.31.mlp.experts.152.up_proj", "model.layers.31.mlp.experts.153.up_proj", "model.layers.31.mlp.experts.154.up_proj", "model.layers.31.mlp.experts.155.up_proj", "model.layers.31.mlp.experts.156.up_proj", "model.layers.31.mlp.experts.157.up_proj", "model.layers.31.mlp.experts.158.up_proj", "model.layers.31.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0006350597366690663, "dbits": 2516582400 }, { "dkld": -0.0009248911403119675, "dbits": 5033164800 }, { "dkld": -0.0009138935245573548, "dbits": 7549747200 }, { "dkld": -0.000978244841098791, "dbits": 12582912000 } ] }, { "idx": 188, "layers": [ "model.layers.31.mlp.experts.0.down_proj", "model.layers.31.mlp.experts.1.down_proj", "model.layers.31.mlp.experts.2.down_proj", "model.layers.31.mlp.experts.3.down_proj", "model.layers.31.mlp.experts.4.down_proj", "model.layers.31.mlp.experts.5.down_proj", "model.layers.31.mlp.experts.6.down_proj", "model.layers.31.mlp.experts.7.down_proj", "model.layers.31.mlp.experts.8.down_proj", "model.layers.31.mlp.experts.9.down_proj", "model.layers.31.mlp.experts.10.down_proj", "model.layers.31.mlp.experts.11.down_proj", "model.layers.31.mlp.experts.12.down_proj", "model.layers.31.mlp.experts.13.down_proj", "model.layers.31.mlp.experts.14.down_proj", "model.layers.31.mlp.experts.15.down_proj", "model.layers.31.mlp.experts.16.down_proj", "model.layers.31.mlp.experts.17.down_proj", "model.layers.31.mlp.experts.18.down_proj", "model.layers.31.mlp.experts.19.down_proj", "model.layers.31.mlp.experts.20.down_proj", "model.layers.31.mlp.experts.21.down_proj", "model.layers.31.mlp.experts.22.down_proj", "model.layers.31.mlp.experts.23.down_proj", "model.layers.31.mlp.experts.24.down_proj", "model.layers.31.mlp.experts.25.down_proj", "model.layers.31.mlp.experts.26.down_proj", "model.layers.31.mlp.experts.27.down_proj", "model.layers.31.mlp.experts.28.down_proj", "model.layers.31.mlp.experts.29.down_proj", "model.layers.31.mlp.experts.30.down_proj", "model.layers.31.mlp.experts.31.down_proj", "model.layers.31.mlp.experts.32.down_proj", "model.layers.31.mlp.experts.33.down_proj", "model.layers.31.mlp.experts.34.down_proj", "model.layers.31.mlp.experts.35.down_proj", "model.layers.31.mlp.experts.36.down_proj", "model.layers.31.mlp.experts.37.down_proj", "model.layers.31.mlp.experts.38.down_proj", "model.layers.31.mlp.experts.39.down_proj", "model.layers.31.mlp.experts.40.down_proj", "model.layers.31.mlp.experts.41.down_proj", "model.layers.31.mlp.experts.42.down_proj", "model.layers.31.mlp.experts.43.down_proj", "model.layers.31.mlp.experts.44.down_proj", "model.layers.31.mlp.experts.45.down_proj", "model.layers.31.mlp.experts.46.down_proj", "model.layers.31.mlp.experts.47.down_proj", "model.layers.31.mlp.experts.48.down_proj", "model.layers.31.mlp.experts.49.down_proj", "model.layers.31.mlp.experts.50.down_proj", "model.layers.31.mlp.experts.51.down_proj", "model.layers.31.mlp.experts.52.down_proj", "model.layers.31.mlp.experts.53.down_proj", "model.layers.31.mlp.experts.54.down_proj", "model.layers.31.mlp.experts.55.down_proj", "model.layers.31.mlp.experts.56.down_proj", "model.layers.31.mlp.experts.57.down_proj", "model.layers.31.mlp.experts.58.down_proj", "model.layers.31.mlp.experts.59.down_proj", "model.layers.31.mlp.experts.60.down_proj", "model.layers.31.mlp.experts.61.down_proj", "model.layers.31.mlp.experts.62.down_proj", "model.layers.31.mlp.experts.63.down_proj", "model.layers.31.mlp.experts.64.down_proj", "model.layers.31.mlp.experts.65.down_proj", "model.layers.31.mlp.experts.66.down_proj", "model.layers.31.mlp.experts.67.down_proj", "model.layers.31.mlp.experts.68.down_proj", "model.layers.31.mlp.experts.69.down_proj", "model.layers.31.mlp.experts.70.down_proj", "model.layers.31.mlp.experts.71.down_proj", "model.layers.31.mlp.experts.72.down_proj", "model.layers.31.mlp.experts.73.down_proj", "model.layers.31.mlp.experts.74.down_proj", "model.layers.31.mlp.experts.75.down_proj", "model.layers.31.mlp.experts.76.down_proj", "model.layers.31.mlp.experts.77.down_proj", "model.layers.31.mlp.experts.78.down_proj", "model.layers.31.mlp.experts.79.down_proj", "model.layers.31.mlp.experts.80.down_proj", "model.layers.31.mlp.experts.81.down_proj", "model.layers.31.mlp.experts.82.down_proj", "model.layers.31.mlp.experts.83.down_proj", "model.layers.31.mlp.experts.84.down_proj", "model.layers.31.mlp.experts.85.down_proj", "model.layers.31.mlp.experts.86.down_proj", "model.layers.31.mlp.experts.87.down_proj", "model.layers.31.mlp.experts.88.down_proj", "model.layers.31.mlp.experts.89.down_proj", "model.layers.31.mlp.experts.90.down_proj", "model.layers.31.mlp.experts.91.down_proj", "model.layers.31.mlp.experts.92.down_proj", "model.layers.31.mlp.experts.93.down_proj", "model.layers.31.mlp.experts.94.down_proj", "model.layers.31.mlp.experts.95.down_proj", "model.layers.31.mlp.experts.96.down_proj", "model.layers.31.mlp.experts.97.down_proj", "model.layers.31.mlp.experts.98.down_proj", "model.layers.31.mlp.experts.99.down_proj", "model.layers.31.mlp.experts.100.down_proj", "model.layers.31.mlp.experts.101.down_proj", "model.layers.31.mlp.experts.102.down_proj", "model.layers.31.mlp.experts.103.down_proj", "model.layers.31.mlp.experts.104.down_proj", "model.layers.31.mlp.experts.105.down_proj", "model.layers.31.mlp.experts.106.down_proj", "model.layers.31.mlp.experts.107.down_proj", "model.layers.31.mlp.experts.108.down_proj", "model.layers.31.mlp.experts.109.down_proj", "model.layers.31.mlp.experts.110.down_proj", "model.layers.31.mlp.experts.111.down_proj", "model.layers.31.mlp.experts.112.down_proj", "model.layers.31.mlp.experts.113.down_proj", "model.layers.31.mlp.experts.114.down_proj", "model.layers.31.mlp.experts.115.down_proj", "model.layers.31.mlp.experts.116.down_proj", "model.layers.31.mlp.experts.117.down_proj", "model.layers.31.mlp.experts.118.down_proj", "model.layers.31.mlp.experts.119.down_proj", "model.layers.31.mlp.experts.120.down_proj", "model.layers.31.mlp.experts.121.down_proj", "model.layers.31.mlp.experts.122.down_proj", "model.layers.31.mlp.experts.123.down_proj", "model.layers.31.mlp.experts.124.down_proj", "model.layers.31.mlp.experts.125.down_proj", "model.layers.31.mlp.experts.126.down_proj", "model.layers.31.mlp.experts.127.down_proj", "model.layers.31.mlp.experts.128.down_proj", "model.layers.31.mlp.experts.129.down_proj", "model.layers.31.mlp.experts.130.down_proj", "model.layers.31.mlp.experts.131.down_proj", "model.layers.31.mlp.experts.132.down_proj", "model.layers.31.mlp.experts.133.down_proj", "model.layers.31.mlp.experts.134.down_proj", "model.layers.31.mlp.experts.135.down_proj", "model.layers.31.mlp.experts.136.down_proj", "model.layers.31.mlp.experts.137.down_proj", "model.layers.31.mlp.experts.138.down_proj", "model.layers.31.mlp.experts.139.down_proj", "model.layers.31.mlp.experts.140.down_proj", "model.layers.31.mlp.experts.141.down_proj", "model.layers.31.mlp.experts.142.down_proj", "model.layers.31.mlp.experts.143.down_proj", "model.layers.31.mlp.experts.144.down_proj", "model.layers.31.mlp.experts.145.down_proj", "model.layers.31.mlp.experts.146.down_proj", "model.layers.31.mlp.experts.147.down_proj", "model.layers.31.mlp.experts.148.down_proj", "model.layers.31.mlp.experts.149.down_proj", "model.layers.31.mlp.experts.150.down_proj", "model.layers.31.mlp.experts.151.down_proj", "model.layers.31.mlp.experts.152.down_proj", "model.layers.31.mlp.experts.153.down_proj", "model.layers.31.mlp.experts.154.down_proj", "model.layers.31.mlp.experts.155.down_proj", "model.layers.31.mlp.experts.156.down_proj", "model.layers.31.mlp.experts.157.down_proj", "model.layers.31.mlp.experts.158.down_proj", "model.layers.31.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -6.74067996442318e-05, "dbits": 1258291200 }, { "dkld": -0.00039897076785565255, "dbits": 2516582400 }, { "dkld": -0.0004045315086841639, "dbits": 3774873600 }, { "dkld": -0.00043018441647291183, "dbits": 6291456000 } ] }, { "idx": 189, "layers": [ "model.layers.32.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0005651433020830154, "dbits": 62914560 }, { "dkld": 0.0006235813722014344, "dbits": 125829120 }, { "dkld": 0.00047706346958875656, "dbits": 188743680 }, { "dkld": 0.0005637466907501137, "dbits": 314572800 } ] }, { "idx": 190, "layers": [ "model.layers.32.self_attn.k_proj", "model.layers.32.self_attn.v_proj" ], "candidates": [ { "dkld": -1.4560855925091487e-05, "dbits": 10485760 }, { "dkld": 0.00041446369141340256, "dbits": 20971520 }, { "dkld": 0.0001778709702193737, "dbits": 31457280 }, { "dkld": -0.00013557747006416598, "dbits": 52428800 } ] }, { "idx": 191, "layers": [ "model.layers.32.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0007000209763646237, "dbits": 62914560 }, { "dkld": -0.00018016807734966278, "dbits": 125829120 }, { "dkld": -2.1189730614434854e-05, "dbits": 188743680 }, { "dkld": -2.9335822910078746e-05, "dbits": 314572800 } ] }, { "idx": 192, "layers": [ "model.layers.32.mlp.shared_experts.gate_proj", "model.layers.32.mlp.shared_experts.up_proj", "model.layers.32.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0011922221630811664, "dbits": 23592960 }, { "dkld": 0.0016415394842624609, "dbits": 47185920 }, { "dkld": 0.0014527064748108276, "dbits": 70778880 }, { "dkld": 0.001523420400917519, "dbits": 117964800 } ] }, { "idx": 193, "layers": [ "model.layers.32.mlp.experts.0.gate_proj", "model.layers.32.mlp.experts.1.gate_proj", "model.layers.32.mlp.experts.2.gate_proj", "model.layers.32.mlp.experts.3.gate_proj", "model.layers.32.mlp.experts.4.gate_proj", "model.layers.32.mlp.experts.5.gate_proj", "model.layers.32.mlp.experts.6.gate_proj", "model.layers.32.mlp.experts.7.gate_proj", "model.layers.32.mlp.experts.8.gate_proj", "model.layers.32.mlp.experts.9.gate_proj", "model.layers.32.mlp.experts.10.gate_proj", "model.layers.32.mlp.experts.11.gate_proj", "model.layers.32.mlp.experts.12.gate_proj", "model.layers.32.mlp.experts.13.gate_proj", "model.layers.32.mlp.experts.14.gate_proj", "model.layers.32.mlp.experts.15.gate_proj", "model.layers.32.mlp.experts.16.gate_proj", "model.layers.32.mlp.experts.17.gate_proj", "model.layers.32.mlp.experts.18.gate_proj", "model.layers.32.mlp.experts.19.gate_proj", "model.layers.32.mlp.experts.20.gate_proj", "model.layers.32.mlp.experts.21.gate_proj", "model.layers.32.mlp.experts.22.gate_proj", "model.layers.32.mlp.experts.23.gate_proj", "model.layers.32.mlp.experts.24.gate_proj", "model.layers.32.mlp.experts.25.gate_proj", "model.layers.32.mlp.experts.26.gate_proj", "model.layers.32.mlp.experts.27.gate_proj", "model.layers.32.mlp.experts.28.gate_proj", "model.layers.32.mlp.experts.29.gate_proj", "model.layers.32.mlp.experts.30.gate_proj", "model.layers.32.mlp.experts.31.gate_proj", "model.layers.32.mlp.experts.32.gate_proj", "model.layers.32.mlp.experts.33.gate_proj", "model.layers.32.mlp.experts.34.gate_proj", "model.layers.32.mlp.experts.35.gate_proj", "model.layers.32.mlp.experts.36.gate_proj", "model.layers.32.mlp.experts.37.gate_proj", "model.layers.32.mlp.experts.38.gate_proj", "model.layers.32.mlp.experts.39.gate_proj", "model.layers.32.mlp.experts.40.gate_proj", "model.layers.32.mlp.experts.41.gate_proj", "model.layers.32.mlp.experts.42.gate_proj", "model.layers.32.mlp.experts.43.gate_proj", "model.layers.32.mlp.experts.44.gate_proj", "model.layers.32.mlp.experts.45.gate_proj", "model.layers.32.mlp.experts.46.gate_proj", "model.layers.32.mlp.experts.47.gate_proj", "model.layers.32.mlp.experts.48.gate_proj", "model.layers.32.mlp.experts.49.gate_proj", "model.layers.32.mlp.experts.50.gate_proj", "model.layers.32.mlp.experts.51.gate_proj", "model.layers.32.mlp.experts.52.gate_proj", "model.layers.32.mlp.experts.53.gate_proj", "model.layers.32.mlp.experts.54.gate_proj", "model.layers.32.mlp.experts.55.gate_proj", "model.layers.32.mlp.experts.56.gate_proj", "model.layers.32.mlp.experts.57.gate_proj", "model.layers.32.mlp.experts.58.gate_proj", "model.layers.32.mlp.experts.59.gate_proj", "model.layers.32.mlp.experts.60.gate_proj", "model.layers.32.mlp.experts.61.gate_proj", "model.layers.32.mlp.experts.62.gate_proj", "model.layers.32.mlp.experts.63.gate_proj", "model.layers.32.mlp.experts.64.gate_proj", "model.layers.32.mlp.experts.65.gate_proj", "model.layers.32.mlp.experts.66.gate_proj", "model.layers.32.mlp.experts.67.gate_proj", "model.layers.32.mlp.experts.68.gate_proj", "model.layers.32.mlp.experts.69.gate_proj", "model.layers.32.mlp.experts.70.gate_proj", "model.layers.32.mlp.experts.71.gate_proj", "model.layers.32.mlp.experts.72.gate_proj", "model.layers.32.mlp.experts.73.gate_proj", "model.layers.32.mlp.experts.74.gate_proj", "model.layers.32.mlp.experts.75.gate_proj", "model.layers.32.mlp.experts.76.gate_proj", "model.layers.32.mlp.experts.77.gate_proj", "model.layers.32.mlp.experts.78.gate_proj", "model.layers.32.mlp.experts.79.gate_proj", "model.layers.32.mlp.experts.80.gate_proj", "model.layers.32.mlp.experts.81.gate_proj", "model.layers.32.mlp.experts.82.gate_proj", "model.layers.32.mlp.experts.83.gate_proj", "model.layers.32.mlp.experts.84.gate_proj", "model.layers.32.mlp.experts.85.gate_proj", "model.layers.32.mlp.experts.86.gate_proj", "model.layers.32.mlp.experts.87.gate_proj", "model.layers.32.mlp.experts.88.gate_proj", "model.layers.32.mlp.experts.89.gate_proj", "model.layers.32.mlp.experts.90.gate_proj", "model.layers.32.mlp.experts.91.gate_proj", "model.layers.32.mlp.experts.92.gate_proj", "model.layers.32.mlp.experts.93.gate_proj", "model.layers.32.mlp.experts.94.gate_proj", "model.layers.32.mlp.experts.95.gate_proj", "model.layers.32.mlp.experts.96.gate_proj", "model.layers.32.mlp.experts.97.gate_proj", "model.layers.32.mlp.experts.98.gate_proj", "model.layers.32.mlp.experts.99.gate_proj", "model.layers.32.mlp.experts.100.gate_proj", "model.layers.32.mlp.experts.101.gate_proj", "model.layers.32.mlp.experts.102.gate_proj", "model.layers.32.mlp.experts.103.gate_proj", "model.layers.32.mlp.experts.104.gate_proj", "model.layers.32.mlp.experts.105.gate_proj", "model.layers.32.mlp.experts.106.gate_proj", "model.layers.32.mlp.experts.107.gate_proj", "model.layers.32.mlp.experts.108.gate_proj", "model.layers.32.mlp.experts.109.gate_proj", "model.layers.32.mlp.experts.110.gate_proj", "model.layers.32.mlp.experts.111.gate_proj", "model.layers.32.mlp.experts.112.gate_proj", "model.layers.32.mlp.experts.113.gate_proj", "model.layers.32.mlp.experts.114.gate_proj", "model.layers.32.mlp.experts.115.gate_proj", "model.layers.32.mlp.experts.116.gate_proj", "model.layers.32.mlp.experts.117.gate_proj", "model.layers.32.mlp.experts.118.gate_proj", "model.layers.32.mlp.experts.119.gate_proj", "model.layers.32.mlp.experts.120.gate_proj", "model.layers.32.mlp.experts.121.gate_proj", "model.layers.32.mlp.experts.122.gate_proj", "model.layers.32.mlp.experts.123.gate_proj", "model.layers.32.mlp.experts.124.gate_proj", "model.layers.32.mlp.experts.125.gate_proj", "model.layers.32.mlp.experts.126.gate_proj", "model.layers.32.mlp.experts.127.gate_proj", "model.layers.32.mlp.experts.128.gate_proj", "model.layers.32.mlp.experts.129.gate_proj", "model.layers.32.mlp.experts.130.gate_proj", "model.layers.32.mlp.experts.131.gate_proj", "model.layers.32.mlp.experts.132.gate_proj", "model.layers.32.mlp.experts.133.gate_proj", "model.layers.32.mlp.experts.134.gate_proj", "model.layers.32.mlp.experts.135.gate_proj", "model.layers.32.mlp.experts.136.gate_proj", "model.layers.32.mlp.experts.137.gate_proj", "model.layers.32.mlp.experts.138.gate_proj", "model.layers.32.mlp.experts.139.gate_proj", "model.layers.32.mlp.experts.140.gate_proj", "model.layers.32.mlp.experts.141.gate_proj", "model.layers.32.mlp.experts.142.gate_proj", "model.layers.32.mlp.experts.143.gate_proj", "model.layers.32.mlp.experts.144.gate_proj", "model.layers.32.mlp.experts.145.gate_proj", "model.layers.32.mlp.experts.146.gate_proj", "model.layers.32.mlp.experts.147.gate_proj", "model.layers.32.mlp.experts.148.gate_proj", "model.layers.32.mlp.experts.149.gate_proj", "model.layers.32.mlp.experts.150.gate_proj", "model.layers.32.mlp.experts.151.gate_proj", "model.layers.32.mlp.experts.152.gate_proj", "model.layers.32.mlp.experts.153.gate_proj", "model.layers.32.mlp.experts.154.gate_proj", "model.layers.32.mlp.experts.155.gate_proj", "model.layers.32.mlp.experts.156.gate_proj", "model.layers.32.mlp.experts.157.gate_proj", "model.layers.32.mlp.experts.158.gate_proj", "model.layers.32.mlp.experts.159.gate_proj", "model.layers.32.mlp.experts.0.up_proj", "model.layers.32.mlp.experts.1.up_proj", "model.layers.32.mlp.experts.2.up_proj", "model.layers.32.mlp.experts.3.up_proj", "model.layers.32.mlp.experts.4.up_proj", "model.layers.32.mlp.experts.5.up_proj", "model.layers.32.mlp.experts.6.up_proj", "model.layers.32.mlp.experts.7.up_proj", "model.layers.32.mlp.experts.8.up_proj", "model.layers.32.mlp.experts.9.up_proj", "model.layers.32.mlp.experts.10.up_proj", "model.layers.32.mlp.experts.11.up_proj", "model.layers.32.mlp.experts.12.up_proj", "model.layers.32.mlp.experts.13.up_proj", "model.layers.32.mlp.experts.14.up_proj", "model.layers.32.mlp.experts.15.up_proj", "model.layers.32.mlp.experts.16.up_proj", "model.layers.32.mlp.experts.17.up_proj", "model.layers.32.mlp.experts.18.up_proj", "model.layers.32.mlp.experts.19.up_proj", "model.layers.32.mlp.experts.20.up_proj", "model.layers.32.mlp.experts.21.up_proj", "model.layers.32.mlp.experts.22.up_proj", "model.layers.32.mlp.experts.23.up_proj", "model.layers.32.mlp.experts.24.up_proj", "model.layers.32.mlp.experts.25.up_proj", "model.layers.32.mlp.experts.26.up_proj", "model.layers.32.mlp.experts.27.up_proj", "model.layers.32.mlp.experts.28.up_proj", "model.layers.32.mlp.experts.29.up_proj", "model.layers.32.mlp.experts.30.up_proj", "model.layers.32.mlp.experts.31.up_proj", "model.layers.32.mlp.experts.32.up_proj", "model.layers.32.mlp.experts.33.up_proj", "model.layers.32.mlp.experts.34.up_proj", "model.layers.32.mlp.experts.35.up_proj", "model.layers.32.mlp.experts.36.up_proj", "model.layers.32.mlp.experts.37.up_proj", "model.layers.32.mlp.experts.38.up_proj", "model.layers.32.mlp.experts.39.up_proj", "model.layers.32.mlp.experts.40.up_proj", "model.layers.32.mlp.experts.41.up_proj", "model.layers.32.mlp.experts.42.up_proj", "model.layers.32.mlp.experts.43.up_proj", "model.layers.32.mlp.experts.44.up_proj", "model.layers.32.mlp.experts.45.up_proj", "model.layers.32.mlp.experts.46.up_proj", "model.layers.32.mlp.experts.47.up_proj", "model.layers.32.mlp.experts.48.up_proj", "model.layers.32.mlp.experts.49.up_proj", "model.layers.32.mlp.experts.50.up_proj", "model.layers.32.mlp.experts.51.up_proj", "model.layers.32.mlp.experts.52.up_proj", "model.layers.32.mlp.experts.53.up_proj", "model.layers.32.mlp.experts.54.up_proj", "model.layers.32.mlp.experts.55.up_proj", "model.layers.32.mlp.experts.56.up_proj", "model.layers.32.mlp.experts.57.up_proj", "model.layers.32.mlp.experts.58.up_proj", "model.layers.32.mlp.experts.59.up_proj", "model.layers.32.mlp.experts.60.up_proj", "model.layers.32.mlp.experts.61.up_proj", "model.layers.32.mlp.experts.62.up_proj", "model.layers.32.mlp.experts.63.up_proj", "model.layers.32.mlp.experts.64.up_proj", "model.layers.32.mlp.experts.65.up_proj", "model.layers.32.mlp.experts.66.up_proj", "model.layers.32.mlp.experts.67.up_proj", "model.layers.32.mlp.experts.68.up_proj", "model.layers.32.mlp.experts.69.up_proj", "model.layers.32.mlp.experts.70.up_proj", "model.layers.32.mlp.experts.71.up_proj", "model.layers.32.mlp.experts.72.up_proj", "model.layers.32.mlp.experts.73.up_proj", "model.layers.32.mlp.experts.74.up_proj", "model.layers.32.mlp.experts.75.up_proj", "model.layers.32.mlp.experts.76.up_proj", "model.layers.32.mlp.experts.77.up_proj", "model.layers.32.mlp.experts.78.up_proj", "model.layers.32.mlp.experts.79.up_proj", "model.layers.32.mlp.experts.80.up_proj", "model.layers.32.mlp.experts.81.up_proj", "model.layers.32.mlp.experts.82.up_proj", "model.layers.32.mlp.experts.83.up_proj", "model.layers.32.mlp.experts.84.up_proj", "model.layers.32.mlp.experts.85.up_proj", "model.layers.32.mlp.experts.86.up_proj", "model.layers.32.mlp.experts.87.up_proj", "model.layers.32.mlp.experts.88.up_proj", "model.layers.32.mlp.experts.89.up_proj", "model.layers.32.mlp.experts.90.up_proj", "model.layers.32.mlp.experts.91.up_proj", "model.layers.32.mlp.experts.92.up_proj", "model.layers.32.mlp.experts.93.up_proj", "model.layers.32.mlp.experts.94.up_proj", "model.layers.32.mlp.experts.95.up_proj", "model.layers.32.mlp.experts.96.up_proj", "model.layers.32.mlp.experts.97.up_proj", "model.layers.32.mlp.experts.98.up_proj", "model.layers.32.mlp.experts.99.up_proj", "model.layers.32.mlp.experts.100.up_proj", "model.layers.32.mlp.experts.101.up_proj", "model.layers.32.mlp.experts.102.up_proj", "model.layers.32.mlp.experts.103.up_proj", "model.layers.32.mlp.experts.104.up_proj", "model.layers.32.mlp.experts.105.up_proj", "model.layers.32.mlp.experts.106.up_proj", "model.layers.32.mlp.experts.107.up_proj", "model.layers.32.mlp.experts.108.up_proj", "model.layers.32.mlp.experts.109.up_proj", "model.layers.32.mlp.experts.110.up_proj", "model.layers.32.mlp.experts.111.up_proj", "model.layers.32.mlp.experts.112.up_proj", "model.layers.32.mlp.experts.113.up_proj", "model.layers.32.mlp.experts.114.up_proj", "model.layers.32.mlp.experts.115.up_proj", "model.layers.32.mlp.experts.116.up_proj", "model.layers.32.mlp.experts.117.up_proj", "model.layers.32.mlp.experts.118.up_proj", "model.layers.32.mlp.experts.119.up_proj", "model.layers.32.mlp.experts.120.up_proj", "model.layers.32.mlp.experts.121.up_proj", "model.layers.32.mlp.experts.122.up_proj", "model.layers.32.mlp.experts.123.up_proj", "model.layers.32.mlp.experts.124.up_proj", "model.layers.32.mlp.experts.125.up_proj", "model.layers.32.mlp.experts.126.up_proj", "model.layers.32.mlp.experts.127.up_proj", "model.layers.32.mlp.experts.128.up_proj", "model.layers.32.mlp.experts.129.up_proj", "model.layers.32.mlp.experts.130.up_proj", "model.layers.32.mlp.experts.131.up_proj", "model.layers.32.mlp.experts.132.up_proj", "model.layers.32.mlp.experts.133.up_proj", "model.layers.32.mlp.experts.134.up_proj", "model.layers.32.mlp.experts.135.up_proj", "model.layers.32.mlp.experts.136.up_proj", "model.layers.32.mlp.experts.137.up_proj", "model.layers.32.mlp.experts.138.up_proj", "model.layers.32.mlp.experts.139.up_proj", "model.layers.32.mlp.experts.140.up_proj", "model.layers.32.mlp.experts.141.up_proj", "model.layers.32.mlp.experts.142.up_proj", "model.layers.32.mlp.experts.143.up_proj", "model.layers.32.mlp.experts.144.up_proj", "model.layers.32.mlp.experts.145.up_proj", "model.layers.32.mlp.experts.146.up_proj", "model.layers.32.mlp.experts.147.up_proj", "model.layers.32.mlp.experts.148.up_proj", "model.layers.32.mlp.experts.149.up_proj", "model.layers.32.mlp.experts.150.up_proj", "model.layers.32.mlp.experts.151.up_proj", "model.layers.32.mlp.experts.152.up_proj", "model.layers.32.mlp.experts.153.up_proj", "model.layers.32.mlp.experts.154.up_proj", "model.layers.32.mlp.experts.155.up_proj", "model.layers.32.mlp.experts.156.up_proj", "model.layers.32.mlp.experts.157.up_proj", "model.layers.32.mlp.experts.158.up_proj", "model.layers.32.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00025029424577951986, "dbits": 2516582400 }, { "dkld": -0.000378178991377362, "dbits": 5033164800 }, { "dkld": -0.00036031100898981094, "dbits": 7549747200 }, { "dkld": -0.0004301173612475395, "dbits": 12582912000 } ] }, { "idx": 194, "layers": [ "model.layers.32.mlp.experts.0.down_proj", "model.layers.32.mlp.experts.1.down_proj", "model.layers.32.mlp.experts.2.down_proj", "model.layers.32.mlp.experts.3.down_proj", "model.layers.32.mlp.experts.4.down_proj", "model.layers.32.mlp.experts.5.down_proj", "model.layers.32.mlp.experts.6.down_proj", "model.layers.32.mlp.experts.7.down_proj", "model.layers.32.mlp.experts.8.down_proj", "model.layers.32.mlp.experts.9.down_proj", "model.layers.32.mlp.experts.10.down_proj", "model.layers.32.mlp.experts.11.down_proj", "model.layers.32.mlp.experts.12.down_proj", "model.layers.32.mlp.experts.13.down_proj", "model.layers.32.mlp.experts.14.down_proj", "model.layers.32.mlp.experts.15.down_proj", "model.layers.32.mlp.experts.16.down_proj", "model.layers.32.mlp.experts.17.down_proj", "model.layers.32.mlp.experts.18.down_proj", "model.layers.32.mlp.experts.19.down_proj", "model.layers.32.mlp.experts.20.down_proj", "model.layers.32.mlp.experts.21.down_proj", "model.layers.32.mlp.experts.22.down_proj", "model.layers.32.mlp.experts.23.down_proj", "model.layers.32.mlp.experts.24.down_proj", "model.layers.32.mlp.experts.25.down_proj", "model.layers.32.mlp.experts.26.down_proj", "model.layers.32.mlp.experts.27.down_proj", "model.layers.32.mlp.experts.28.down_proj", "model.layers.32.mlp.experts.29.down_proj", "model.layers.32.mlp.experts.30.down_proj", "model.layers.32.mlp.experts.31.down_proj", "model.layers.32.mlp.experts.32.down_proj", "model.layers.32.mlp.experts.33.down_proj", "model.layers.32.mlp.experts.34.down_proj", "model.layers.32.mlp.experts.35.down_proj", "model.layers.32.mlp.experts.36.down_proj", "model.layers.32.mlp.experts.37.down_proj", "model.layers.32.mlp.experts.38.down_proj", "model.layers.32.mlp.experts.39.down_proj", "model.layers.32.mlp.experts.40.down_proj", "model.layers.32.mlp.experts.41.down_proj", "model.layers.32.mlp.experts.42.down_proj", "model.layers.32.mlp.experts.43.down_proj", "model.layers.32.mlp.experts.44.down_proj", "model.layers.32.mlp.experts.45.down_proj", "model.layers.32.mlp.experts.46.down_proj", "model.layers.32.mlp.experts.47.down_proj", "model.layers.32.mlp.experts.48.down_proj", "model.layers.32.mlp.experts.49.down_proj", "model.layers.32.mlp.experts.50.down_proj", "model.layers.32.mlp.experts.51.down_proj", "model.layers.32.mlp.experts.52.down_proj", "model.layers.32.mlp.experts.53.down_proj", "model.layers.32.mlp.experts.54.down_proj", "model.layers.32.mlp.experts.55.down_proj", "model.layers.32.mlp.experts.56.down_proj", "model.layers.32.mlp.experts.57.down_proj", "model.layers.32.mlp.experts.58.down_proj", "model.layers.32.mlp.experts.59.down_proj", "model.layers.32.mlp.experts.60.down_proj", "model.layers.32.mlp.experts.61.down_proj", "model.layers.32.mlp.experts.62.down_proj", "model.layers.32.mlp.experts.63.down_proj", "model.layers.32.mlp.experts.64.down_proj", "model.layers.32.mlp.experts.65.down_proj", "model.layers.32.mlp.experts.66.down_proj", "model.layers.32.mlp.experts.67.down_proj", "model.layers.32.mlp.experts.68.down_proj", "model.layers.32.mlp.experts.69.down_proj", "model.layers.32.mlp.experts.70.down_proj", "model.layers.32.mlp.experts.71.down_proj", "model.layers.32.mlp.experts.72.down_proj", "model.layers.32.mlp.experts.73.down_proj", "model.layers.32.mlp.experts.74.down_proj", "model.layers.32.mlp.experts.75.down_proj", "model.layers.32.mlp.experts.76.down_proj", "model.layers.32.mlp.experts.77.down_proj", "model.layers.32.mlp.experts.78.down_proj", "model.layers.32.mlp.experts.79.down_proj", "model.layers.32.mlp.experts.80.down_proj", "model.layers.32.mlp.experts.81.down_proj", "model.layers.32.mlp.experts.82.down_proj", "model.layers.32.mlp.experts.83.down_proj", "model.layers.32.mlp.experts.84.down_proj", "model.layers.32.mlp.experts.85.down_proj", "model.layers.32.mlp.experts.86.down_proj", "model.layers.32.mlp.experts.87.down_proj", "model.layers.32.mlp.experts.88.down_proj", "model.layers.32.mlp.experts.89.down_proj", "model.layers.32.mlp.experts.90.down_proj", "model.layers.32.mlp.experts.91.down_proj", "model.layers.32.mlp.experts.92.down_proj", "model.layers.32.mlp.experts.93.down_proj", "model.layers.32.mlp.experts.94.down_proj", "model.layers.32.mlp.experts.95.down_proj", "model.layers.32.mlp.experts.96.down_proj", "model.layers.32.mlp.experts.97.down_proj", "model.layers.32.mlp.experts.98.down_proj", "model.layers.32.mlp.experts.99.down_proj", "model.layers.32.mlp.experts.100.down_proj", "model.layers.32.mlp.experts.101.down_proj", "model.layers.32.mlp.experts.102.down_proj", "model.layers.32.mlp.experts.103.down_proj", "model.layers.32.mlp.experts.104.down_proj", "model.layers.32.mlp.experts.105.down_proj", "model.layers.32.mlp.experts.106.down_proj", "model.layers.32.mlp.experts.107.down_proj", "model.layers.32.mlp.experts.108.down_proj", "model.layers.32.mlp.experts.109.down_proj", "model.layers.32.mlp.experts.110.down_proj", "model.layers.32.mlp.experts.111.down_proj", "model.layers.32.mlp.experts.112.down_proj", "model.layers.32.mlp.experts.113.down_proj", "model.layers.32.mlp.experts.114.down_proj", "model.layers.32.mlp.experts.115.down_proj", "model.layers.32.mlp.experts.116.down_proj", "model.layers.32.mlp.experts.117.down_proj", "model.layers.32.mlp.experts.118.down_proj", "model.layers.32.mlp.experts.119.down_proj", "model.layers.32.mlp.experts.120.down_proj", "model.layers.32.mlp.experts.121.down_proj", "model.layers.32.mlp.experts.122.down_proj", "model.layers.32.mlp.experts.123.down_proj", "model.layers.32.mlp.experts.124.down_proj", "model.layers.32.mlp.experts.125.down_proj", "model.layers.32.mlp.experts.126.down_proj", "model.layers.32.mlp.experts.127.down_proj", "model.layers.32.mlp.experts.128.down_proj", "model.layers.32.mlp.experts.129.down_proj", "model.layers.32.mlp.experts.130.down_proj", "model.layers.32.mlp.experts.131.down_proj", "model.layers.32.mlp.experts.132.down_proj", "model.layers.32.mlp.experts.133.down_proj", "model.layers.32.mlp.experts.134.down_proj", "model.layers.32.mlp.experts.135.down_proj", "model.layers.32.mlp.experts.136.down_proj", "model.layers.32.mlp.experts.137.down_proj", "model.layers.32.mlp.experts.138.down_proj", "model.layers.32.mlp.experts.139.down_proj", "model.layers.32.mlp.experts.140.down_proj", "model.layers.32.mlp.experts.141.down_proj", "model.layers.32.mlp.experts.142.down_proj", "model.layers.32.mlp.experts.143.down_proj", "model.layers.32.mlp.experts.144.down_proj", "model.layers.32.mlp.experts.145.down_proj", "model.layers.32.mlp.experts.146.down_proj", "model.layers.32.mlp.experts.147.down_proj", "model.layers.32.mlp.experts.148.down_proj", "model.layers.32.mlp.experts.149.down_proj", "model.layers.32.mlp.experts.150.down_proj", "model.layers.32.mlp.experts.151.down_proj", "model.layers.32.mlp.experts.152.down_proj", "model.layers.32.mlp.experts.153.down_proj", "model.layers.32.mlp.experts.154.down_proj", "model.layers.32.mlp.experts.155.down_proj", "model.layers.32.mlp.experts.156.down_proj", "model.layers.32.mlp.experts.157.down_proj", "model.layers.32.mlp.experts.158.down_proj", "model.layers.32.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00029995786026121296, "dbits": 1258291200 }, { "dkld": -0.00043670833110809326, "dbits": 2516582400 }, { "dkld": -0.0004905861802399214, "dbits": 3774873600 }, { "dkld": -0.0005160320550203351, "dbits": 6291456000 } ] }, { "idx": 195, "layers": [ "model.layers.33.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0004228494130075061, "dbits": 62914560 }, { "dkld": -0.00039727417752147276, "dbits": 125829120 }, { "dkld": -0.00016763536259532807, "dbits": 188743680 }, { "dkld": -0.0003268556669354522, "dbits": 314572800 } ] }, { "idx": 196, "layers": [ "model.layers.33.self_attn.k_proj", "model.layers.33.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0004012415185570828, "dbits": 10485760 }, { "dkld": -0.0004238716326653985, "dbits": 20971520 }, { "dkld": -0.0003662116825580708, "dbits": 31457280 }, { "dkld": -0.0004650078713893946, "dbits": 52428800 } ] }, { "idx": 197, "layers": [ "model.layers.33.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0008159829303622246, "dbits": 62914560 }, { "dkld": 0.0003806145861744853, "dbits": 125829120 }, { "dkld": 0.0007402637973427661, "dbits": 188743680 }, { "dkld": 0.0007023323327302877, "dbits": 314572800 } ] }, { "idx": 198, "layers": [ "model.layers.33.mlp.shared_experts.gate_proj", "model.layers.33.mlp.shared_experts.up_proj", "model.layers.33.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.001440229546278718, "dbits": 23592960 }, { "dkld": -0.002001705765724185, "dbits": 47185920 }, { "dkld": -0.0016329297795891762, "dbits": 70778880 }, { "dkld": -0.0013877931982278907, "dbits": 117964800 } ] }, { "idx": 199, "layers": [ "model.layers.33.mlp.experts.0.gate_proj", "model.layers.33.mlp.experts.1.gate_proj", "model.layers.33.mlp.experts.2.gate_proj", "model.layers.33.mlp.experts.3.gate_proj", "model.layers.33.mlp.experts.4.gate_proj", "model.layers.33.mlp.experts.5.gate_proj", "model.layers.33.mlp.experts.6.gate_proj", "model.layers.33.mlp.experts.7.gate_proj", "model.layers.33.mlp.experts.8.gate_proj", "model.layers.33.mlp.experts.9.gate_proj", "model.layers.33.mlp.experts.10.gate_proj", "model.layers.33.mlp.experts.11.gate_proj", "model.layers.33.mlp.experts.12.gate_proj", "model.layers.33.mlp.experts.13.gate_proj", "model.layers.33.mlp.experts.14.gate_proj", "model.layers.33.mlp.experts.15.gate_proj", "model.layers.33.mlp.experts.16.gate_proj", "model.layers.33.mlp.experts.17.gate_proj", "model.layers.33.mlp.experts.18.gate_proj", "model.layers.33.mlp.experts.19.gate_proj", "model.layers.33.mlp.experts.20.gate_proj", "model.layers.33.mlp.experts.21.gate_proj", "model.layers.33.mlp.experts.22.gate_proj", "model.layers.33.mlp.experts.23.gate_proj", "model.layers.33.mlp.experts.24.gate_proj", "model.layers.33.mlp.experts.25.gate_proj", "model.layers.33.mlp.experts.26.gate_proj", "model.layers.33.mlp.experts.27.gate_proj", "model.layers.33.mlp.experts.28.gate_proj", "model.layers.33.mlp.experts.29.gate_proj", "model.layers.33.mlp.experts.30.gate_proj", "model.layers.33.mlp.experts.31.gate_proj", "model.layers.33.mlp.experts.32.gate_proj", "model.layers.33.mlp.experts.33.gate_proj", "model.layers.33.mlp.experts.34.gate_proj", "model.layers.33.mlp.experts.35.gate_proj", "model.layers.33.mlp.experts.36.gate_proj", "model.layers.33.mlp.experts.37.gate_proj", "model.layers.33.mlp.experts.38.gate_proj", "model.layers.33.mlp.experts.39.gate_proj", "model.layers.33.mlp.experts.40.gate_proj", "model.layers.33.mlp.experts.41.gate_proj", "model.layers.33.mlp.experts.42.gate_proj", "model.layers.33.mlp.experts.43.gate_proj", "model.layers.33.mlp.experts.44.gate_proj", "model.layers.33.mlp.experts.45.gate_proj", "model.layers.33.mlp.experts.46.gate_proj", "model.layers.33.mlp.experts.47.gate_proj", "model.layers.33.mlp.experts.48.gate_proj", "model.layers.33.mlp.experts.49.gate_proj", "model.layers.33.mlp.experts.50.gate_proj", "model.layers.33.mlp.experts.51.gate_proj", "model.layers.33.mlp.experts.52.gate_proj", "model.layers.33.mlp.experts.53.gate_proj", "model.layers.33.mlp.experts.54.gate_proj", "model.layers.33.mlp.experts.55.gate_proj", "model.layers.33.mlp.experts.56.gate_proj", "model.layers.33.mlp.experts.57.gate_proj", "model.layers.33.mlp.experts.58.gate_proj", "model.layers.33.mlp.experts.59.gate_proj", "model.layers.33.mlp.experts.60.gate_proj", "model.layers.33.mlp.experts.61.gate_proj", "model.layers.33.mlp.experts.62.gate_proj", "model.layers.33.mlp.experts.63.gate_proj", "model.layers.33.mlp.experts.64.gate_proj", "model.layers.33.mlp.experts.65.gate_proj", "model.layers.33.mlp.experts.66.gate_proj", "model.layers.33.mlp.experts.67.gate_proj", "model.layers.33.mlp.experts.68.gate_proj", "model.layers.33.mlp.experts.69.gate_proj", "model.layers.33.mlp.experts.70.gate_proj", "model.layers.33.mlp.experts.71.gate_proj", "model.layers.33.mlp.experts.72.gate_proj", "model.layers.33.mlp.experts.73.gate_proj", "model.layers.33.mlp.experts.74.gate_proj", "model.layers.33.mlp.experts.75.gate_proj", "model.layers.33.mlp.experts.76.gate_proj", "model.layers.33.mlp.experts.77.gate_proj", "model.layers.33.mlp.experts.78.gate_proj", "model.layers.33.mlp.experts.79.gate_proj", "model.layers.33.mlp.experts.80.gate_proj", "model.layers.33.mlp.experts.81.gate_proj", "model.layers.33.mlp.experts.82.gate_proj", "model.layers.33.mlp.experts.83.gate_proj", "model.layers.33.mlp.experts.84.gate_proj", "model.layers.33.mlp.experts.85.gate_proj", "model.layers.33.mlp.experts.86.gate_proj", "model.layers.33.mlp.experts.87.gate_proj", "model.layers.33.mlp.experts.88.gate_proj", "model.layers.33.mlp.experts.89.gate_proj", "model.layers.33.mlp.experts.90.gate_proj", "model.layers.33.mlp.experts.91.gate_proj", "model.layers.33.mlp.experts.92.gate_proj", "model.layers.33.mlp.experts.93.gate_proj", "model.layers.33.mlp.experts.94.gate_proj", "model.layers.33.mlp.experts.95.gate_proj", "model.layers.33.mlp.experts.96.gate_proj", "model.layers.33.mlp.experts.97.gate_proj", "model.layers.33.mlp.experts.98.gate_proj", "model.layers.33.mlp.experts.99.gate_proj", "model.layers.33.mlp.experts.100.gate_proj", "model.layers.33.mlp.experts.101.gate_proj", "model.layers.33.mlp.experts.102.gate_proj", "model.layers.33.mlp.experts.103.gate_proj", "model.layers.33.mlp.experts.104.gate_proj", "model.layers.33.mlp.experts.105.gate_proj", "model.layers.33.mlp.experts.106.gate_proj", "model.layers.33.mlp.experts.107.gate_proj", "model.layers.33.mlp.experts.108.gate_proj", "model.layers.33.mlp.experts.109.gate_proj", "model.layers.33.mlp.experts.110.gate_proj", "model.layers.33.mlp.experts.111.gate_proj", "model.layers.33.mlp.experts.112.gate_proj", "model.layers.33.mlp.experts.113.gate_proj", "model.layers.33.mlp.experts.114.gate_proj", "model.layers.33.mlp.experts.115.gate_proj", "model.layers.33.mlp.experts.116.gate_proj", "model.layers.33.mlp.experts.117.gate_proj", "model.layers.33.mlp.experts.118.gate_proj", "model.layers.33.mlp.experts.119.gate_proj", "model.layers.33.mlp.experts.120.gate_proj", "model.layers.33.mlp.experts.121.gate_proj", "model.layers.33.mlp.experts.122.gate_proj", "model.layers.33.mlp.experts.123.gate_proj", "model.layers.33.mlp.experts.124.gate_proj", "model.layers.33.mlp.experts.125.gate_proj", "model.layers.33.mlp.experts.126.gate_proj", "model.layers.33.mlp.experts.127.gate_proj", "model.layers.33.mlp.experts.128.gate_proj", "model.layers.33.mlp.experts.129.gate_proj", "model.layers.33.mlp.experts.130.gate_proj", "model.layers.33.mlp.experts.131.gate_proj", "model.layers.33.mlp.experts.132.gate_proj", "model.layers.33.mlp.experts.133.gate_proj", "model.layers.33.mlp.experts.134.gate_proj", "model.layers.33.mlp.experts.135.gate_proj", "model.layers.33.mlp.experts.136.gate_proj", "model.layers.33.mlp.experts.137.gate_proj", "model.layers.33.mlp.experts.138.gate_proj", "model.layers.33.mlp.experts.139.gate_proj", "model.layers.33.mlp.experts.140.gate_proj", "model.layers.33.mlp.experts.141.gate_proj", "model.layers.33.mlp.experts.142.gate_proj", "model.layers.33.mlp.experts.143.gate_proj", "model.layers.33.mlp.experts.144.gate_proj", "model.layers.33.mlp.experts.145.gate_proj", "model.layers.33.mlp.experts.146.gate_proj", "model.layers.33.mlp.experts.147.gate_proj", "model.layers.33.mlp.experts.148.gate_proj", "model.layers.33.mlp.experts.149.gate_proj", "model.layers.33.mlp.experts.150.gate_proj", "model.layers.33.mlp.experts.151.gate_proj", "model.layers.33.mlp.experts.152.gate_proj", "model.layers.33.mlp.experts.153.gate_proj", "model.layers.33.mlp.experts.154.gate_proj", "model.layers.33.mlp.experts.155.gate_proj", "model.layers.33.mlp.experts.156.gate_proj", "model.layers.33.mlp.experts.157.gate_proj", "model.layers.33.mlp.experts.158.gate_proj", "model.layers.33.mlp.experts.159.gate_proj", "model.layers.33.mlp.experts.0.up_proj", "model.layers.33.mlp.experts.1.up_proj", "model.layers.33.mlp.experts.2.up_proj", "model.layers.33.mlp.experts.3.up_proj", "model.layers.33.mlp.experts.4.up_proj", "model.layers.33.mlp.experts.5.up_proj", "model.layers.33.mlp.experts.6.up_proj", "model.layers.33.mlp.experts.7.up_proj", "model.layers.33.mlp.experts.8.up_proj", "model.layers.33.mlp.experts.9.up_proj", "model.layers.33.mlp.experts.10.up_proj", "model.layers.33.mlp.experts.11.up_proj", "model.layers.33.mlp.experts.12.up_proj", "model.layers.33.mlp.experts.13.up_proj", "model.layers.33.mlp.experts.14.up_proj", "model.layers.33.mlp.experts.15.up_proj", "model.layers.33.mlp.experts.16.up_proj", "model.layers.33.mlp.experts.17.up_proj", "model.layers.33.mlp.experts.18.up_proj", "model.layers.33.mlp.experts.19.up_proj", "model.layers.33.mlp.experts.20.up_proj", "model.layers.33.mlp.experts.21.up_proj", "model.layers.33.mlp.experts.22.up_proj", "model.layers.33.mlp.experts.23.up_proj", "model.layers.33.mlp.experts.24.up_proj", "model.layers.33.mlp.experts.25.up_proj", "model.layers.33.mlp.experts.26.up_proj", "model.layers.33.mlp.experts.27.up_proj", "model.layers.33.mlp.experts.28.up_proj", "model.layers.33.mlp.experts.29.up_proj", "model.layers.33.mlp.experts.30.up_proj", "model.layers.33.mlp.experts.31.up_proj", "model.layers.33.mlp.experts.32.up_proj", "model.layers.33.mlp.experts.33.up_proj", "model.layers.33.mlp.experts.34.up_proj", "model.layers.33.mlp.experts.35.up_proj", "model.layers.33.mlp.experts.36.up_proj", "model.layers.33.mlp.experts.37.up_proj", "model.layers.33.mlp.experts.38.up_proj", "model.layers.33.mlp.experts.39.up_proj", "model.layers.33.mlp.experts.40.up_proj", "model.layers.33.mlp.experts.41.up_proj", "model.layers.33.mlp.experts.42.up_proj", "model.layers.33.mlp.experts.43.up_proj", "model.layers.33.mlp.experts.44.up_proj", "model.layers.33.mlp.experts.45.up_proj", "model.layers.33.mlp.experts.46.up_proj", "model.layers.33.mlp.experts.47.up_proj", "model.layers.33.mlp.experts.48.up_proj", "model.layers.33.mlp.experts.49.up_proj", "model.layers.33.mlp.experts.50.up_proj", "model.layers.33.mlp.experts.51.up_proj", "model.layers.33.mlp.experts.52.up_proj", "model.layers.33.mlp.experts.53.up_proj", "model.layers.33.mlp.experts.54.up_proj", "model.layers.33.mlp.experts.55.up_proj", "model.layers.33.mlp.experts.56.up_proj", "model.layers.33.mlp.experts.57.up_proj", "model.layers.33.mlp.experts.58.up_proj", "model.layers.33.mlp.experts.59.up_proj", "model.layers.33.mlp.experts.60.up_proj", "model.layers.33.mlp.experts.61.up_proj", "model.layers.33.mlp.experts.62.up_proj", "model.layers.33.mlp.experts.63.up_proj", "model.layers.33.mlp.experts.64.up_proj", "model.layers.33.mlp.experts.65.up_proj", "model.layers.33.mlp.experts.66.up_proj", "model.layers.33.mlp.experts.67.up_proj", "model.layers.33.mlp.experts.68.up_proj", "model.layers.33.mlp.experts.69.up_proj", "model.layers.33.mlp.experts.70.up_proj", "model.layers.33.mlp.experts.71.up_proj", "model.layers.33.mlp.experts.72.up_proj", "model.layers.33.mlp.experts.73.up_proj", "model.layers.33.mlp.experts.74.up_proj", "model.layers.33.mlp.experts.75.up_proj", "model.layers.33.mlp.experts.76.up_proj", "model.layers.33.mlp.experts.77.up_proj", "model.layers.33.mlp.experts.78.up_proj", "model.layers.33.mlp.experts.79.up_proj", "model.layers.33.mlp.experts.80.up_proj", "model.layers.33.mlp.experts.81.up_proj", "model.layers.33.mlp.experts.82.up_proj", "model.layers.33.mlp.experts.83.up_proj", "model.layers.33.mlp.experts.84.up_proj", "model.layers.33.mlp.experts.85.up_proj", "model.layers.33.mlp.experts.86.up_proj", "model.layers.33.mlp.experts.87.up_proj", "model.layers.33.mlp.experts.88.up_proj", "model.layers.33.mlp.experts.89.up_proj", "model.layers.33.mlp.experts.90.up_proj", "model.layers.33.mlp.experts.91.up_proj", "model.layers.33.mlp.experts.92.up_proj", "model.layers.33.mlp.experts.93.up_proj", "model.layers.33.mlp.experts.94.up_proj", "model.layers.33.mlp.experts.95.up_proj", "model.layers.33.mlp.experts.96.up_proj", "model.layers.33.mlp.experts.97.up_proj", "model.layers.33.mlp.experts.98.up_proj", "model.layers.33.mlp.experts.99.up_proj", "model.layers.33.mlp.experts.100.up_proj", "model.layers.33.mlp.experts.101.up_proj", "model.layers.33.mlp.experts.102.up_proj", "model.layers.33.mlp.experts.103.up_proj", "model.layers.33.mlp.experts.104.up_proj", "model.layers.33.mlp.experts.105.up_proj", "model.layers.33.mlp.experts.106.up_proj", "model.layers.33.mlp.experts.107.up_proj", "model.layers.33.mlp.experts.108.up_proj", "model.layers.33.mlp.experts.109.up_proj", "model.layers.33.mlp.experts.110.up_proj", "model.layers.33.mlp.experts.111.up_proj", "model.layers.33.mlp.experts.112.up_proj", "model.layers.33.mlp.experts.113.up_proj", "model.layers.33.mlp.experts.114.up_proj", "model.layers.33.mlp.experts.115.up_proj", "model.layers.33.mlp.experts.116.up_proj", "model.layers.33.mlp.experts.117.up_proj", "model.layers.33.mlp.experts.118.up_proj", "model.layers.33.mlp.experts.119.up_proj", "model.layers.33.mlp.experts.120.up_proj", "model.layers.33.mlp.experts.121.up_proj", "model.layers.33.mlp.experts.122.up_proj", "model.layers.33.mlp.experts.123.up_proj", "model.layers.33.mlp.experts.124.up_proj", "model.layers.33.mlp.experts.125.up_proj", "model.layers.33.mlp.experts.126.up_proj", "model.layers.33.mlp.experts.127.up_proj", "model.layers.33.mlp.experts.128.up_proj", "model.layers.33.mlp.experts.129.up_proj", "model.layers.33.mlp.experts.130.up_proj", "model.layers.33.mlp.experts.131.up_proj", "model.layers.33.mlp.experts.132.up_proj", "model.layers.33.mlp.experts.133.up_proj", "model.layers.33.mlp.experts.134.up_proj", "model.layers.33.mlp.experts.135.up_proj", "model.layers.33.mlp.experts.136.up_proj", "model.layers.33.mlp.experts.137.up_proj", "model.layers.33.mlp.experts.138.up_proj", "model.layers.33.mlp.experts.139.up_proj", "model.layers.33.mlp.experts.140.up_proj", "model.layers.33.mlp.experts.141.up_proj", "model.layers.33.mlp.experts.142.up_proj", "model.layers.33.mlp.experts.143.up_proj", "model.layers.33.mlp.experts.144.up_proj", "model.layers.33.mlp.experts.145.up_proj", "model.layers.33.mlp.experts.146.up_proj", "model.layers.33.mlp.experts.147.up_proj", "model.layers.33.mlp.experts.148.up_proj", "model.layers.33.mlp.experts.149.up_proj", "model.layers.33.mlp.experts.150.up_proj", "model.layers.33.mlp.experts.151.up_proj", "model.layers.33.mlp.experts.152.up_proj", "model.layers.33.mlp.experts.153.up_proj", "model.layers.33.mlp.experts.154.up_proj", "model.layers.33.mlp.experts.155.up_proj", "model.layers.33.mlp.experts.156.up_proj", "model.layers.33.mlp.experts.157.up_proj", "model.layers.33.mlp.experts.158.up_proj", "model.layers.33.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 0.0002651311457157052, "dbits": 2516582400 }, { "dkld": 0.0001340130344033158, "dbits": 5033164800 }, { "dkld": -2.875998616219677e-05, "dbits": 7549747200 }, { "dkld": 8.525997400282703e-05, "dbits": 12582912000 } ] }, { "idx": 200, "layers": [ "model.layers.33.mlp.experts.0.down_proj", "model.layers.33.mlp.experts.1.down_proj", "model.layers.33.mlp.experts.2.down_proj", "model.layers.33.mlp.experts.3.down_proj", "model.layers.33.mlp.experts.4.down_proj", "model.layers.33.mlp.experts.5.down_proj", "model.layers.33.mlp.experts.6.down_proj", "model.layers.33.mlp.experts.7.down_proj", "model.layers.33.mlp.experts.8.down_proj", "model.layers.33.mlp.experts.9.down_proj", "model.layers.33.mlp.experts.10.down_proj", "model.layers.33.mlp.experts.11.down_proj", "model.layers.33.mlp.experts.12.down_proj", "model.layers.33.mlp.experts.13.down_proj", "model.layers.33.mlp.experts.14.down_proj", "model.layers.33.mlp.experts.15.down_proj", "model.layers.33.mlp.experts.16.down_proj", "model.layers.33.mlp.experts.17.down_proj", "model.layers.33.mlp.experts.18.down_proj", "model.layers.33.mlp.experts.19.down_proj", "model.layers.33.mlp.experts.20.down_proj", "model.layers.33.mlp.experts.21.down_proj", "model.layers.33.mlp.experts.22.down_proj", "model.layers.33.mlp.experts.23.down_proj", "model.layers.33.mlp.experts.24.down_proj", "model.layers.33.mlp.experts.25.down_proj", "model.layers.33.mlp.experts.26.down_proj", "model.layers.33.mlp.experts.27.down_proj", "model.layers.33.mlp.experts.28.down_proj", "model.layers.33.mlp.experts.29.down_proj", "model.layers.33.mlp.experts.30.down_proj", "model.layers.33.mlp.experts.31.down_proj", "model.layers.33.mlp.experts.32.down_proj", "model.layers.33.mlp.experts.33.down_proj", "model.layers.33.mlp.experts.34.down_proj", "model.layers.33.mlp.experts.35.down_proj", "model.layers.33.mlp.experts.36.down_proj", "model.layers.33.mlp.experts.37.down_proj", "model.layers.33.mlp.experts.38.down_proj", "model.layers.33.mlp.experts.39.down_proj", "model.layers.33.mlp.experts.40.down_proj", "model.layers.33.mlp.experts.41.down_proj", "model.layers.33.mlp.experts.42.down_proj", "model.layers.33.mlp.experts.43.down_proj", "model.layers.33.mlp.experts.44.down_proj", "model.layers.33.mlp.experts.45.down_proj", "model.layers.33.mlp.experts.46.down_proj", "model.layers.33.mlp.experts.47.down_proj", "model.layers.33.mlp.experts.48.down_proj", "model.layers.33.mlp.experts.49.down_proj", "model.layers.33.mlp.experts.50.down_proj", "model.layers.33.mlp.experts.51.down_proj", "model.layers.33.mlp.experts.52.down_proj", "model.layers.33.mlp.experts.53.down_proj", "model.layers.33.mlp.experts.54.down_proj", "model.layers.33.mlp.experts.55.down_proj", "model.layers.33.mlp.experts.56.down_proj", "model.layers.33.mlp.experts.57.down_proj", "model.layers.33.mlp.experts.58.down_proj", "model.layers.33.mlp.experts.59.down_proj", "model.layers.33.mlp.experts.60.down_proj", "model.layers.33.mlp.experts.61.down_proj", "model.layers.33.mlp.experts.62.down_proj", "model.layers.33.mlp.experts.63.down_proj", "model.layers.33.mlp.experts.64.down_proj", "model.layers.33.mlp.experts.65.down_proj", "model.layers.33.mlp.experts.66.down_proj", "model.layers.33.mlp.experts.67.down_proj", "model.layers.33.mlp.experts.68.down_proj", "model.layers.33.mlp.experts.69.down_proj", "model.layers.33.mlp.experts.70.down_proj", "model.layers.33.mlp.experts.71.down_proj", "model.layers.33.mlp.experts.72.down_proj", "model.layers.33.mlp.experts.73.down_proj", "model.layers.33.mlp.experts.74.down_proj", "model.layers.33.mlp.experts.75.down_proj", "model.layers.33.mlp.experts.76.down_proj", "model.layers.33.mlp.experts.77.down_proj", "model.layers.33.mlp.experts.78.down_proj", "model.layers.33.mlp.experts.79.down_proj", "model.layers.33.mlp.experts.80.down_proj", "model.layers.33.mlp.experts.81.down_proj", "model.layers.33.mlp.experts.82.down_proj", "model.layers.33.mlp.experts.83.down_proj", "model.layers.33.mlp.experts.84.down_proj", "model.layers.33.mlp.experts.85.down_proj", "model.layers.33.mlp.experts.86.down_proj", "model.layers.33.mlp.experts.87.down_proj", "model.layers.33.mlp.experts.88.down_proj", "model.layers.33.mlp.experts.89.down_proj", "model.layers.33.mlp.experts.90.down_proj", "model.layers.33.mlp.experts.91.down_proj", "model.layers.33.mlp.experts.92.down_proj", "model.layers.33.mlp.experts.93.down_proj", "model.layers.33.mlp.experts.94.down_proj", "model.layers.33.mlp.experts.95.down_proj", "model.layers.33.mlp.experts.96.down_proj", "model.layers.33.mlp.experts.97.down_proj", "model.layers.33.mlp.experts.98.down_proj", "model.layers.33.mlp.experts.99.down_proj", "model.layers.33.mlp.experts.100.down_proj", "model.layers.33.mlp.experts.101.down_proj", "model.layers.33.mlp.experts.102.down_proj", "model.layers.33.mlp.experts.103.down_proj", "model.layers.33.mlp.experts.104.down_proj", "model.layers.33.mlp.experts.105.down_proj", "model.layers.33.mlp.experts.106.down_proj", "model.layers.33.mlp.experts.107.down_proj", "model.layers.33.mlp.experts.108.down_proj", "model.layers.33.mlp.experts.109.down_proj", "model.layers.33.mlp.experts.110.down_proj", "model.layers.33.mlp.experts.111.down_proj", "model.layers.33.mlp.experts.112.down_proj", "model.layers.33.mlp.experts.113.down_proj", "model.layers.33.mlp.experts.114.down_proj", "model.layers.33.mlp.experts.115.down_proj", "model.layers.33.mlp.experts.116.down_proj", "model.layers.33.mlp.experts.117.down_proj", "model.layers.33.mlp.experts.118.down_proj", "model.layers.33.mlp.experts.119.down_proj", "model.layers.33.mlp.experts.120.down_proj", "model.layers.33.mlp.experts.121.down_proj", "model.layers.33.mlp.experts.122.down_proj", "model.layers.33.mlp.experts.123.down_proj", "model.layers.33.mlp.experts.124.down_proj", "model.layers.33.mlp.experts.125.down_proj", "model.layers.33.mlp.experts.126.down_proj", "model.layers.33.mlp.experts.127.down_proj", "model.layers.33.mlp.experts.128.down_proj", "model.layers.33.mlp.experts.129.down_proj", "model.layers.33.mlp.experts.130.down_proj", "model.layers.33.mlp.experts.131.down_proj", "model.layers.33.mlp.experts.132.down_proj", "model.layers.33.mlp.experts.133.down_proj", "model.layers.33.mlp.experts.134.down_proj", "model.layers.33.mlp.experts.135.down_proj", "model.layers.33.mlp.experts.136.down_proj", "model.layers.33.mlp.experts.137.down_proj", "model.layers.33.mlp.experts.138.down_proj", "model.layers.33.mlp.experts.139.down_proj", "model.layers.33.mlp.experts.140.down_proj", "model.layers.33.mlp.experts.141.down_proj", "model.layers.33.mlp.experts.142.down_proj", "model.layers.33.mlp.experts.143.down_proj", "model.layers.33.mlp.experts.144.down_proj", "model.layers.33.mlp.experts.145.down_proj", "model.layers.33.mlp.experts.146.down_proj", "model.layers.33.mlp.experts.147.down_proj", "model.layers.33.mlp.experts.148.down_proj", "model.layers.33.mlp.experts.149.down_proj", "model.layers.33.mlp.experts.150.down_proj", "model.layers.33.mlp.experts.151.down_proj", "model.layers.33.mlp.experts.152.down_proj", "model.layers.33.mlp.experts.153.down_proj", "model.layers.33.mlp.experts.154.down_proj", "model.layers.33.mlp.experts.155.down_proj", "model.layers.33.mlp.experts.156.down_proj", "model.layers.33.mlp.experts.157.down_proj", "model.layers.33.mlp.experts.158.down_proj", "model.layers.33.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -8.167251944542209e-05, "dbits": 1258291200 }, { "dkld": -0.0002475157380104065, "dbits": 2516582400 }, { "dkld": -0.00032366849482060034, "dbits": 3774873600 }, { "dkld": -0.00029706722125411034, "dbits": 6291456000 } ] }, { "idx": 201, "layers": [ "model.layers.34.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00032833088189362925, "dbits": 62914560 }, { "dkld": 4.9786642193766495e-06, "dbits": 125829120 }, { "dkld": -0.0001090809702873341, "dbits": 188743680 }, { "dkld": -4.500988870859146e-05, "dbits": 314572800 } ] }, { "idx": 202, "layers": [ "model.layers.34.self_attn.k_proj", "model.layers.34.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00021633952856063565, "dbits": 10485760 }, { "dkld": -0.00010542785748840489, "dbits": 20971520 }, { "dkld": -5.7884305715572015e-05, "dbits": 31457280 }, { "dkld": -0.00014040041714907803, "dbits": 52428800 } ] }, { "idx": 203, "layers": [ "model.layers.34.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0008010376244783485, "dbits": 62914560 }, { "dkld": -0.0006421463564038332, "dbits": 125829120 }, { "dkld": -0.0004405749030411271, "dbits": 188743680 }, { "dkld": -0.0005903886631131172, "dbits": 314572800 } ] }, { "idx": 204, "layers": [ "model.layers.34.mlp.shared_experts.gate_proj", "model.layers.34.mlp.shared_experts.up_proj", "model.layers.34.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0024050856940448367, "dbits": 23592960 }, { "dkld": -0.0028472309932112694, "dbits": 47185920 }, { "dkld": -0.0029979782178998077, "dbits": 70778880 }, { "dkld": -0.0030028412118554143, "dbits": 117964800 } ] }, { "idx": 205, "layers": [ "model.layers.34.mlp.experts.0.gate_proj", "model.layers.34.mlp.experts.1.gate_proj", "model.layers.34.mlp.experts.2.gate_proj", "model.layers.34.mlp.experts.3.gate_proj", "model.layers.34.mlp.experts.4.gate_proj", "model.layers.34.mlp.experts.5.gate_proj", "model.layers.34.mlp.experts.6.gate_proj", "model.layers.34.mlp.experts.7.gate_proj", "model.layers.34.mlp.experts.8.gate_proj", "model.layers.34.mlp.experts.9.gate_proj", "model.layers.34.mlp.experts.10.gate_proj", "model.layers.34.mlp.experts.11.gate_proj", "model.layers.34.mlp.experts.12.gate_proj", "model.layers.34.mlp.experts.13.gate_proj", "model.layers.34.mlp.experts.14.gate_proj", "model.layers.34.mlp.experts.15.gate_proj", "model.layers.34.mlp.experts.16.gate_proj", "model.layers.34.mlp.experts.17.gate_proj", "model.layers.34.mlp.experts.18.gate_proj", "model.layers.34.mlp.experts.19.gate_proj", "model.layers.34.mlp.experts.20.gate_proj", "model.layers.34.mlp.experts.21.gate_proj", "model.layers.34.mlp.experts.22.gate_proj", "model.layers.34.mlp.experts.23.gate_proj", "model.layers.34.mlp.experts.24.gate_proj", "model.layers.34.mlp.experts.25.gate_proj", "model.layers.34.mlp.experts.26.gate_proj", "model.layers.34.mlp.experts.27.gate_proj", "model.layers.34.mlp.experts.28.gate_proj", "model.layers.34.mlp.experts.29.gate_proj", "model.layers.34.mlp.experts.30.gate_proj", "model.layers.34.mlp.experts.31.gate_proj", "model.layers.34.mlp.experts.32.gate_proj", "model.layers.34.mlp.experts.33.gate_proj", "model.layers.34.mlp.experts.34.gate_proj", "model.layers.34.mlp.experts.35.gate_proj", "model.layers.34.mlp.experts.36.gate_proj", "model.layers.34.mlp.experts.37.gate_proj", "model.layers.34.mlp.experts.38.gate_proj", "model.layers.34.mlp.experts.39.gate_proj", "model.layers.34.mlp.experts.40.gate_proj", "model.layers.34.mlp.experts.41.gate_proj", "model.layers.34.mlp.experts.42.gate_proj", "model.layers.34.mlp.experts.43.gate_proj", "model.layers.34.mlp.experts.44.gate_proj", "model.layers.34.mlp.experts.45.gate_proj", "model.layers.34.mlp.experts.46.gate_proj", "model.layers.34.mlp.experts.47.gate_proj", "model.layers.34.mlp.experts.48.gate_proj", "model.layers.34.mlp.experts.49.gate_proj", "model.layers.34.mlp.experts.50.gate_proj", "model.layers.34.mlp.experts.51.gate_proj", "model.layers.34.mlp.experts.52.gate_proj", "model.layers.34.mlp.experts.53.gate_proj", "model.layers.34.mlp.experts.54.gate_proj", "model.layers.34.mlp.experts.55.gate_proj", "model.layers.34.mlp.experts.56.gate_proj", "model.layers.34.mlp.experts.57.gate_proj", "model.layers.34.mlp.experts.58.gate_proj", "model.layers.34.mlp.experts.59.gate_proj", "model.layers.34.mlp.experts.60.gate_proj", "model.layers.34.mlp.experts.61.gate_proj", "model.layers.34.mlp.experts.62.gate_proj", "model.layers.34.mlp.experts.63.gate_proj", "model.layers.34.mlp.experts.64.gate_proj", "model.layers.34.mlp.experts.65.gate_proj", "model.layers.34.mlp.experts.66.gate_proj", "model.layers.34.mlp.experts.67.gate_proj", "model.layers.34.mlp.experts.68.gate_proj", "model.layers.34.mlp.experts.69.gate_proj", "model.layers.34.mlp.experts.70.gate_proj", "model.layers.34.mlp.experts.71.gate_proj", "model.layers.34.mlp.experts.72.gate_proj", "model.layers.34.mlp.experts.73.gate_proj", "model.layers.34.mlp.experts.74.gate_proj", "model.layers.34.mlp.experts.75.gate_proj", "model.layers.34.mlp.experts.76.gate_proj", "model.layers.34.mlp.experts.77.gate_proj", "model.layers.34.mlp.experts.78.gate_proj", "model.layers.34.mlp.experts.79.gate_proj", "model.layers.34.mlp.experts.80.gate_proj", "model.layers.34.mlp.experts.81.gate_proj", "model.layers.34.mlp.experts.82.gate_proj", "model.layers.34.mlp.experts.83.gate_proj", "model.layers.34.mlp.experts.84.gate_proj", "model.layers.34.mlp.experts.85.gate_proj", "model.layers.34.mlp.experts.86.gate_proj", "model.layers.34.mlp.experts.87.gate_proj", "model.layers.34.mlp.experts.88.gate_proj", "model.layers.34.mlp.experts.89.gate_proj", "model.layers.34.mlp.experts.90.gate_proj", "model.layers.34.mlp.experts.91.gate_proj", "model.layers.34.mlp.experts.92.gate_proj", "model.layers.34.mlp.experts.93.gate_proj", "model.layers.34.mlp.experts.94.gate_proj", "model.layers.34.mlp.experts.95.gate_proj", "model.layers.34.mlp.experts.96.gate_proj", "model.layers.34.mlp.experts.97.gate_proj", "model.layers.34.mlp.experts.98.gate_proj", "model.layers.34.mlp.experts.99.gate_proj", "model.layers.34.mlp.experts.100.gate_proj", "model.layers.34.mlp.experts.101.gate_proj", "model.layers.34.mlp.experts.102.gate_proj", "model.layers.34.mlp.experts.103.gate_proj", "model.layers.34.mlp.experts.104.gate_proj", "model.layers.34.mlp.experts.105.gate_proj", "model.layers.34.mlp.experts.106.gate_proj", "model.layers.34.mlp.experts.107.gate_proj", "model.layers.34.mlp.experts.108.gate_proj", "model.layers.34.mlp.experts.109.gate_proj", "model.layers.34.mlp.experts.110.gate_proj", "model.layers.34.mlp.experts.111.gate_proj", "model.layers.34.mlp.experts.112.gate_proj", "model.layers.34.mlp.experts.113.gate_proj", "model.layers.34.mlp.experts.114.gate_proj", "model.layers.34.mlp.experts.115.gate_proj", "model.layers.34.mlp.experts.116.gate_proj", "model.layers.34.mlp.experts.117.gate_proj", "model.layers.34.mlp.experts.118.gate_proj", "model.layers.34.mlp.experts.119.gate_proj", "model.layers.34.mlp.experts.120.gate_proj", "model.layers.34.mlp.experts.121.gate_proj", "model.layers.34.mlp.experts.122.gate_proj", "model.layers.34.mlp.experts.123.gate_proj", "model.layers.34.mlp.experts.124.gate_proj", "model.layers.34.mlp.experts.125.gate_proj", "model.layers.34.mlp.experts.126.gate_proj", "model.layers.34.mlp.experts.127.gate_proj", "model.layers.34.mlp.experts.128.gate_proj", "model.layers.34.mlp.experts.129.gate_proj", "model.layers.34.mlp.experts.130.gate_proj", "model.layers.34.mlp.experts.131.gate_proj", "model.layers.34.mlp.experts.132.gate_proj", "model.layers.34.mlp.experts.133.gate_proj", "model.layers.34.mlp.experts.134.gate_proj", "model.layers.34.mlp.experts.135.gate_proj", "model.layers.34.mlp.experts.136.gate_proj", "model.layers.34.mlp.experts.137.gate_proj", "model.layers.34.mlp.experts.138.gate_proj", "model.layers.34.mlp.experts.139.gate_proj", "model.layers.34.mlp.experts.140.gate_proj", "model.layers.34.mlp.experts.141.gate_proj", "model.layers.34.mlp.experts.142.gate_proj", "model.layers.34.mlp.experts.143.gate_proj", "model.layers.34.mlp.experts.144.gate_proj", "model.layers.34.mlp.experts.145.gate_proj", "model.layers.34.mlp.experts.146.gate_proj", "model.layers.34.mlp.experts.147.gate_proj", "model.layers.34.mlp.experts.148.gate_proj", "model.layers.34.mlp.experts.149.gate_proj", "model.layers.34.mlp.experts.150.gate_proj", "model.layers.34.mlp.experts.151.gate_proj", "model.layers.34.mlp.experts.152.gate_proj", "model.layers.34.mlp.experts.153.gate_proj", "model.layers.34.mlp.experts.154.gate_proj", "model.layers.34.mlp.experts.155.gate_proj", "model.layers.34.mlp.experts.156.gate_proj", "model.layers.34.mlp.experts.157.gate_proj", "model.layers.34.mlp.experts.158.gate_proj", "model.layers.34.mlp.experts.159.gate_proj", "model.layers.34.mlp.experts.0.up_proj", "model.layers.34.mlp.experts.1.up_proj", "model.layers.34.mlp.experts.2.up_proj", "model.layers.34.mlp.experts.3.up_proj", "model.layers.34.mlp.experts.4.up_proj", "model.layers.34.mlp.experts.5.up_proj", "model.layers.34.mlp.experts.6.up_proj", "model.layers.34.mlp.experts.7.up_proj", "model.layers.34.mlp.experts.8.up_proj", "model.layers.34.mlp.experts.9.up_proj", "model.layers.34.mlp.experts.10.up_proj", "model.layers.34.mlp.experts.11.up_proj", "model.layers.34.mlp.experts.12.up_proj", "model.layers.34.mlp.experts.13.up_proj", "model.layers.34.mlp.experts.14.up_proj", "model.layers.34.mlp.experts.15.up_proj", "model.layers.34.mlp.experts.16.up_proj", "model.layers.34.mlp.experts.17.up_proj", "model.layers.34.mlp.experts.18.up_proj", "model.layers.34.mlp.experts.19.up_proj", "model.layers.34.mlp.experts.20.up_proj", "model.layers.34.mlp.experts.21.up_proj", "model.layers.34.mlp.experts.22.up_proj", "model.layers.34.mlp.experts.23.up_proj", "model.layers.34.mlp.experts.24.up_proj", "model.layers.34.mlp.experts.25.up_proj", "model.layers.34.mlp.experts.26.up_proj", "model.layers.34.mlp.experts.27.up_proj", "model.layers.34.mlp.experts.28.up_proj", "model.layers.34.mlp.experts.29.up_proj", "model.layers.34.mlp.experts.30.up_proj", "model.layers.34.mlp.experts.31.up_proj", "model.layers.34.mlp.experts.32.up_proj", "model.layers.34.mlp.experts.33.up_proj", "model.layers.34.mlp.experts.34.up_proj", "model.layers.34.mlp.experts.35.up_proj", "model.layers.34.mlp.experts.36.up_proj", "model.layers.34.mlp.experts.37.up_proj", "model.layers.34.mlp.experts.38.up_proj", "model.layers.34.mlp.experts.39.up_proj", "model.layers.34.mlp.experts.40.up_proj", "model.layers.34.mlp.experts.41.up_proj", "model.layers.34.mlp.experts.42.up_proj", "model.layers.34.mlp.experts.43.up_proj", "model.layers.34.mlp.experts.44.up_proj", "model.layers.34.mlp.experts.45.up_proj", "model.layers.34.mlp.experts.46.up_proj", "model.layers.34.mlp.experts.47.up_proj", "model.layers.34.mlp.experts.48.up_proj", "model.layers.34.mlp.experts.49.up_proj", "model.layers.34.mlp.experts.50.up_proj", "model.layers.34.mlp.experts.51.up_proj", "model.layers.34.mlp.experts.52.up_proj", "model.layers.34.mlp.experts.53.up_proj", "model.layers.34.mlp.experts.54.up_proj", "model.layers.34.mlp.experts.55.up_proj", "model.layers.34.mlp.experts.56.up_proj", "model.layers.34.mlp.experts.57.up_proj", "model.layers.34.mlp.experts.58.up_proj", "model.layers.34.mlp.experts.59.up_proj", "model.layers.34.mlp.experts.60.up_proj", "model.layers.34.mlp.experts.61.up_proj", "model.layers.34.mlp.experts.62.up_proj", "model.layers.34.mlp.experts.63.up_proj", "model.layers.34.mlp.experts.64.up_proj", "model.layers.34.mlp.experts.65.up_proj", "model.layers.34.mlp.experts.66.up_proj", "model.layers.34.mlp.experts.67.up_proj", "model.layers.34.mlp.experts.68.up_proj", "model.layers.34.mlp.experts.69.up_proj", "model.layers.34.mlp.experts.70.up_proj", "model.layers.34.mlp.experts.71.up_proj", "model.layers.34.mlp.experts.72.up_proj", "model.layers.34.mlp.experts.73.up_proj", "model.layers.34.mlp.experts.74.up_proj", "model.layers.34.mlp.experts.75.up_proj", "model.layers.34.mlp.experts.76.up_proj", "model.layers.34.mlp.experts.77.up_proj", "model.layers.34.mlp.experts.78.up_proj", "model.layers.34.mlp.experts.79.up_proj", "model.layers.34.mlp.experts.80.up_proj", "model.layers.34.mlp.experts.81.up_proj", "model.layers.34.mlp.experts.82.up_proj", "model.layers.34.mlp.experts.83.up_proj", "model.layers.34.mlp.experts.84.up_proj", "model.layers.34.mlp.experts.85.up_proj", "model.layers.34.mlp.experts.86.up_proj", "model.layers.34.mlp.experts.87.up_proj", "model.layers.34.mlp.experts.88.up_proj", "model.layers.34.mlp.experts.89.up_proj", "model.layers.34.mlp.experts.90.up_proj", "model.layers.34.mlp.experts.91.up_proj", "model.layers.34.mlp.experts.92.up_proj", "model.layers.34.mlp.experts.93.up_proj", "model.layers.34.mlp.experts.94.up_proj", "model.layers.34.mlp.experts.95.up_proj", "model.layers.34.mlp.experts.96.up_proj", "model.layers.34.mlp.experts.97.up_proj", "model.layers.34.mlp.experts.98.up_proj", "model.layers.34.mlp.experts.99.up_proj", "model.layers.34.mlp.experts.100.up_proj", "model.layers.34.mlp.experts.101.up_proj", "model.layers.34.mlp.experts.102.up_proj", "model.layers.34.mlp.experts.103.up_proj", "model.layers.34.mlp.experts.104.up_proj", "model.layers.34.mlp.experts.105.up_proj", "model.layers.34.mlp.experts.106.up_proj", "model.layers.34.mlp.experts.107.up_proj", "model.layers.34.mlp.experts.108.up_proj", "model.layers.34.mlp.experts.109.up_proj", "model.layers.34.mlp.experts.110.up_proj", "model.layers.34.mlp.experts.111.up_proj", "model.layers.34.mlp.experts.112.up_proj", "model.layers.34.mlp.experts.113.up_proj", "model.layers.34.mlp.experts.114.up_proj", "model.layers.34.mlp.experts.115.up_proj", "model.layers.34.mlp.experts.116.up_proj", "model.layers.34.mlp.experts.117.up_proj", "model.layers.34.mlp.experts.118.up_proj", "model.layers.34.mlp.experts.119.up_proj", "model.layers.34.mlp.experts.120.up_proj", "model.layers.34.mlp.experts.121.up_proj", "model.layers.34.mlp.experts.122.up_proj", "model.layers.34.mlp.experts.123.up_proj", "model.layers.34.mlp.experts.124.up_proj", "model.layers.34.mlp.experts.125.up_proj", "model.layers.34.mlp.experts.126.up_proj", "model.layers.34.mlp.experts.127.up_proj", "model.layers.34.mlp.experts.128.up_proj", "model.layers.34.mlp.experts.129.up_proj", "model.layers.34.mlp.experts.130.up_proj", "model.layers.34.mlp.experts.131.up_proj", "model.layers.34.mlp.experts.132.up_proj", "model.layers.34.mlp.experts.133.up_proj", "model.layers.34.mlp.experts.134.up_proj", "model.layers.34.mlp.experts.135.up_proj", "model.layers.34.mlp.experts.136.up_proj", "model.layers.34.mlp.experts.137.up_proj", "model.layers.34.mlp.experts.138.up_proj", "model.layers.34.mlp.experts.139.up_proj", "model.layers.34.mlp.experts.140.up_proj", "model.layers.34.mlp.experts.141.up_proj", "model.layers.34.mlp.experts.142.up_proj", "model.layers.34.mlp.experts.143.up_proj", "model.layers.34.mlp.experts.144.up_proj", "model.layers.34.mlp.experts.145.up_proj", "model.layers.34.mlp.experts.146.up_proj", "model.layers.34.mlp.experts.147.up_proj", "model.layers.34.mlp.experts.148.up_proj", "model.layers.34.mlp.experts.149.up_proj", "model.layers.34.mlp.experts.150.up_proj", "model.layers.34.mlp.experts.151.up_proj", "model.layers.34.mlp.experts.152.up_proj", "model.layers.34.mlp.experts.153.up_proj", "model.layers.34.mlp.experts.154.up_proj", "model.layers.34.mlp.experts.155.up_proj", "model.layers.34.mlp.experts.156.up_proj", "model.layers.34.mlp.experts.157.up_proj", "model.layers.34.mlp.experts.158.up_proj", "model.layers.34.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0008448420092463521, "dbits": 2516582400 }, { "dkld": -0.0010472472757101087, "dbits": 5033164800 }, { "dkld": -0.0008958917111158426, "dbits": 7549747200 }, { "dkld": -0.0009859177283942783, "dbits": 12582912000 } ] }, { "idx": 206, "layers": [ "model.layers.34.mlp.experts.0.down_proj", "model.layers.34.mlp.experts.1.down_proj", "model.layers.34.mlp.experts.2.down_proj", "model.layers.34.mlp.experts.3.down_proj", "model.layers.34.mlp.experts.4.down_proj", "model.layers.34.mlp.experts.5.down_proj", "model.layers.34.mlp.experts.6.down_proj", "model.layers.34.mlp.experts.7.down_proj", "model.layers.34.mlp.experts.8.down_proj", "model.layers.34.mlp.experts.9.down_proj", "model.layers.34.mlp.experts.10.down_proj", "model.layers.34.mlp.experts.11.down_proj", "model.layers.34.mlp.experts.12.down_proj", "model.layers.34.mlp.experts.13.down_proj", "model.layers.34.mlp.experts.14.down_proj", "model.layers.34.mlp.experts.15.down_proj", "model.layers.34.mlp.experts.16.down_proj", "model.layers.34.mlp.experts.17.down_proj", "model.layers.34.mlp.experts.18.down_proj", "model.layers.34.mlp.experts.19.down_proj", "model.layers.34.mlp.experts.20.down_proj", "model.layers.34.mlp.experts.21.down_proj", "model.layers.34.mlp.experts.22.down_proj", "model.layers.34.mlp.experts.23.down_proj", "model.layers.34.mlp.experts.24.down_proj", "model.layers.34.mlp.experts.25.down_proj", "model.layers.34.mlp.experts.26.down_proj", "model.layers.34.mlp.experts.27.down_proj", "model.layers.34.mlp.experts.28.down_proj", "model.layers.34.mlp.experts.29.down_proj", "model.layers.34.mlp.experts.30.down_proj", "model.layers.34.mlp.experts.31.down_proj", "model.layers.34.mlp.experts.32.down_proj", "model.layers.34.mlp.experts.33.down_proj", "model.layers.34.mlp.experts.34.down_proj", "model.layers.34.mlp.experts.35.down_proj", "model.layers.34.mlp.experts.36.down_proj", "model.layers.34.mlp.experts.37.down_proj", "model.layers.34.mlp.experts.38.down_proj", "model.layers.34.mlp.experts.39.down_proj", "model.layers.34.mlp.experts.40.down_proj", "model.layers.34.mlp.experts.41.down_proj", "model.layers.34.mlp.experts.42.down_proj", "model.layers.34.mlp.experts.43.down_proj", "model.layers.34.mlp.experts.44.down_proj", "model.layers.34.mlp.experts.45.down_proj", "model.layers.34.mlp.experts.46.down_proj", "model.layers.34.mlp.experts.47.down_proj", "model.layers.34.mlp.experts.48.down_proj", "model.layers.34.mlp.experts.49.down_proj", "model.layers.34.mlp.experts.50.down_proj", "model.layers.34.mlp.experts.51.down_proj", "model.layers.34.mlp.experts.52.down_proj", "model.layers.34.mlp.experts.53.down_proj", "model.layers.34.mlp.experts.54.down_proj", "model.layers.34.mlp.experts.55.down_proj", "model.layers.34.mlp.experts.56.down_proj", "model.layers.34.mlp.experts.57.down_proj", "model.layers.34.mlp.experts.58.down_proj", "model.layers.34.mlp.experts.59.down_proj", "model.layers.34.mlp.experts.60.down_proj", "model.layers.34.mlp.experts.61.down_proj", "model.layers.34.mlp.experts.62.down_proj", "model.layers.34.mlp.experts.63.down_proj", "model.layers.34.mlp.experts.64.down_proj", "model.layers.34.mlp.experts.65.down_proj", "model.layers.34.mlp.experts.66.down_proj", "model.layers.34.mlp.experts.67.down_proj", "model.layers.34.mlp.experts.68.down_proj", "model.layers.34.mlp.experts.69.down_proj", "model.layers.34.mlp.experts.70.down_proj", "model.layers.34.mlp.experts.71.down_proj", "model.layers.34.mlp.experts.72.down_proj", "model.layers.34.mlp.experts.73.down_proj", "model.layers.34.mlp.experts.74.down_proj", "model.layers.34.mlp.experts.75.down_proj", "model.layers.34.mlp.experts.76.down_proj", "model.layers.34.mlp.experts.77.down_proj", "model.layers.34.mlp.experts.78.down_proj", "model.layers.34.mlp.experts.79.down_proj", "model.layers.34.mlp.experts.80.down_proj", "model.layers.34.mlp.experts.81.down_proj", "model.layers.34.mlp.experts.82.down_proj", "model.layers.34.mlp.experts.83.down_proj", "model.layers.34.mlp.experts.84.down_proj", "model.layers.34.mlp.experts.85.down_proj", "model.layers.34.mlp.experts.86.down_proj", "model.layers.34.mlp.experts.87.down_proj", "model.layers.34.mlp.experts.88.down_proj", "model.layers.34.mlp.experts.89.down_proj", "model.layers.34.mlp.experts.90.down_proj", "model.layers.34.mlp.experts.91.down_proj", "model.layers.34.mlp.experts.92.down_proj", "model.layers.34.mlp.experts.93.down_proj", "model.layers.34.mlp.experts.94.down_proj", "model.layers.34.mlp.experts.95.down_proj", "model.layers.34.mlp.experts.96.down_proj", "model.layers.34.mlp.experts.97.down_proj", "model.layers.34.mlp.experts.98.down_proj", "model.layers.34.mlp.experts.99.down_proj", "model.layers.34.mlp.experts.100.down_proj", "model.layers.34.mlp.experts.101.down_proj", "model.layers.34.mlp.experts.102.down_proj", "model.layers.34.mlp.experts.103.down_proj", "model.layers.34.mlp.experts.104.down_proj", "model.layers.34.mlp.experts.105.down_proj", "model.layers.34.mlp.experts.106.down_proj", "model.layers.34.mlp.experts.107.down_proj", "model.layers.34.mlp.experts.108.down_proj", "model.layers.34.mlp.experts.109.down_proj", "model.layers.34.mlp.experts.110.down_proj", "model.layers.34.mlp.experts.111.down_proj", "model.layers.34.mlp.experts.112.down_proj", "model.layers.34.mlp.experts.113.down_proj", "model.layers.34.mlp.experts.114.down_proj", "model.layers.34.mlp.experts.115.down_proj", "model.layers.34.mlp.experts.116.down_proj", "model.layers.34.mlp.experts.117.down_proj", "model.layers.34.mlp.experts.118.down_proj", "model.layers.34.mlp.experts.119.down_proj", "model.layers.34.mlp.experts.120.down_proj", "model.layers.34.mlp.experts.121.down_proj", "model.layers.34.mlp.experts.122.down_proj", "model.layers.34.mlp.experts.123.down_proj", "model.layers.34.mlp.experts.124.down_proj", "model.layers.34.mlp.experts.125.down_proj", "model.layers.34.mlp.experts.126.down_proj", "model.layers.34.mlp.experts.127.down_proj", "model.layers.34.mlp.experts.128.down_proj", "model.layers.34.mlp.experts.129.down_proj", "model.layers.34.mlp.experts.130.down_proj", "model.layers.34.mlp.experts.131.down_proj", "model.layers.34.mlp.experts.132.down_proj", "model.layers.34.mlp.experts.133.down_proj", "model.layers.34.mlp.experts.134.down_proj", "model.layers.34.mlp.experts.135.down_proj", "model.layers.34.mlp.experts.136.down_proj", "model.layers.34.mlp.experts.137.down_proj", "model.layers.34.mlp.experts.138.down_proj", "model.layers.34.mlp.experts.139.down_proj", "model.layers.34.mlp.experts.140.down_proj", "model.layers.34.mlp.experts.141.down_proj", "model.layers.34.mlp.experts.142.down_proj", "model.layers.34.mlp.experts.143.down_proj", "model.layers.34.mlp.experts.144.down_proj", "model.layers.34.mlp.experts.145.down_proj", "model.layers.34.mlp.experts.146.down_proj", "model.layers.34.mlp.experts.147.down_proj", "model.layers.34.mlp.experts.148.down_proj", "model.layers.34.mlp.experts.149.down_proj", "model.layers.34.mlp.experts.150.down_proj", "model.layers.34.mlp.experts.151.down_proj", "model.layers.34.mlp.experts.152.down_proj", "model.layers.34.mlp.experts.153.down_proj", "model.layers.34.mlp.experts.154.down_proj", "model.layers.34.mlp.experts.155.down_proj", "model.layers.34.mlp.experts.156.down_proj", "model.layers.34.mlp.experts.157.down_proj", "model.layers.34.mlp.experts.158.down_proj", "model.layers.34.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003975159488618374, "dbits": 1258291200 }, { "dkld": -0.00040723681449890414, "dbits": 2516582400 }, { "dkld": -0.0005251490511000156, "dbits": 3774873600 }, { "dkld": -0.0005291224457323607, "dbits": 6291456000 } ] }, { "idx": 207, "layers": [ "model.layers.35.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0004028655588626945, "dbits": 62914560 }, { "dkld": -0.000608326867222786, "dbits": 125829120 }, { "dkld": -0.0004853099584579551, "dbits": 188743680 }, { "dkld": -0.00048818234354258694, "dbits": 314572800 } ] }, { "idx": 208, "layers": [ "model.layers.35.self_attn.k_proj", "model.layers.35.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0009031804278493011, "dbits": 10485760 }, { "dkld": -0.0009221963584423176, "dbits": 20971520 }, { "dkld": -0.0007484544068575016, "dbits": 31457280 }, { "dkld": -0.0007846629247069414, "dbits": 52428800 } ] }, { "idx": 209, "layers": [ "model.layers.35.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0005360746756196133, "dbits": 62914560 }, { "dkld": -0.0008180148899555317, "dbits": 125829120 }, { "dkld": -0.0005589205771684702, "dbits": 188743680 }, { "dkld": -0.0007574159651994733, "dbits": 314572800 } ] }, { "idx": 210, "layers": [ "model.layers.35.mlp.shared_experts.gate_proj", "model.layers.35.mlp.shared_experts.up_proj", "model.layers.35.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00015495344996452332, "dbits": 23592960 }, { "dkld": -0.00016000904142857153, "dbits": 47185920 }, { "dkld": 0.00017608441412447773, "dbits": 70778880 }, { "dkld": 5.571041256188791e-05, "dbits": 117964800 } ] }, { "idx": 211, "layers": [ "model.layers.35.mlp.experts.0.gate_proj", "model.layers.35.mlp.experts.1.gate_proj", "model.layers.35.mlp.experts.2.gate_proj", "model.layers.35.mlp.experts.3.gate_proj", "model.layers.35.mlp.experts.4.gate_proj", "model.layers.35.mlp.experts.5.gate_proj", "model.layers.35.mlp.experts.6.gate_proj", "model.layers.35.mlp.experts.7.gate_proj", "model.layers.35.mlp.experts.8.gate_proj", "model.layers.35.mlp.experts.9.gate_proj", "model.layers.35.mlp.experts.10.gate_proj", "model.layers.35.mlp.experts.11.gate_proj", "model.layers.35.mlp.experts.12.gate_proj", "model.layers.35.mlp.experts.13.gate_proj", "model.layers.35.mlp.experts.14.gate_proj", "model.layers.35.mlp.experts.15.gate_proj", "model.layers.35.mlp.experts.16.gate_proj", "model.layers.35.mlp.experts.17.gate_proj", "model.layers.35.mlp.experts.18.gate_proj", "model.layers.35.mlp.experts.19.gate_proj", "model.layers.35.mlp.experts.20.gate_proj", "model.layers.35.mlp.experts.21.gate_proj", "model.layers.35.mlp.experts.22.gate_proj", "model.layers.35.mlp.experts.23.gate_proj", "model.layers.35.mlp.experts.24.gate_proj", "model.layers.35.mlp.experts.25.gate_proj", "model.layers.35.mlp.experts.26.gate_proj", "model.layers.35.mlp.experts.27.gate_proj", "model.layers.35.mlp.experts.28.gate_proj", "model.layers.35.mlp.experts.29.gate_proj", "model.layers.35.mlp.experts.30.gate_proj", "model.layers.35.mlp.experts.31.gate_proj", "model.layers.35.mlp.experts.32.gate_proj", "model.layers.35.mlp.experts.33.gate_proj", "model.layers.35.mlp.experts.34.gate_proj", "model.layers.35.mlp.experts.35.gate_proj", "model.layers.35.mlp.experts.36.gate_proj", "model.layers.35.mlp.experts.37.gate_proj", "model.layers.35.mlp.experts.38.gate_proj", "model.layers.35.mlp.experts.39.gate_proj", "model.layers.35.mlp.experts.40.gate_proj", "model.layers.35.mlp.experts.41.gate_proj", "model.layers.35.mlp.experts.42.gate_proj", "model.layers.35.mlp.experts.43.gate_proj", "model.layers.35.mlp.experts.44.gate_proj", "model.layers.35.mlp.experts.45.gate_proj", "model.layers.35.mlp.experts.46.gate_proj", "model.layers.35.mlp.experts.47.gate_proj", "model.layers.35.mlp.experts.48.gate_proj", "model.layers.35.mlp.experts.49.gate_proj", "model.layers.35.mlp.experts.50.gate_proj", "model.layers.35.mlp.experts.51.gate_proj", "model.layers.35.mlp.experts.52.gate_proj", "model.layers.35.mlp.experts.53.gate_proj", "model.layers.35.mlp.experts.54.gate_proj", "model.layers.35.mlp.experts.55.gate_proj", "model.layers.35.mlp.experts.56.gate_proj", "model.layers.35.mlp.experts.57.gate_proj", "model.layers.35.mlp.experts.58.gate_proj", "model.layers.35.mlp.experts.59.gate_proj", "model.layers.35.mlp.experts.60.gate_proj", "model.layers.35.mlp.experts.61.gate_proj", "model.layers.35.mlp.experts.62.gate_proj", "model.layers.35.mlp.experts.63.gate_proj", "model.layers.35.mlp.experts.64.gate_proj", "model.layers.35.mlp.experts.65.gate_proj", "model.layers.35.mlp.experts.66.gate_proj", "model.layers.35.mlp.experts.67.gate_proj", "model.layers.35.mlp.experts.68.gate_proj", "model.layers.35.mlp.experts.69.gate_proj", "model.layers.35.mlp.experts.70.gate_proj", "model.layers.35.mlp.experts.71.gate_proj", "model.layers.35.mlp.experts.72.gate_proj", "model.layers.35.mlp.experts.73.gate_proj", "model.layers.35.mlp.experts.74.gate_proj", "model.layers.35.mlp.experts.75.gate_proj", "model.layers.35.mlp.experts.76.gate_proj", "model.layers.35.mlp.experts.77.gate_proj", "model.layers.35.mlp.experts.78.gate_proj", "model.layers.35.mlp.experts.79.gate_proj", "model.layers.35.mlp.experts.80.gate_proj", "model.layers.35.mlp.experts.81.gate_proj", "model.layers.35.mlp.experts.82.gate_proj", "model.layers.35.mlp.experts.83.gate_proj", "model.layers.35.mlp.experts.84.gate_proj", "model.layers.35.mlp.experts.85.gate_proj", "model.layers.35.mlp.experts.86.gate_proj", "model.layers.35.mlp.experts.87.gate_proj", "model.layers.35.mlp.experts.88.gate_proj", "model.layers.35.mlp.experts.89.gate_proj", "model.layers.35.mlp.experts.90.gate_proj", "model.layers.35.mlp.experts.91.gate_proj", "model.layers.35.mlp.experts.92.gate_proj", "model.layers.35.mlp.experts.93.gate_proj", "model.layers.35.mlp.experts.94.gate_proj", "model.layers.35.mlp.experts.95.gate_proj", "model.layers.35.mlp.experts.96.gate_proj", "model.layers.35.mlp.experts.97.gate_proj", "model.layers.35.mlp.experts.98.gate_proj", "model.layers.35.mlp.experts.99.gate_proj", "model.layers.35.mlp.experts.100.gate_proj", "model.layers.35.mlp.experts.101.gate_proj", "model.layers.35.mlp.experts.102.gate_proj", "model.layers.35.mlp.experts.103.gate_proj", "model.layers.35.mlp.experts.104.gate_proj", "model.layers.35.mlp.experts.105.gate_proj", "model.layers.35.mlp.experts.106.gate_proj", "model.layers.35.mlp.experts.107.gate_proj", "model.layers.35.mlp.experts.108.gate_proj", "model.layers.35.mlp.experts.109.gate_proj", "model.layers.35.mlp.experts.110.gate_proj", "model.layers.35.mlp.experts.111.gate_proj", "model.layers.35.mlp.experts.112.gate_proj", "model.layers.35.mlp.experts.113.gate_proj", "model.layers.35.mlp.experts.114.gate_proj", "model.layers.35.mlp.experts.115.gate_proj", "model.layers.35.mlp.experts.116.gate_proj", "model.layers.35.mlp.experts.117.gate_proj", "model.layers.35.mlp.experts.118.gate_proj", "model.layers.35.mlp.experts.119.gate_proj", "model.layers.35.mlp.experts.120.gate_proj", "model.layers.35.mlp.experts.121.gate_proj", "model.layers.35.mlp.experts.122.gate_proj", "model.layers.35.mlp.experts.123.gate_proj", "model.layers.35.mlp.experts.124.gate_proj", "model.layers.35.mlp.experts.125.gate_proj", "model.layers.35.mlp.experts.126.gate_proj", "model.layers.35.mlp.experts.127.gate_proj", "model.layers.35.mlp.experts.128.gate_proj", "model.layers.35.mlp.experts.129.gate_proj", "model.layers.35.mlp.experts.130.gate_proj", "model.layers.35.mlp.experts.131.gate_proj", "model.layers.35.mlp.experts.132.gate_proj", "model.layers.35.mlp.experts.133.gate_proj", "model.layers.35.mlp.experts.134.gate_proj", "model.layers.35.mlp.experts.135.gate_proj", "model.layers.35.mlp.experts.136.gate_proj", "model.layers.35.mlp.experts.137.gate_proj", "model.layers.35.mlp.experts.138.gate_proj", "model.layers.35.mlp.experts.139.gate_proj", "model.layers.35.mlp.experts.140.gate_proj", "model.layers.35.mlp.experts.141.gate_proj", "model.layers.35.mlp.experts.142.gate_proj", "model.layers.35.mlp.experts.143.gate_proj", "model.layers.35.mlp.experts.144.gate_proj", "model.layers.35.mlp.experts.145.gate_proj", "model.layers.35.mlp.experts.146.gate_proj", "model.layers.35.mlp.experts.147.gate_proj", "model.layers.35.mlp.experts.148.gate_proj", "model.layers.35.mlp.experts.149.gate_proj", "model.layers.35.mlp.experts.150.gate_proj", "model.layers.35.mlp.experts.151.gate_proj", "model.layers.35.mlp.experts.152.gate_proj", "model.layers.35.mlp.experts.153.gate_proj", "model.layers.35.mlp.experts.154.gate_proj", "model.layers.35.mlp.experts.155.gate_proj", "model.layers.35.mlp.experts.156.gate_proj", "model.layers.35.mlp.experts.157.gate_proj", "model.layers.35.mlp.experts.158.gate_proj", "model.layers.35.mlp.experts.159.gate_proj", "model.layers.35.mlp.experts.0.up_proj", "model.layers.35.mlp.experts.1.up_proj", "model.layers.35.mlp.experts.2.up_proj", "model.layers.35.mlp.experts.3.up_proj", "model.layers.35.mlp.experts.4.up_proj", "model.layers.35.mlp.experts.5.up_proj", "model.layers.35.mlp.experts.6.up_proj", "model.layers.35.mlp.experts.7.up_proj", "model.layers.35.mlp.experts.8.up_proj", "model.layers.35.mlp.experts.9.up_proj", "model.layers.35.mlp.experts.10.up_proj", "model.layers.35.mlp.experts.11.up_proj", "model.layers.35.mlp.experts.12.up_proj", "model.layers.35.mlp.experts.13.up_proj", "model.layers.35.mlp.experts.14.up_proj", "model.layers.35.mlp.experts.15.up_proj", "model.layers.35.mlp.experts.16.up_proj", "model.layers.35.mlp.experts.17.up_proj", "model.layers.35.mlp.experts.18.up_proj", "model.layers.35.mlp.experts.19.up_proj", "model.layers.35.mlp.experts.20.up_proj", "model.layers.35.mlp.experts.21.up_proj", "model.layers.35.mlp.experts.22.up_proj", "model.layers.35.mlp.experts.23.up_proj", "model.layers.35.mlp.experts.24.up_proj", "model.layers.35.mlp.experts.25.up_proj", "model.layers.35.mlp.experts.26.up_proj", "model.layers.35.mlp.experts.27.up_proj", "model.layers.35.mlp.experts.28.up_proj", "model.layers.35.mlp.experts.29.up_proj", "model.layers.35.mlp.experts.30.up_proj", "model.layers.35.mlp.experts.31.up_proj", "model.layers.35.mlp.experts.32.up_proj", "model.layers.35.mlp.experts.33.up_proj", "model.layers.35.mlp.experts.34.up_proj", "model.layers.35.mlp.experts.35.up_proj", "model.layers.35.mlp.experts.36.up_proj", "model.layers.35.mlp.experts.37.up_proj", "model.layers.35.mlp.experts.38.up_proj", "model.layers.35.mlp.experts.39.up_proj", "model.layers.35.mlp.experts.40.up_proj", "model.layers.35.mlp.experts.41.up_proj", "model.layers.35.mlp.experts.42.up_proj", "model.layers.35.mlp.experts.43.up_proj", "model.layers.35.mlp.experts.44.up_proj", "model.layers.35.mlp.experts.45.up_proj", "model.layers.35.mlp.experts.46.up_proj", "model.layers.35.mlp.experts.47.up_proj", "model.layers.35.mlp.experts.48.up_proj", "model.layers.35.mlp.experts.49.up_proj", "model.layers.35.mlp.experts.50.up_proj", "model.layers.35.mlp.experts.51.up_proj", "model.layers.35.mlp.experts.52.up_proj", "model.layers.35.mlp.experts.53.up_proj", "model.layers.35.mlp.experts.54.up_proj", "model.layers.35.mlp.experts.55.up_proj", "model.layers.35.mlp.experts.56.up_proj", "model.layers.35.mlp.experts.57.up_proj", "model.layers.35.mlp.experts.58.up_proj", "model.layers.35.mlp.experts.59.up_proj", "model.layers.35.mlp.experts.60.up_proj", "model.layers.35.mlp.experts.61.up_proj", "model.layers.35.mlp.experts.62.up_proj", "model.layers.35.mlp.experts.63.up_proj", "model.layers.35.mlp.experts.64.up_proj", "model.layers.35.mlp.experts.65.up_proj", "model.layers.35.mlp.experts.66.up_proj", "model.layers.35.mlp.experts.67.up_proj", "model.layers.35.mlp.experts.68.up_proj", "model.layers.35.mlp.experts.69.up_proj", "model.layers.35.mlp.experts.70.up_proj", "model.layers.35.mlp.experts.71.up_proj", "model.layers.35.mlp.experts.72.up_proj", "model.layers.35.mlp.experts.73.up_proj", "model.layers.35.mlp.experts.74.up_proj", "model.layers.35.mlp.experts.75.up_proj", "model.layers.35.mlp.experts.76.up_proj", "model.layers.35.mlp.experts.77.up_proj", "model.layers.35.mlp.experts.78.up_proj", "model.layers.35.mlp.experts.79.up_proj", "model.layers.35.mlp.experts.80.up_proj", "model.layers.35.mlp.experts.81.up_proj", "model.layers.35.mlp.experts.82.up_proj", "model.layers.35.mlp.experts.83.up_proj", "model.layers.35.mlp.experts.84.up_proj", "model.layers.35.mlp.experts.85.up_proj", "model.layers.35.mlp.experts.86.up_proj", "model.layers.35.mlp.experts.87.up_proj", "model.layers.35.mlp.experts.88.up_proj", "model.layers.35.mlp.experts.89.up_proj", "model.layers.35.mlp.experts.90.up_proj", "model.layers.35.mlp.experts.91.up_proj", "model.layers.35.mlp.experts.92.up_proj", "model.layers.35.mlp.experts.93.up_proj", "model.layers.35.mlp.experts.94.up_proj", "model.layers.35.mlp.experts.95.up_proj", "model.layers.35.mlp.experts.96.up_proj", "model.layers.35.mlp.experts.97.up_proj", "model.layers.35.mlp.experts.98.up_proj", "model.layers.35.mlp.experts.99.up_proj", "model.layers.35.mlp.experts.100.up_proj", "model.layers.35.mlp.experts.101.up_proj", "model.layers.35.mlp.experts.102.up_proj", "model.layers.35.mlp.experts.103.up_proj", "model.layers.35.mlp.experts.104.up_proj", "model.layers.35.mlp.experts.105.up_proj", "model.layers.35.mlp.experts.106.up_proj", "model.layers.35.mlp.experts.107.up_proj", "model.layers.35.mlp.experts.108.up_proj", "model.layers.35.mlp.experts.109.up_proj", "model.layers.35.mlp.experts.110.up_proj", "model.layers.35.mlp.experts.111.up_proj", "model.layers.35.mlp.experts.112.up_proj", "model.layers.35.mlp.experts.113.up_proj", "model.layers.35.mlp.experts.114.up_proj", "model.layers.35.mlp.experts.115.up_proj", "model.layers.35.mlp.experts.116.up_proj", "model.layers.35.mlp.experts.117.up_proj", "model.layers.35.mlp.experts.118.up_proj", "model.layers.35.mlp.experts.119.up_proj", "model.layers.35.mlp.experts.120.up_proj", "model.layers.35.mlp.experts.121.up_proj", "model.layers.35.mlp.experts.122.up_proj", "model.layers.35.mlp.experts.123.up_proj", "model.layers.35.mlp.experts.124.up_proj", "model.layers.35.mlp.experts.125.up_proj", "model.layers.35.mlp.experts.126.up_proj", "model.layers.35.mlp.experts.127.up_proj", "model.layers.35.mlp.experts.128.up_proj", "model.layers.35.mlp.experts.129.up_proj", "model.layers.35.mlp.experts.130.up_proj", "model.layers.35.mlp.experts.131.up_proj", "model.layers.35.mlp.experts.132.up_proj", "model.layers.35.mlp.experts.133.up_proj", "model.layers.35.mlp.experts.134.up_proj", "model.layers.35.mlp.experts.135.up_proj", "model.layers.35.mlp.experts.136.up_proj", "model.layers.35.mlp.experts.137.up_proj", "model.layers.35.mlp.experts.138.up_proj", "model.layers.35.mlp.experts.139.up_proj", "model.layers.35.mlp.experts.140.up_proj", "model.layers.35.mlp.experts.141.up_proj", "model.layers.35.mlp.experts.142.up_proj", "model.layers.35.mlp.experts.143.up_proj", "model.layers.35.mlp.experts.144.up_proj", "model.layers.35.mlp.experts.145.up_proj", "model.layers.35.mlp.experts.146.up_proj", "model.layers.35.mlp.experts.147.up_proj", "model.layers.35.mlp.experts.148.up_proj", "model.layers.35.mlp.experts.149.up_proj", "model.layers.35.mlp.experts.150.up_proj", "model.layers.35.mlp.experts.151.up_proj", "model.layers.35.mlp.experts.152.up_proj", "model.layers.35.mlp.experts.153.up_proj", "model.layers.35.mlp.experts.154.up_proj", "model.layers.35.mlp.experts.155.up_proj", "model.layers.35.mlp.experts.156.up_proj", "model.layers.35.mlp.experts.157.up_proj", "model.layers.35.mlp.experts.158.up_proj", "model.layers.35.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0008833657018840341, "dbits": 2516582400 }, { "dkld": -0.001081555150449276, "dbits": 5033164800 }, { "dkld": -0.0011784057132899761, "dbits": 7549747200 }, { "dkld": -0.0011956688016653144, "dbits": 12582912000 } ] }, { "idx": 212, "layers": [ "model.layers.35.mlp.experts.0.down_proj", "model.layers.35.mlp.experts.1.down_proj", "model.layers.35.mlp.experts.2.down_proj", "model.layers.35.mlp.experts.3.down_proj", "model.layers.35.mlp.experts.4.down_proj", "model.layers.35.mlp.experts.5.down_proj", "model.layers.35.mlp.experts.6.down_proj", "model.layers.35.mlp.experts.7.down_proj", "model.layers.35.mlp.experts.8.down_proj", "model.layers.35.mlp.experts.9.down_proj", "model.layers.35.mlp.experts.10.down_proj", "model.layers.35.mlp.experts.11.down_proj", "model.layers.35.mlp.experts.12.down_proj", "model.layers.35.mlp.experts.13.down_proj", "model.layers.35.mlp.experts.14.down_proj", "model.layers.35.mlp.experts.15.down_proj", "model.layers.35.mlp.experts.16.down_proj", "model.layers.35.mlp.experts.17.down_proj", "model.layers.35.mlp.experts.18.down_proj", "model.layers.35.mlp.experts.19.down_proj", "model.layers.35.mlp.experts.20.down_proj", "model.layers.35.mlp.experts.21.down_proj", "model.layers.35.mlp.experts.22.down_proj", "model.layers.35.mlp.experts.23.down_proj", "model.layers.35.mlp.experts.24.down_proj", "model.layers.35.mlp.experts.25.down_proj", "model.layers.35.mlp.experts.26.down_proj", "model.layers.35.mlp.experts.27.down_proj", "model.layers.35.mlp.experts.28.down_proj", "model.layers.35.mlp.experts.29.down_proj", "model.layers.35.mlp.experts.30.down_proj", "model.layers.35.mlp.experts.31.down_proj", "model.layers.35.mlp.experts.32.down_proj", "model.layers.35.mlp.experts.33.down_proj", "model.layers.35.mlp.experts.34.down_proj", "model.layers.35.mlp.experts.35.down_proj", "model.layers.35.mlp.experts.36.down_proj", "model.layers.35.mlp.experts.37.down_proj", "model.layers.35.mlp.experts.38.down_proj", "model.layers.35.mlp.experts.39.down_proj", "model.layers.35.mlp.experts.40.down_proj", "model.layers.35.mlp.experts.41.down_proj", "model.layers.35.mlp.experts.42.down_proj", "model.layers.35.mlp.experts.43.down_proj", "model.layers.35.mlp.experts.44.down_proj", "model.layers.35.mlp.experts.45.down_proj", "model.layers.35.mlp.experts.46.down_proj", "model.layers.35.mlp.experts.47.down_proj", "model.layers.35.mlp.experts.48.down_proj", "model.layers.35.mlp.experts.49.down_proj", "model.layers.35.mlp.experts.50.down_proj", "model.layers.35.mlp.experts.51.down_proj", "model.layers.35.mlp.experts.52.down_proj", "model.layers.35.mlp.experts.53.down_proj", "model.layers.35.mlp.experts.54.down_proj", "model.layers.35.mlp.experts.55.down_proj", "model.layers.35.mlp.experts.56.down_proj", "model.layers.35.mlp.experts.57.down_proj", "model.layers.35.mlp.experts.58.down_proj", "model.layers.35.mlp.experts.59.down_proj", "model.layers.35.mlp.experts.60.down_proj", "model.layers.35.mlp.experts.61.down_proj", "model.layers.35.mlp.experts.62.down_proj", "model.layers.35.mlp.experts.63.down_proj", "model.layers.35.mlp.experts.64.down_proj", "model.layers.35.mlp.experts.65.down_proj", "model.layers.35.mlp.experts.66.down_proj", "model.layers.35.mlp.experts.67.down_proj", "model.layers.35.mlp.experts.68.down_proj", "model.layers.35.mlp.experts.69.down_proj", "model.layers.35.mlp.experts.70.down_proj", "model.layers.35.mlp.experts.71.down_proj", "model.layers.35.mlp.experts.72.down_proj", "model.layers.35.mlp.experts.73.down_proj", "model.layers.35.mlp.experts.74.down_proj", "model.layers.35.mlp.experts.75.down_proj", "model.layers.35.mlp.experts.76.down_proj", "model.layers.35.mlp.experts.77.down_proj", "model.layers.35.mlp.experts.78.down_proj", "model.layers.35.mlp.experts.79.down_proj", "model.layers.35.mlp.experts.80.down_proj", "model.layers.35.mlp.experts.81.down_proj", "model.layers.35.mlp.experts.82.down_proj", "model.layers.35.mlp.experts.83.down_proj", "model.layers.35.mlp.experts.84.down_proj", "model.layers.35.mlp.experts.85.down_proj", "model.layers.35.mlp.experts.86.down_proj", "model.layers.35.mlp.experts.87.down_proj", "model.layers.35.mlp.experts.88.down_proj", "model.layers.35.mlp.experts.89.down_proj", "model.layers.35.mlp.experts.90.down_proj", "model.layers.35.mlp.experts.91.down_proj", "model.layers.35.mlp.experts.92.down_proj", "model.layers.35.mlp.experts.93.down_proj", "model.layers.35.mlp.experts.94.down_proj", "model.layers.35.mlp.experts.95.down_proj", "model.layers.35.mlp.experts.96.down_proj", "model.layers.35.mlp.experts.97.down_proj", "model.layers.35.mlp.experts.98.down_proj", "model.layers.35.mlp.experts.99.down_proj", "model.layers.35.mlp.experts.100.down_proj", "model.layers.35.mlp.experts.101.down_proj", "model.layers.35.mlp.experts.102.down_proj", "model.layers.35.mlp.experts.103.down_proj", "model.layers.35.mlp.experts.104.down_proj", "model.layers.35.mlp.experts.105.down_proj", "model.layers.35.mlp.experts.106.down_proj", "model.layers.35.mlp.experts.107.down_proj", "model.layers.35.mlp.experts.108.down_proj", "model.layers.35.mlp.experts.109.down_proj", "model.layers.35.mlp.experts.110.down_proj", "model.layers.35.mlp.experts.111.down_proj", "model.layers.35.mlp.experts.112.down_proj", "model.layers.35.mlp.experts.113.down_proj", "model.layers.35.mlp.experts.114.down_proj", "model.layers.35.mlp.experts.115.down_proj", "model.layers.35.mlp.experts.116.down_proj", "model.layers.35.mlp.experts.117.down_proj", "model.layers.35.mlp.experts.118.down_proj", "model.layers.35.mlp.experts.119.down_proj", "model.layers.35.mlp.experts.120.down_proj", "model.layers.35.mlp.experts.121.down_proj", "model.layers.35.mlp.experts.122.down_proj", "model.layers.35.mlp.experts.123.down_proj", "model.layers.35.mlp.experts.124.down_proj", "model.layers.35.mlp.experts.125.down_proj", "model.layers.35.mlp.experts.126.down_proj", "model.layers.35.mlp.experts.127.down_proj", "model.layers.35.mlp.experts.128.down_proj", "model.layers.35.mlp.experts.129.down_proj", "model.layers.35.mlp.experts.130.down_proj", "model.layers.35.mlp.experts.131.down_proj", "model.layers.35.mlp.experts.132.down_proj", "model.layers.35.mlp.experts.133.down_proj", "model.layers.35.mlp.experts.134.down_proj", "model.layers.35.mlp.experts.135.down_proj", "model.layers.35.mlp.experts.136.down_proj", "model.layers.35.mlp.experts.137.down_proj", "model.layers.35.mlp.experts.138.down_proj", "model.layers.35.mlp.experts.139.down_proj", "model.layers.35.mlp.experts.140.down_proj", "model.layers.35.mlp.experts.141.down_proj", "model.layers.35.mlp.experts.142.down_proj", "model.layers.35.mlp.experts.143.down_proj", "model.layers.35.mlp.experts.144.down_proj", "model.layers.35.mlp.experts.145.down_proj", "model.layers.35.mlp.experts.146.down_proj", "model.layers.35.mlp.experts.147.down_proj", "model.layers.35.mlp.experts.148.down_proj", "model.layers.35.mlp.experts.149.down_proj", "model.layers.35.mlp.experts.150.down_proj", "model.layers.35.mlp.experts.151.down_proj", "model.layers.35.mlp.experts.152.down_proj", "model.layers.35.mlp.experts.153.down_proj", "model.layers.35.mlp.experts.154.down_proj", "model.layers.35.mlp.experts.155.down_proj", "model.layers.35.mlp.experts.156.down_proj", "model.layers.35.mlp.experts.157.down_proj", "model.layers.35.mlp.experts.158.down_proj", "model.layers.35.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0005300923250615708, "dbits": 1258291200 }, { "dkld": -0.0007056375965476119, "dbits": 2516582400 }, { "dkld": -0.0007744002155959662, "dbits": 3774873600 }, { "dkld": -0.00080056581646204, "dbits": 6291456000 } ] }, { "idx": 213, "layers": [ "model.layers.36.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0010281080380082075, "dbits": 62914560 }, { "dkld": 0.0009049266576766912, "dbits": 125829120 }, { "dkld": 0.0006357384845614433, "dbits": 188743680 }, { "dkld": 0.0006356300786137525, "dbits": 314572800 } ] }, { "idx": 214, "layers": [ "model.layers.36.self_attn.k_proj", "model.layers.36.self_attn.v_proj" ], "candidates": [ { "dkld": -7.682777941227237e-05, "dbits": 10485760 }, { "dkld": -0.0017570486292243087, "dbits": 20971520 }, { "dkld": -0.001575288269668823, "dbits": 31457280 }, { "dkld": -0.001598224975168705, "dbits": 52428800 } ] }, { "idx": 215, "layers": [ "model.layers.36.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0021030277013778798, "dbits": 62914560 }, { "dkld": -0.0021900677122175805, "dbits": 125829120 }, { "dkld": -0.002140116505324849, "dbits": 188743680 }, { "dkld": -0.0022501504980027703, "dbits": 314572800 } ] }, { "idx": 216, "layers": [ "model.layers.36.mlp.shared_experts.gate_proj", "model.layers.36.mlp.shared_experts.up_proj", "model.layers.36.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0017040768638253295, "dbits": 23592960 }, { "dkld": -0.001356167532503616, "dbits": 47185920 }, { "dkld": -0.001436361018568283, "dbits": 70778880 }, { "dkld": -0.0013651088811457157, "dbits": 117964800 } ] }, { "idx": 217, "layers": [ "model.layers.36.mlp.experts.0.gate_proj", "model.layers.36.mlp.experts.1.gate_proj", "model.layers.36.mlp.experts.2.gate_proj", "model.layers.36.mlp.experts.3.gate_proj", "model.layers.36.mlp.experts.4.gate_proj", "model.layers.36.mlp.experts.5.gate_proj", "model.layers.36.mlp.experts.6.gate_proj", "model.layers.36.mlp.experts.7.gate_proj", "model.layers.36.mlp.experts.8.gate_proj", "model.layers.36.mlp.experts.9.gate_proj", "model.layers.36.mlp.experts.10.gate_proj", "model.layers.36.mlp.experts.11.gate_proj", "model.layers.36.mlp.experts.12.gate_proj", "model.layers.36.mlp.experts.13.gate_proj", "model.layers.36.mlp.experts.14.gate_proj", "model.layers.36.mlp.experts.15.gate_proj", "model.layers.36.mlp.experts.16.gate_proj", "model.layers.36.mlp.experts.17.gate_proj", "model.layers.36.mlp.experts.18.gate_proj", "model.layers.36.mlp.experts.19.gate_proj", "model.layers.36.mlp.experts.20.gate_proj", "model.layers.36.mlp.experts.21.gate_proj", "model.layers.36.mlp.experts.22.gate_proj", "model.layers.36.mlp.experts.23.gate_proj", "model.layers.36.mlp.experts.24.gate_proj", "model.layers.36.mlp.experts.25.gate_proj", "model.layers.36.mlp.experts.26.gate_proj", "model.layers.36.mlp.experts.27.gate_proj", "model.layers.36.mlp.experts.28.gate_proj", "model.layers.36.mlp.experts.29.gate_proj", "model.layers.36.mlp.experts.30.gate_proj", "model.layers.36.mlp.experts.31.gate_proj", "model.layers.36.mlp.experts.32.gate_proj", "model.layers.36.mlp.experts.33.gate_proj", "model.layers.36.mlp.experts.34.gate_proj", "model.layers.36.mlp.experts.35.gate_proj", "model.layers.36.mlp.experts.36.gate_proj", "model.layers.36.mlp.experts.37.gate_proj", "model.layers.36.mlp.experts.38.gate_proj", "model.layers.36.mlp.experts.39.gate_proj", "model.layers.36.mlp.experts.40.gate_proj", "model.layers.36.mlp.experts.41.gate_proj", "model.layers.36.mlp.experts.42.gate_proj", "model.layers.36.mlp.experts.43.gate_proj", "model.layers.36.mlp.experts.44.gate_proj", "model.layers.36.mlp.experts.45.gate_proj", "model.layers.36.mlp.experts.46.gate_proj", "model.layers.36.mlp.experts.47.gate_proj", "model.layers.36.mlp.experts.48.gate_proj", "model.layers.36.mlp.experts.49.gate_proj", "model.layers.36.mlp.experts.50.gate_proj", "model.layers.36.mlp.experts.51.gate_proj", "model.layers.36.mlp.experts.52.gate_proj", "model.layers.36.mlp.experts.53.gate_proj", "model.layers.36.mlp.experts.54.gate_proj", "model.layers.36.mlp.experts.55.gate_proj", "model.layers.36.mlp.experts.56.gate_proj", "model.layers.36.mlp.experts.57.gate_proj", "model.layers.36.mlp.experts.58.gate_proj", "model.layers.36.mlp.experts.59.gate_proj", "model.layers.36.mlp.experts.60.gate_proj", "model.layers.36.mlp.experts.61.gate_proj", "model.layers.36.mlp.experts.62.gate_proj", "model.layers.36.mlp.experts.63.gate_proj", "model.layers.36.mlp.experts.64.gate_proj", "model.layers.36.mlp.experts.65.gate_proj", "model.layers.36.mlp.experts.66.gate_proj", "model.layers.36.mlp.experts.67.gate_proj", "model.layers.36.mlp.experts.68.gate_proj", "model.layers.36.mlp.experts.69.gate_proj", "model.layers.36.mlp.experts.70.gate_proj", "model.layers.36.mlp.experts.71.gate_proj", "model.layers.36.mlp.experts.72.gate_proj", "model.layers.36.mlp.experts.73.gate_proj", "model.layers.36.mlp.experts.74.gate_proj", "model.layers.36.mlp.experts.75.gate_proj", "model.layers.36.mlp.experts.76.gate_proj", "model.layers.36.mlp.experts.77.gate_proj", "model.layers.36.mlp.experts.78.gate_proj", "model.layers.36.mlp.experts.79.gate_proj", "model.layers.36.mlp.experts.80.gate_proj", "model.layers.36.mlp.experts.81.gate_proj", "model.layers.36.mlp.experts.82.gate_proj", "model.layers.36.mlp.experts.83.gate_proj", "model.layers.36.mlp.experts.84.gate_proj", "model.layers.36.mlp.experts.85.gate_proj", "model.layers.36.mlp.experts.86.gate_proj", "model.layers.36.mlp.experts.87.gate_proj", "model.layers.36.mlp.experts.88.gate_proj", "model.layers.36.mlp.experts.89.gate_proj", "model.layers.36.mlp.experts.90.gate_proj", "model.layers.36.mlp.experts.91.gate_proj", "model.layers.36.mlp.experts.92.gate_proj", "model.layers.36.mlp.experts.93.gate_proj", "model.layers.36.mlp.experts.94.gate_proj", "model.layers.36.mlp.experts.95.gate_proj", "model.layers.36.mlp.experts.96.gate_proj", "model.layers.36.mlp.experts.97.gate_proj", "model.layers.36.mlp.experts.98.gate_proj", "model.layers.36.mlp.experts.99.gate_proj", "model.layers.36.mlp.experts.100.gate_proj", "model.layers.36.mlp.experts.101.gate_proj", "model.layers.36.mlp.experts.102.gate_proj", "model.layers.36.mlp.experts.103.gate_proj", "model.layers.36.mlp.experts.104.gate_proj", "model.layers.36.mlp.experts.105.gate_proj", "model.layers.36.mlp.experts.106.gate_proj", "model.layers.36.mlp.experts.107.gate_proj", "model.layers.36.mlp.experts.108.gate_proj", "model.layers.36.mlp.experts.109.gate_proj", "model.layers.36.mlp.experts.110.gate_proj", "model.layers.36.mlp.experts.111.gate_proj", "model.layers.36.mlp.experts.112.gate_proj", "model.layers.36.mlp.experts.113.gate_proj", "model.layers.36.mlp.experts.114.gate_proj", "model.layers.36.mlp.experts.115.gate_proj", "model.layers.36.mlp.experts.116.gate_proj", "model.layers.36.mlp.experts.117.gate_proj", "model.layers.36.mlp.experts.118.gate_proj", "model.layers.36.mlp.experts.119.gate_proj", "model.layers.36.mlp.experts.120.gate_proj", "model.layers.36.mlp.experts.121.gate_proj", "model.layers.36.mlp.experts.122.gate_proj", "model.layers.36.mlp.experts.123.gate_proj", "model.layers.36.mlp.experts.124.gate_proj", "model.layers.36.mlp.experts.125.gate_proj", "model.layers.36.mlp.experts.126.gate_proj", "model.layers.36.mlp.experts.127.gate_proj", "model.layers.36.mlp.experts.128.gate_proj", "model.layers.36.mlp.experts.129.gate_proj", "model.layers.36.mlp.experts.130.gate_proj", "model.layers.36.mlp.experts.131.gate_proj", "model.layers.36.mlp.experts.132.gate_proj", "model.layers.36.mlp.experts.133.gate_proj", "model.layers.36.mlp.experts.134.gate_proj", "model.layers.36.mlp.experts.135.gate_proj", "model.layers.36.mlp.experts.136.gate_proj", "model.layers.36.mlp.experts.137.gate_proj", "model.layers.36.mlp.experts.138.gate_proj", "model.layers.36.mlp.experts.139.gate_proj", "model.layers.36.mlp.experts.140.gate_proj", "model.layers.36.mlp.experts.141.gate_proj", "model.layers.36.mlp.experts.142.gate_proj", "model.layers.36.mlp.experts.143.gate_proj", "model.layers.36.mlp.experts.144.gate_proj", "model.layers.36.mlp.experts.145.gate_proj", "model.layers.36.mlp.experts.146.gate_proj", "model.layers.36.mlp.experts.147.gate_proj", "model.layers.36.mlp.experts.148.gate_proj", "model.layers.36.mlp.experts.149.gate_proj", "model.layers.36.mlp.experts.150.gate_proj", "model.layers.36.mlp.experts.151.gate_proj", "model.layers.36.mlp.experts.152.gate_proj", "model.layers.36.mlp.experts.153.gate_proj", "model.layers.36.mlp.experts.154.gate_proj", "model.layers.36.mlp.experts.155.gate_proj", "model.layers.36.mlp.experts.156.gate_proj", "model.layers.36.mlp.experts.157.gate_proj", "model.layers.36.mlp.experts.158.gate_proj", "model.layers.36.mlp.experts.159.gate_proj", "model.layers.36.mlp.experts.0.up_proj", "model.layers.36.mlp.experts.1.up_proj", "model.layers.36.mlp.experts.2.up_proj", "model.layers.36.mlp.experts.3.up_proj", "model.layers.36.mlp.experts.4.up_proj", "model.layers.36.mlp.experts.5.up_proj", "model.layers.36.mlp.experts.6.up_proj", "model.layers.36.mlp.experts.7.up_proj", "model.layers.36.mlp.experts.8.up_proj", "model.layers.36.mlp.experts.9.up_proj", "model.layers.36.mlp.experts.10.up_proj", "model.layers.36.mlp.experts.11.up_proj", "model.layers.36.mlp.experts.12.up_proj", "model.layers.36.mlp.experts.13.up_proj", "model.layers.36.mlp.experts.14.up_proj", "model.layers.36.mlp.experts.15.up_proj", "model.layers.36.mlp.experts.16.up_proj", "model.layers.36.mlp.experts.17.up_proj", "model.layers.36.mlp.experts.18.up_proj", "model.layers.36.mlp.experts.19.up_proj", "model.layers.36.mlp.experts.20.up_proj", "model.layers.36.mlp.experts.21.up_proj", "model.layers.36.mlp.experts.22.up_proj", "model.layers.36.mlp.experts.23.up_proj", "model.layers.36.mlp.experts.24.up_proj", "model.layers.36.mlp.experts.25.up_proj", "model.layers.36.mlp.experts.26.up_proj", "model.layers.36.mlp.experts.27.up_proj", "model.layers.36.mlp.experts.28.up_proj", "model.layers.36.mlp.experts.29.up_proj", "model.layers.36.mlp.experts.30.up_proj", "model.layers.36.mlp.experts.31.up_proj", "model.layers.36.mlp.experts.32.up_proj", "model.layers.36.mlp.experts.33.up_proj", "model.layers.36.mlp.experts.34.up_proj", "model.layers.36.mlp.experts.35.up_proj", "model.layers.36.mlp.experts.36.up_proj", "model.layers.36.mlp.experts.37.up_proj", "model.layers.36.mlp.experts.38.up_proj", "model.layers.36.mlp.experts.39.up_proj", "model.layers.36.mlp.experts.40.up_proj", "model.layers.36.mlp.experts.41.up_proj", "model.layers.36.mlp.experts.42.up_proj", "model.layers.36.mlp.experts.43.up_proj", "model.layers.36.mlp.experts.44.up_proj", "model.layers.36.mlp.experts.45.up_proj", "model.layers.36.mlp.experts.46.up_proj", "model.layers.36.mlp.experts.47.up_proj", "model.layers.36.mlp.experts.48.up_proj", "model.layers.36.mlp.experts.49.up_proj", "model.layers.36.mlp.experts.50.up_proj", "model.layers.36.mlp.experts.51.up_proj", "model.layers.36.mlp.experts.52.up_proj", "model.layers.36.mlp.experts.53.up_proj", "model.layers.36.mlp.experts.54.up_proj", "model.layers.36.mlp.experts.55.up_proj", "model.layers.36.mlp.experts.56.up_proj", "model.layers.36.mlp.experts.57.up_proj", "model.layers.36.mlp.experts.58.up_proj", "model.layers.36.mlp.experts.59.up_proj", "model.layers.36.mlp.experts.60.up_proj", "model.layers.36.mlp.experts.61.up_proj", "model.layers.36.mlp.experts.62.up_proj", "model.layers.36.mlp.experts.63.up_proj", "model.layers.36.mlp.experts.64.up_proj", "model.layers.36.mlp.experts.65.up_proj", "model.layers.36.mlp.experts.66.up_proj", "model.layers.36.mlp.experts.67.up_proj", "model.layers.36.mlp.experts.68.up_proj", "model.layers.36.mlp.experts.69.up_proj", "model.layers.36.mlp.experts.70.up_proj", "model.layers.36.mlp.experts.71.up_proj", "model.layers.36.mlp.experts.72.up_proj", "model.layers.36.mlp.experts.73.up_proj", "model.layers.36.mlp.experts.74.up_proj", "model.layers.36.mlp.experts.75.up_proj", "model.layers.36.mlp.experts.76.up_proj", "model.layers.36.mlp.experts.77.up_proj", "model.layers.36.mlp.experts.78.up_proj", "model.layers.36.mlp.experts.79.up_proj", "model.layers.36.mlp.experts.80.up_proj", "model.layers.36.mlp.experts.81.up_proj", "model.layers.36.mlp.experts.82.up_proj", "model.layers.36.mlp.experts.83.up_proj", "model.layers.36.mlp.experts.84.up_proj", "model.layers.36.mlp.experts.85.up_proj", "model.layers.36.mlp.experts.86.up_proj", "model.layers.36.mlp.experts.87.up_proj", "model.layers.36.mlp.experts.88.up_proj", "model.layers.36.mlp.experts.89.up_proj", "model.layers.36.mlp.experts.90.up_proj", "model.layers.36.mlp.experts.91.up_proj", "model.layers.36.mlp.experts.92.up_proj", "model.layers.36.mlp.experts.93.up_proj", "model.layers.36.mlp.experts.94.up_proj", "model.layers.36.mlp.experts.95.up_proj", "model.layers.36.mlp.experts.96.up_proj", "model.layers.36.mlp.experts.97.up_proj", "model.layers.36.mlp.experts.98.up_proj", "model.layers.36.mlp.experts.99.up_proj", "model.layers.36.mlp.experts.100.up_proj", "model.layers.36.mlp.experts.101.up_proj", "model.layers.36.mlp.experts.102.up_proj", "model.layers.36.mlp.experts.103.up_proj", "model.layers.36.mlp.experts.104.up_proj", "model.layers.36.mlp.experts.105.up_proj", "model.layers.36.mlp.experts.106.up_proj", "model.layers.36.mlp.experts.107.up_proj", "model.layers.36.mlp.experts.108.up_proj", "model.layers.36.mlp.experts.109.up_proj", "model.layers.36.mlp.experts.110.up_proj", "model.layers.36.mlp.experts.111.up_proj", "model.layers.36.mlp.experts.112.up_proj", "model.layers.36.mlp.experts.113.up_proj", "model.layers.36.mlp.experts.114.up_proj", "model.layers.36.mlp.experts.115.up_proj", "model.layers.36.mlp.experts.116.up_proj", "model.layers.36.mlp.experts.117.up_proj", "model.layers.36.mlp.experts.118.up_proj", "model.layers.36.mlp.experts.119.up_proj", "model.layers.36.mlp.experts.120.up_proj", "model.layers.36.mlp.experts.121.up_proj", "model.layers.36.mlp.experts.122.up_proj", "model.layers.36.mlp.experts.123.up_proj", "model.layers.36.mlp.experts.124.up_proj", "model.layers.36.mlp.experts.125.up_proj", "model.layers.36.mlp.experts.126.up_proj", "model.layers.36.mlp.experts.127.up_proj", "model.layers.36.mlp.experts.128.up_proj", "model.layers.36.mlp.experts.129.up_proj", "model.layers.36.mlp.experts.130.up_proj", "model.layers.36.mlp.experts.131.up_proj", "model.layers.36.mlp.experts.132.up_proj", "model.layers.36.mlp.experts.133.up_proj", "model.layers.36.mlp.experts.134.up_proj", "model.layers.36.mlp.experts.135.up_proj", "model.layers.36.mlp.experts.136.up_proj", "model.layers.36.mlp.experts.137.up_proj", "model.layers.36.mlp.experts.138.up_proj", "model.layers.36.mlp.experts.139.up_proj", "model.layers.36.mlp.experts.140.up_proj", "model.layers.36.mlp.experts.141.up_proj", "model.layers.36.mlp.experts.142.up_proj", "model.layers.36.mlp.experts.143.up_proj", "model.layers.36.mlp.experts.144.up_proj", "model.layers.36.mlp.experts.145.up_proj", "model.layers.36.mlp.experts.146.up_proj", "model.layers.36.mlp.experts.147.up_proj", "model.layers.36.mlp.experts.148.up_proj", "model.layers.36.mlp.experts.149.up_proj", "model.layers.36.mlp.experts.150.up_proj", "model.layers.36.mlp.experts.151.up_proj", "model.layers.36.mlp.experts.152.up_proj", "model.layers.36.mlp.experts.153.up_proj", "model.layers.36.mlp.experts.154.up_proj", "model.layers.36.mlp.experts.155.up_proj", "model.layers.36.mlp.experts.156.up_proj", "model.layers.36.mlp.experts.157.up_proj", "model.layers.36.mlp.experts.158.up_proj", "model.layers.36.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00022512022405862808, "dbits": 2516582400 }, { "dkld": -0.0005300514400005396, "dbits": 5033164800 }, { "dkld": -0.0005990936420857962, "dbits": 7549747200 }, { "dkld": -0.0006079554557800321, "dbits": 12582912000 } ] }, { "idx": 218, "layers": [ "model.layers.36.mlp.experts.0.down_proj", "model.layers.36.mlp.experts.1.down_proj", "model.layers.36.mlp.experts.2.down_proj", "model.layers.36.mlp.experts.3.down_proj", "model.layers.36.mlp.experts.4.down_proj", "model.layers.36.mlp.experts.5.down_proj", "model.layers.36.mlp.experts.6.down_proj", "model.layers.36.mlp.experts.7.down_proj", "model.layers.36.mlp.experts.8.down_proj", "model.layers.36.mlp.experts.9.down_proj", "model.layers.36.mlp.experts.10.down_proj", "model.layers.36.mlp.experts.11.down_proj", "model.layers.36.mlp.experts.12.down_proj", "model.layers.36.mlp.experts.13.down_proj", "model.layers.36.mlp.experts.14.down_proj", "model.layers.36.mlp.experts.15.down_proj", "model.layers.36.mlp.experts.16.down_proj", "model.layers.36.mlp.experts.17.down_proj", "model.layers.36.mlp.experts.18.down_proj", "model.layers.36.mlp.experts.19.down_proj", "model.layers.36.mlp.experts.20.down_proj", "model.layers.36.mlp.experts.21.down_proj", "model.layers.36.mlp.experts.22.down_proj", "model.layers.36.mlp.experts.23.down_proj", "model.layers.36.mlp.experts.24.down_proj", "model.layers.36.mlp.experts.25.down_proj", "model.layers.36.mlp.experts.26.down_proj", "model.layers.36.mlp.experts.27.down_proj", "model.layers.36.mlp.experts.28.down_proj", "model.layers.36.mlp.experts.29.down_proj", "model.layers.36.mlp.experts.30.down_proj", "model.layers.36.mlp.experts.31.down_proj", "model.layers.36.mlp.experts.32.down_proj", "model.layers.36.mlp.experts.33.down_proj", "model.layers.36.mlp.experts.34.down_proj", "model.layers.36.mlp.experts.35.down_proj", "model.layers.36.mlp.experts.36.down_proj", "model.layers.36.mlp.experts.37.down_proj", "model.layers.36.mlp.experts.38.down_proj", "model.layers.36.mlp.experts.39.down_proj", "model.layers.36.mlp.experts.40.down_proj", "model.layers.36.mlp.experts.41.down_proj", "model.layers.36.mlp.experts.42.down_proj", "model.layers.36.mlp.experts.43.down_proj", "model.layers.36.mlp.experts.44.down_proj", "model.layers.36.mlp.experts.45.down_proj", "model.layers.36.mlp.experts.46.down_proj", "model.layers.36.mlp.experts.47.down_proj", "model.layers.36.mlp.experts.48.down_proj", "model.layers.36.mlp.experts.49.down_proj", "model.layers.36.mlp.experts.50.down_proj", "model.layers.36.mlp.experts.51.down_proj", "model.layers.36.mlp.experts.52.down_proj", "model.layers.36.mlp.experts.53.down_proj", "model.layers.36.mlp.experts.54.down_proj", "model.layers.36.mlp.experts.55.down_proj", "model.layers.36.mlp.experts.56.down_proj", "model.layers.36.mlp.experts.57.down_proj", "model.layers.36.mlp.experts.58.down_proj", "model.layers.36.mlp.experts.59.down_proj", "model.layers.36.mlp.experts.60.down_proj", "model.layers.36.mlp.experts.61.down_proj", "model.layers.36.mlp.experts.62.down_proj", "model.layers.36.mlp.experts.63.down_proj", "model.layers.36.mlp.experts.64.down_proj", "model.layers.36.mlp.experts.65.down_proj", "model.layers.36.mlp.experts.66.down_proj", "model.layers.36.mlp.experts.67.down_proj", "model.layers.36.mlp.experts.68.down_proj", "model.layers.36.mlp.experts.69.down_proj", "model.layers.36.mlp.experts.70.down_proj", "model.layers.36.mlp.experts.71.down_proj", "model.layers.36.mlp.experts.72.down_proj", "model.layers.36.mlp.experts.73.down_proj", "model.layers.36.mlp.experts.74.down_proj", "model.layers.36.mlp.experts.75.down_proj", "model.layers.36.mlp.experts.76.down_proj", "model.layers.36.mlp.experts.77.down_proj", "model.layers.36.mlp.experts.78.down_proj", "model.layers.36.mlp.experts.79.down_proj", "model.layers.36.mlp.experts.80.down_proj", "model.layers.36.mlp.experts.81.down_proj", "model.layers.36.mlp.experts.82.down_proj", "model.layers.36.mlp.experts.83.down_proj", "model.layers.36.mlp.experts.84.down_proj", "model.layers.36.mlp.experts.85.down_proj", "model.layers.36.mlp.experts.86.down_proj", "model.layers.36.mlp.experts.87.down_proj", "model.layers.36.mlp.experts.88.down_proj", "model.layers.36.mlp.experts.89.down_proj", "model.layers.36.mlp.experts.90.down_proj", "model.layers.36.mlp.experts.91.down_proj", "model.layers.36.mlp.experts.92.down_proj", "model.layers.36.mlp.experts.93.down_proj", "model.layers.36.mlp.experts.94.down_proj", "model.layers.36.mlp.experts.95.down_proj", "model.layers.36.mlp.experts.96.down_proj", "model.layers.36.mlp.experts.97.down_proj", "model.layers.36.mlp.experts.98.down_proj", "model.layers.36.mlp.experts.99.down_proj", "model.layers.36.mlp.experts.100.down_proj", "model.layers.36.mlp.experts.101.down_proj", "model.layers.36.mlp.experts.102.down_proj", "model.layers.36.mlp.experts.103.down_proj", "model.layers.36.mlp.experts.104.down_proj", "model.layers.36.mlp.experts.105.down_proj", "model.layers.36.mlp.experts.106.down_proj", "model.layers.36.mlp.experts.107.down_proj", "model.layers.36.mlp.experts.108.down_proj", "model.layers.36.mlp.experts.109.down_proj", "model.layers.36.mlp.experts.110.down_proj", "model.layers.36.mlp.experts.111.down_proj", "model.layers.36.mlp.experts.112.down_proj", "model.layers.36.mlp.experts.113.down_proj", "model.layers.36.mlp.experts.114.down_proj", "model.layers.36.mlp.experts.115.down_proj", "model.layers.36.mlp.experts.116.down_proj", "model.layers.36.mlp.experts.117.down_proj", "model.layers.36.mlp.experts.118.down_proj", "model.layers.36.mlp.experts.119.down_proj", "model.layers.36.mlp.experts.120.down_proj", "model.layers.36.mlp.experts.121.down_proj", "model.layers.36.mlp.experts.122.down_proj", "model.layers.36.mlp.experts.123.down_proj", "model.layers.36.mlp.experts.124.down_proj", "model.layers.36.mlp.experts.125.down_proj", "model.layers.36.mlp.experts.126.down_proj", "model.layers.36.mlp.experts.127.down_proj", "model.layers.36.mlp.experts.128.down_proj", "model.layers.36.mlp.experts.129.down_proj", "model.layers.36.mlp.experts.130.down_proj", "model.layers.36.mlp.experts.131.down_proj", "model.layers.36.mlp.experts.132.down_proj", "model.layers.36.mlp.experts.133.down_proj", "model.layers.36.mlp.experts.134.down_proj", "model.layers.36.mlp.experts.135.down_proj", "model.layers.36.mlp.experts.136.down_proj", "model.layers.36.mlp.experts.137.down_proj", "model.layers.36.mlp.experts.138.down_proj", "model.layers.36.mlp.experts.139.down_proj", "model.layers.36.mlp.experts.140.down_proj", "model.layers.36.mlp.experts.141.down_proj", "model.layers.36.mlp.experts.142.down_proj", "model.layers.36.mlp.experts.143.down_proj", "model.layers.36.mlp.experts.144.down_proj", "model.layers.36.mlp.experts.145.down_proj", "model.layers.36.mlp.experts.146.down_proj", "model.layers.36.mlp.experts.147.down_proj", "model.layers.36.mlp.experts.148.down_proj", "model.layers.36.mlp.experts.149.down_proj", "model.layers.36.mlp.experts.150.down_proj", "model.layers.36.mlp.experts.151.down_proj", "model.layers.36.mlp.experts.152.down_proj", "model.layers.36.mlp.experts.153.down_proj", "model.layers.36.mlp.experts.154.down_proj", "model.layers.36.mlp.experts.155.down_proj", "model.layers.36.mlp.experts.156.down_proj", "model.layers.36.mlp.experts.157.down_proj", "model.layers.36.mlp.experts.158.down_proj", "model.layers.36.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.000280844699591401, "dbits": 1258291200 }, { "dkld": -0.00036254292353987694, "dbits": 2516582400 }, { "dkld": -0.0005087334662675913, "dbits": 3774873600 }, { "dkld": -0.000508292391896259, "dbits": 6291456000 } ] }, { "idx": 219, "layers": [ "model.layers.37.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0007827311754226712, "dbits": 62914560 }, { "dkld": -0.0005726687610149439, "dbits": 125829120 }, { "dkld": -0.0006047803908586502, "dbits": 188743680 }, { "dkld": -0.0006246911361813545, "dbits": 314572800 } ] }, { "idx": 220, "layers": [ "model.layers.37.self_attn.k_proj", "model.layers.37.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00029728254303335033, "dbits": 10485760 }, { "dkld": -0.0002482743002474308, "dbits": 20971520 }, { "dkld": 1.7994828522202577e-05, "dbits": 31457280 }, { "dkld": -0.00028539467602968493, "dbits": 52428800 } ] }, { "idx": 221, "layers": [ "model.layers.37.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0018476307392120417, "dbits": 62914560 }, { "dkld": -0.001344950124621394, "dbits": 125829120 }, { "dkld": -0.0012411110103130452, "dbits": 188743680 }, { "dkld": -0.0011636901646852604, "dbits": 314572800 } ] }, { "idx": 222, "layers": [ "model.layers.37.mlp.shared_experts.gate_proj", "model.layers.37.mlp.shared_experts.up_proj", "model.layers.37.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0008200546726584435, "dbits": 23592960 }, { "dkld": 0.0009524287655949537, "dbits": 47185920 }, { "dkld": 0.0007745385169982827, "dbits": 70778880 }, { "dkld": 0.0007255906239151955, "dbits": 117964800 } ] }, { "idx": 223, "layers": [ "model.layers.37.mlp.experts.0.gate_proj", "model.layers.37.mlp.experts.1.gate_proj", "model.layers.37.mlp.experts.2.gate_proj", "model.layers.37.mlp.experts.3.gate_proj", "model.layers.37.mlp.experts.4.gate_proj", "model.layers.37.mlp.experts.5.gate_proj", "model.layers.37.mlp.experts.6.gate_proj", "model.layers.37.mlp.experts.7.gate_proj", "model.layers.37.mlp.experts.8.gate_proj", "model.layers.37.mlp.experts.9.gate_proj", "model.layers.37.mlp.experts.10.gate_proj", "model.layers.37.mlp.experts.11.gate_proj", "model.layers.37.mlp.experts.12.gate_proj", "model.layers.37.mlp.experts.13.gate_proj", "model.layers.37.mlp.experts.14.gate_proj", "model.layers.37.mlp.experts.15.gate_proj", "model.layers.37.mlp.experts.16.gate_proj", "model.layers.37.mlp.experts.17.gate_proj", "model.layers.37.mlp.experts.18.gate_proj", "model.layers.37.mlp.experts.19.gate_proj", "model.layers.37.mlp.experts.20.gate_proj", "model.layers.37.mlp.experts.21.gate_proj", "model.layers.37.mlp.experts.22.gate_proj", "model.layers.37.mlp.experts.23.gate_proj", "model.layers.37.mlp.experts.24.gate_proj", "model.layers.37.mlp.experts.25.gate_proj", "model.layers.37.mlp.experts.26.gate_proj", "model.layers.37.mlp.experts.27.gate_proj", "model.layers.37.mlp.experts.28.gate_proj", "model.layers.37.mlp.experts.29.gate_proj", "model.layers.37.mlp.experts.30.gate_proj", "model.layers.37.mlp.experts.31.gate_proj", "model.layers.37.mlp.experts.32.gate_proj", "model.layers.37.mlp.experts.33.gate_proj", "model.layers.37.mlp.experts.34.gate_proj", "model.layers.37.mlp.experts.35.gate_proj", "model.layers.37.mlp.experts.36.gate_proj", "model.layers.37.mlp.experts.37.gate_proj", "model.layers.37.mlp.experts.38.gate_proj", "model.layers.37.mlp.experts.39.gate_proj", "model.layers.37.mlp.experts.40.gate_proj", "model.layers.37.mlp.experts.41.gate_proj", "model.layers.37.mlp.experts.42.gate_proj", "model.layers.37.mlp.experts.43.gate_proj", "model.layers.37.mlp.experts.44.gate_proj", "model.layers.37.mlp.experts.45.gate_proj", "model.layers.37.mlp.experts.46.gate_proj", "model.layers.37.mlp.experts.47.gate_proj", "model.layers.37.mlp.experts.48.gate_proj", "model.layers.37.mlp.experts.49.gate_proj", "model.layers.37.mlp.experts.50.gate_proj", "model.layers.37.mlp.experts.51.gate_proj", "model.layers.37.mlp.experts.52.gate_proj", "model.layers.37.mlp.experts.53.gate_proj", "model.layers.37.mlp.experts.54.gate_proj", "model.layers.37.mlp.experts.55.gate_proj", "model.layers.37.mlp.experts.56.gate_proj", "model.layers.37.mlp.experts.57.gate_proj", "model.layers.37.mlp.experts.58.gate_proj", "model.layers.37.mlp.experts.59.gate_proj", "model.layers.37.mlp.experts.60.gate_proj", "model.layers.37.mlp.experts.61.gate_proj", "model.layers.37.mlp.experts.62.gate_proj", "model.layers.37.mlp.experts.63.gate_proj", "model.layers.37.mlp.experts.64.gate_proj", "model.layers.37.mlp.experts.65.gate_proj", "model.layers.37.mlp.experts.66.gate_proj", "model.layers.37.mlp.experts.67.gate_proj", "model.layers.37.mlp.experts.68.gate_proj", "model.layers.37.mlp.experts.69.gate_proj", "model.layers.37.mlp.experts.70.gate_proj", "model.layers.37.mlp.experts.71.gate_proj", "model.layers.37.mlp.experts.72.gate_proj", "model.layers.37.mlp.experts.73.gate_proj", "model.layers.37.mlp.experts.74.gate_proj", "model.layers.37.mlp.experts.75.gate_proj", "model.layers.37.mlp.experts.76.gate_proj", "model.layers.37.mlp.experts.77.gate_proj", "model.layers.37.mlp.experts.78.gate_proj", "model.layers.37.mlp.experts.79.gate_proj", "model.layers.37.mlp.experts.80.gate_proj", "model.layers.37.mlp.experts.81.gate_proj", "model.layers.37.mlp.experts.82.gate_proj", "model.layers.37.mlp.experts.83.gate_proj", "model.layers.37.mlp.experts.84.gate_proj", "model.layers.37.mlp.experts.85.gate_proj", "model.layers.37.mlp.experts.86.gate_proj", "model.layers.37.mlp.experts.87.gate_proj", "model.layers.37.mlp.experts.88.gate_proj", "model.layers.37.mlp.experts.89.gate_proj", "model.layers.37.mlp.experts.90.gate_proj", "model.layers.37.mlp.experts.91.gate_proj", "model.layers.37.mlp.experts.92.gate_proj", "model.layers.37.mlp.experts.93.gate_proj", "model.layers.37.mlp.experts.94.gate_proj", "model.layers.37.mlp.experts.95.gate_proj", "model.layers.37.mlp.experts.96.gate_proj", "model.layers.37.mlp.experts.97.gate_proj", "model.layers.37.mlp.experts.98.gate_proj", "model.layers.37.mlp.experts.99.gate_proj", "model.layers.37.mlp.experts.100.gate_proj", "model.layers.37.mlp.experts.101.gate_proj", "model.layers.37.mlp.experts.102.gate_proj", "model.layers.37.mlp.experts.103.gate_proj", "model.layers.37.mlp.experts.104.gate_proj", "model.layers.37.mlp.experts.105.gate_proj", "model.layers.37.mlp.experts.106.gate_proj", "model.layers.37.mlp.experts.107.gate_proj", "model.layers.37.mlp.experts.108.gate_proj", "model.layers.37.mlp.experts.109.gate_proj", "model.layers.37.mlp.experts.110.gate_proj", "model.layers.37.mlp.experts.111.gate_proj", "model.layers.37.mlp.experts.112.gate_proj", "model.layers.37.mlp.experts.113.gate_proj", "model.layers.37.mlp.experts.114.gate_proj", "model.layers.37.mlp.experts.115.gate_proj", "model.layers.37.mlp.experts.116.gate_proj", "model.layers.37.mlp.experts.117.gate_proj", "model.layers.37.mlp.experts.118.gate_proj", "model.layers.37.mlp.experts.119.gate_proj", "model.layers.37.mlp.experts.120.gate_proj", "model.layers.37.mlp.experts.121.gate_proj", "model.layers.37.mlp.experts.122.gate_proj", "model.layers.37.mlp.experts.123.gate_proj", "model.layers.37.mlp.experts.124.gate_proj", "model.layers.37.mlp.experts.125.gate_proj", "model.layers.37.mlp.experts.126.gate_proj", "model.layers.37.mlp.experts.127.gate_proj", "model.layers.37.mlp.experts.128.gate_proj", "model.layers.37.mlp.experts.129.gate_proj", "model.layers.37.mlp.experts.130.gate_proj", "model.layers.37.mlp.experts.131.gate_proj", "model.layers.37.mlp.experts.132.gate_proj", "model.layers.37.mlp.experts.133.gate_proj", "model.layers.37.mlp.experts.134.gate_proj", "model.layers.37.mlp.experts.135.gate_proj", "model.layers.37.mlp.experts.136.gate_proj", "model.layers.37.mlp.experts.137.gate_proj", "model.layers.37.mlp.experts.138.gate_proj", "model.layers.37.mlp.experts.139.gate_proj", "model.layers.37.mlp.experts.140.gate_proj", "model.layers.37.mlp.experts.141.gate_proj", "model.layers.37.mlp.experts.142.gate_proj", "model.layers.37.mlp.experts.143.gate_proj", "model.layers.37.mlp.experts.144.gate_proj", "model.layers.37.mlp.experts.145.gate_proj", "model.layers.37.mlp.experts.146.gate_proj", "model.layers.37.mlp.experts.147.gate_proj", "model.layers.37.mlp.experts.148.gate_proj", "model.layers.37.mlp.experts.149.gate_proj", "model.layers.37.mlp.experts.150.gate_proj", "model.layers.37.mlp.experts.151.gate_proj", "model.layers.37.mlp.experts.152.gate_proj", "model.layers.37.mlp.experts.153.gate_proj", "model.layers.37.mlp.experts.154.gate_proj", "model.layers.37.mlp.experts.155.gate_proj", "model.layers.37.mlp.experts.156.gate_proj", "model.layers.37.mlp.experts.157.gate_proj", "model.layers.37.mlp.experts.158.gate_proj", "model.layers.37.mlp.experts.159.gate_proj", "model.layers.37.mlp.experts.0.up_proj", "model.layers.37.mlp.experts.1.up_proj", "model.layers.37.mlp.experts.2.up_proj", "model.layers.37.mlp.experts.3.up_proj", "model.layers.37.mlp.experts.4.up_proj", "model.layers.37.mlp.experts.5.up_proj", "model.layers.37.mlp.experts.6.up_proj", "model.layers.37.mlp.experts.7.up_proj", "model.layers.37.mlp.experts.8.up_proj", "model.layers.37.mlp.experts.9.up_proj", "model.layers.37.mlp.experts.10.up_proj", "model.layers.37.mlp.experts.11.up_proj", "model.layers.37.mlp.experts.12.up_proj", "model.layers.37.mlp.experts.13.up_proj", "model.layers.37.mlp.experts.14.up_proj", "model.layers.37.mlp.experts.15.up_proj", "model.layers.37.mlp.experts.16.up_proj", "model.layers.37.mlp.experts.17.up_proj", "model.layers.37.mlp.experts.18.up_proj", "model.layers.37.mlp.experts.19.up_proj", "model.layers.37.mlp.experts.20.up_proj", "model.layers.37.mlp.experts.21.up_proj", "model.layers.37.mlp.experts.22.up_proj", "model.layers.37.mlp.experts.23.up_proj", "model.layers.37.mlp.experts.24.up_proj", "model.layers.37.mlp.experts.25.up_proj", "model.layers.37.mlp.experts.26.up_proj", "model.layers.37.mlp.experts.27.up_proj", "model.layers.37.mlp.experts.28.up_proj", "model.layers.37.mlp.experts.29.up_proj", "model.layers.37.mlp.experts.30.up_proj", "model.layers.37.mlp.experts.31.up_proj", "model.layers.37.mlp.experts.32.up_proj", "model.layers.37.mlp.experts.33.up_proj", "model.layers.37.mlp.experts.34.up_proj", "model.layers.37.mlp.experts.35.up_proj", "model.layers.37.mlp.experts.36.up_proj", "model.layers.37.mlp.experts.37.up_proj", "model.layers.37.mlp.experts.38.up_proj", "model.layers.37.mlp.experts.39.up_proj", "model.layers.37.mlp.experts.40.up_proj", "model.layers.37.mlp.experts.41.up_proj", "model.layers.37.mlp.experts.42.up_proj", "model.layers.37.mlp.experts.43.up_proj", "model.layers.37.mlp.experts.44.up_proj", "model.layers.37.mlp.experts.45.up_proj", "model.layers.37.mlp.experts.46.up_proj", "model.layers.37.mlp.experts.47.up_proj", "model.layers.37.mlp.experts.48.up_proj", "model.layers.37.mlp.experts.49.up_proj", "model.layers.37.mlp.experts.50.up_proj", "model.layers.37.mlp.experts.51.up_proj", "model.layers.37.mlp.experts.52.up_proj", "model.layers.37.mlp.experts.53.up_proj", "model.layers.37.mlp.experts.54.up_proj", "model.layers.37.mlp.experts.55.up_proj", "model.layers.37.mlp.experts.56.up_proj", "model.layers.37.mlp.experts.57.up_proj", "model.layers.37.mlp.experts.58.up_proj", "model.layers.37.mlp.experts.59.up_proj", "model.layers.37.mlp.experts.60.up_proj", "model.layers.37.mlp.experts.61.up_proj", "model.layers.37.mlp.experts.62.up_proj", "model.layers.37.mlp.experts.63.up_proj", "model.layers.37.mlp.experts.64.up_proj", "model.layers.37.mlp.experts.65.up_proj", "model.layers.37.mlp.experts.66.up_proj", "model.layers.37.mlp.experts.67.up_proj", "model.layers.37.mlp.experts.68.up_proj", "model.layers.37.mlp.experts.69.up_proj", "model.layers.37.mlp.experts.70.up_proj", "model.layers.37.mlp.experts.71.up_proj", "model.layers.37.mlp.experts.72.up_proj", "model.layers.37.mlp.experts.73.up_proj", "model.layers.37.mlp.experts.74.up_proj", "model.layers.37.mlp.experts.75.up_proj", "model.layers.37.mlp.experts.76.up_proj", "model.layers.37.mlp.experts.77.up_proj", "model.layers.37.mlp.experts.78.up_proj", "model.layers.37.mlp.experts.79.up_proj", "model.layers.37.mlp.experts.80.up_proj", "model.layers.37.mlp.experts.81.up_proj", "model.layers.37.mlp.experts.82.up_proj", "model.layers.37.mlp.experts.83.up_proj", "model.layers.37.mlp.experts.84.up_proj", "model.layers.37.mlp.experts.85.up_proj", "model.layers.37.mlp.experts.86.up_proj", "model.layers.37.mlp.experts.87.up_proj", "model.layers.37.mlp.experts.88.up_proj", "model.layers.37.mlp.experts.89.up_proj", "model.layers.37.mlp.experts.90.up_proj", "model.layers.37.mlp.experts.91.up_proj", "model.layers.37.mlp.experts.92.up_proj", "model.layers.37.mlp.experts.93.up_proj", "model.layers.37.mlp.experts.94.up_proj", "model.layers.37.mlp.experts.95.up_proj", "model.layers.37.mlp.experts.96.up_proj", "model.layers.37.mlp.experts.97.up_proj", "model.layers.37.mlp.experts.98.up_proj", "model.layers.37.mlp.experts.99.up_proj", "model.layers.37.mlp.experts.100.up_proj", "model.layers.37.mlp.experts.101.up_proj", "model.layers.37.mlp.experts.102.up_proj", "model.layers.37.mlp.experts.103.up_proj", "model.layers.37.mlp.experts.104.up_proj", "model.layers.37.mlp.experts.105.up_proj", "model.layers.37.mlp.experts.106.up_proj", "model.layers.37.mlp.experts.107.up_proj", "model.layers.37.mlp.experts.108.up_proj", "model.layers.37.mlp.experts.109.up_proj", "model.layers.37.mlp.experts.110.up_proj", "model.layers.37.mlp.experts.111.up_proj", "model.layers.37.mlp.experts.112.up_proj", "model.layers.37.mlp.experts.113.up_proj", "model.layers.37.mlp.experts.114.up_proj", "model.layers.37.mlp.experts.115.up_proj", "model.layers.37.mlp.experts.116.up_proj", "model.layers.37.mlp.experts.117.up_proj", "model.layers.37.mlp.experts.118.up_proj", "model.layers.37.mlp.experts.119.up_proj", "model.layers.37.mlp.experts.120.up_proj", "model.layers.37.mlp.experts.121.up_proj", "model.layers.37.mlp.experts.122.up_proj", "model.layers.37.mlp.experts.123.up_proj", "model.layers.37.mlp.experts.124.up_proj", "model.layers.37.mlp.experts.125.up_proj", "model.layers.37.mlp.experts.126.up_proj", "model.layers.37.mlp.experts.127.up_proj", "model.layers.37.mlp.experts.128.up_proj", "model.layers.37.mlp.experts.129.up_proj", "model.layers.37.mlp.experts.130.up_proj", "model.layers.37.mlp.experts.131.up_proj", "model.layers.37.mlp.experts.132.up_proj", "model.layers.37.mlp.experts.133.up_proj", "model.layers.37.mlp.experts.134.up_proj", "model.layers.37.mlp.experts.135.up_proj", "model.layers.37.mlp.experts.136.up_proj", "model.layers.37.mlp.experts.137.up_proj", "model.layers.37.mlp.experts.138.up_proj", "model.layers.37.mlp.experts.139.up_proj", "model.layers.37.mlp.experts.140.up_proj", "model.layers.37.mlp.experts.141.up_proj", "model.layers.37.mlp.experts.142.up_proj", "model.layers.37.mlp.experts.143.up_proj", "model.layers.37.mlp.experts.144.up_proj", "model.layers.37.mlp.experts.145.up_proj", "model.layers.37.mlp.experts.146.up_proj", "model.layers.37.mlp.experts.147.up_proj", "model.layers.37.mlp.experts.148.up_proj", "model.layers.37.mlp.experts.149.up_proj", "model.layers.37.mlp.experts.150.up_proj", "model.layers.37.mlp.experts.151.up_proj", "model.layers.37.mlp.experts.152.up_proj", "model.layers.37.mlp.experts.153.up_proj", "model.layers.37.mlp.experts.154.up_proj", "model.layers.37.mlp.experts.155.up_proj", "model.layers.37.mlp.experts.156.up_proj", "model.layers.37.mlp.experts.157.up_proj", "model.layers.37.mlp.experts.158.up_proj", "model.layers.37.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00014797868207097609, "dbits": 2516582400 }, { "dkld": -0.0002407477237284239, "dbits": 5033164800 }, { "dkld": -0.00026134569197894253, "dbits": 7549747200 }, { "dkld": -0.0002515068277716692, "dbits": 12582912000 } ] }, { "idx": 224, "layers": [ "model.layers.37.mlp.experts.0.down_proj", "model.layers.37.mlp.experts.1.down_proj", "model.layers.37.mlp.experts.2.down_proj", "model.layers.37.mlp.experts.3.down_proj", "model.layers.37.mlp.experts.4.down_proj", "model.layers.37.mlp.experts.5.down_proj", "model.layers.37.mlp.experts.6.down_proj", "model.layers.37.mlp.experts.7.down_proj", "model.layers.37.mlp.experts.8.down_proj", "model.layers.37.mlp.experts.9.down_proj", "model.layers.37.mlp.experts.10.down_proj", "model.layers.37.mlp.experts.11.down_proj", "model.layers.37.mlp.experts.12.down_proj", "model.layers.37.mlp.experts.13.down_proj", "model.layers.37.mlp.experts.14.down_proj", "model.layers.37.mlp.experts.15.down_proj", "model.layers.37.mlp.experts.16.down_proj", "model.layers.37.mlp.experts.17.down_proj", "model.layers.37.mlp.experts.18.down_proj", "model.layers.37.mlp.experts.19.down_proj", "model.layers.37.mlp.experts.20.down_proj", "model.layers.37.mlp.experts.21.down_proj", "model.layers.37.mlp.experts.22.down_proj", "model.layers.37.mlp.experts.23.down_proj", "model.layers.37.mlp.experts.24.down_proj", "model.layers.37.mlp.experts.25.down_proj", "model.layers.37.mlp.experts.26.down_proj", "model.layers.37.mlp.experts.27.down_proj", "model.layers.37.mlp.experts.28.down_proj", "model.layers.37.mlp.experts.29.down_proj", "model.layers.37.mlp.experts.30.down_proj", "model.layers.37.mlp.experts.31.down_proj", "model.layers.37.mlp.experts.32.down_proj", "model.layers.37.mlp.experts.33.down_proj", "model.layers.37.mlp.experts.34.down_proj", "model.layers.37.mlp.experts.35.down_proj", "model.layers.37.mlp.experts.36.down_proj", "model.layers.37.mlp.experts.37.down_proj", "model.layers.37.mlp.experts.38.down_proj", "model.layers.37.mlp.experts.39.down_proj", "model.layers.37.mlp.experts.40.down_proj", "model.layers.37.mlp.experts.41.down_proj", "model.layers.37.mlp.experts.42.down_proj", "model.layers.37.mlp.experts.43.down_proj", "model.layers.37.mlp.experts.44.down_proj", "model.layers.37.mlp.experts.45.down_proj", "model.layers.37.mlp.experts.46.down_proj", "model.layers.37.mlp.experts.47.down_proj", "model.layers.37.mlp.experts.48.down_proj", "model.layers.37.mlp.experts.49.down_proj", "model.layers.37.mlp.experts.50.down_proj", "model.layers.37.mlp.experts.51.down_proj", "model.layers.37.mlp.experts.52.down_proj", "model.layers.37.mlp.experts.53.down_proj", "model.layers.37.mlp.experts.54.down_proj", "model.layers.37.mlp.experts.55.down_proj", "model.layers.37.mlp.experts.56.down_proj", "model.layers.37.mlp.experts.57.down_proj", "model.layers.37.mlp.experts.58.down_proj", "model.layers.37.mlp.experts.59.down_proj", "model.layers.37.mlp.experts.60.down_proj", "model.layers.37.mlp.experts.61.down_proj", "model.layers.37.mlp.experts.62.down_proj", "model.layers.37.mlp.experts.63.down_proj", "model.layers.37.mlp.experts.64.down_proj", "model.layers.37.mlp.experts.65.down_proj", "model.layers.37.mlp.experts.66.down_proj", "model.layers.37.mlp.experts.67.down_proj", "model.layers.37.mlp.experts.68.down_proj", "model.layers.37.mlp.experts.69.down_proj", "model.layers.37.mlp.experts.70.down_proj", "model.layers.37.mlp.experts.71.down_proj", "model.layers.37.mlp.experts.72.down_proj", "model.layers.37.mlp.experts.73.down_proj", "model.layers.37.mlp.experts.74.down_proj", "model.layers.37.mlp.experts.75.down_proj", "model.layers.37.mlp.experts.76.down_proj", "model.layers.37.mlp.experts.77.down_proj", "model.layers.37.mlp.experts.78.down_proj", "model.layers.37.mlp.experts.79.down_proj", "model.layers.37.mlp.experts.80.down_proj", "model.layers.37.mlp.experts.81.down_proj", "model.layers.37.mlp.experts.82.down_proj", "model.layers.37.mlp.experts.83.down_proj", "model.layers.37.mlp.experts.84.down_proj", "model.layers.37.mlp.experts.85.down_proj", "model.layers.37.mlp.experts.86.down_proj", "model.layers.37.mlp.experts.87.down_proj", "model.layers.37.mlp.experts.88.down_proj", "model.layers.37.mlp.experts.89.down_proj", "model.layers.37.mlp.experts.90.down_proj", "model.layers.37.mlp.experts.91.down_proj", "model.layers.37.mlp.experts.92.down_proj", "model.layers.37.mlp.experts.93.down_proj", "model.layers.37.mlp.experts.94.down_proj", "model.layers.37.mlp.experts.95.down_proj", "model.layers.37.mlp.experts.96.down_proj", "model.layers.37.mlp.experts.97.down_proj", "model.layers.37.mlp.experts.98.down_proj", "model.layers.37.mlp.experts.99.down_proj", "model.layers.37.mlp.experts.100.down_proj", "model.layers.37.mlp.experts.101.down_proj", "model.layers.37.mlp.experts.102.down_proj", "model.layers.37.mlp.experts.103.down_proj", "model.layers.37.mlp.experts.104.down_proj", "model.layers.37.mlp.experts.105.down_proj", "model.layers.37.mlp.experts.106.down_proj", "model.layers.37.mlp.experts.107.down_proj", "model.layers.37.mlp.experts.108.down_proj", "model.layers.37.mlp.experts.109.down_proj", "model.layers.37.mlp.experts.110.down_proj", "model.layers.37.mlp.experts.111.down_proj", "model.layers.37.mlp.experts.112.down_proj", "model.layers.37.mlp.experts.113.down_proj", "model.layers.37.mlp.experts.114.down_proj", "model.layers.37.mlp.experts.115.down_proj", "model.layers.37.mlp.experts.116.down_proj", "model.layers.37.mlp.experts.117.down_proj", "model.layers.37.mlp.experts.118.down_proj", "model.layers.37.mlp.experts.119.down_proj", "model.layers.37.mlp.experts.120.down_proj", "model.layers.37.mlp.experts.121.down_proj", "model.layers.37.mlp.experts.122.down_proj", "model.layers.37.mlp.experts.123.down_proj", "model.layers.37.mlp.experts.124.down_proj", "model.layers.37.mlp.experts.125.down_proj", "model.layers.37.mlp.experts.126.down_proj", "model.layers.37.mlp.experts.127.down_proj", "model.layers.37.mlp.experts.128.down_proj", "model.layers.37.mlp.experts.129.down_proj", "model.layers.37.mlp.experts.130.down_proj", "model.layers.37.mlp.experts.131.down_proj", "model.layers.37.mlp.experts.132.down_proj", "model.layers.37.mlp.experts.133.down_proj", "model.layers.37.mlp.experts.134.down_proj", "model.layers.37.mlp.experts.135.down_proj", "model.layers.37.mlp.experts.136.down_proj", "model.layers.37.mlp.experts.137.down_proj", "model.layers.37.mlp.experts.138.down_proj", "model.layers.37.mlp.experts.139.down_proj", "model.layers.37.mlp.experts.140.down_proj", "model.layers.37.mlp.experts.141.down_proj", "model.layers.37.mlp.experts.142.down_proj", "model.layers.37.mlp.experts.143.down_proj", "model.layers.37.mlp.experts.144.down_proj", "model.layers.37.mlp.experts.145.down_proj", "model.layers.37.mlp.experts.146.down_proj", "model.layers.37.mlp.experts.147.down_proj", "model.layers.37.mlp.experts.148.down_proj", "model.layers.37.mlp.experts.149.down_proj", "model.layers.37.mlp.experts.150.down_proj", "model.layers.37.mlp.experts.151.down_proj", "model.layers.37.mlp.experts.152.down_proj", "model.layers.37.mlp.experts.153.down_proj", "model.layers.37.mlp.experts.154.down_proj", "model.layers.37.mlp.experts.155.down_proj", "model.layers.37.mlp.experts.156.down_proj", "model.layers.37.mlp.experts.157.down_proj", "model.layers.37.mlp.experts.158.down_proj", "model.layers.37.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00015058675780893482, "dbits": 1258291200 }, { "dkld": -0.00017772465944290716, "dbits": 2516582400 }, { "dkld": -0.0002044240012765014, "dbits": 3774873600 }, { "dkld": -0.0002223406918346882, "dbits": 6291456000 } ] }, { "idx": 225, "layers": [ "model.layers.38.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0010034802369773416, "dbits": 62914560 }, { "dkld": -0.0011996231973171317, "dbits": 125829120 }, { "dkld": -0.0010123202577233398, "dbits": 188743680 }, { "dkld": -0.0010124155320227202, "dbits": 314572800 } ] }, { "idx": 226, "layers": [ "model.layers.38.self_attn.k_proj", "model.layers.38.self_attn.v_proj" ], "candidates": [ { "dkld": -7.316889241337776e-05, "dbits": 10485760 }, { "dkld": -0.00013049133121967316, "dbits": 20971520 }, { "dkld": -5.3746439516555444e-05, "dbits": 31457280 }, { "dkld": -0.0002076433040201775, "dbits": 52428800 } ] }, { "idx": 227, "layers": [ "model.layers.38.self_attn.o_proj" ], "candidates": [ { "dkld": -7.332451641560433e-05, "dbits": 62914560 }, { "dkld": -0.0001657480373978698, "dbits": 125829120 }, { "dkld": -0.00016838330775499621, "dbits": 188743680 }, { "dkld": -0.00019404925405980267, "dbits": 314572800 } ] }, { "idx": 228, "layers": [ "model.layers.38.mlp.shared_experts.gate_proj", "model.layers.38.mlp.shared_experts.up_proj", "model.layers.38.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 1.7666164785615224e-05, "dbits": 23592960 }, { "dkld": 0.00027898680418729505, "dbits": 47185920 }, { "dkld": 0.0004289051517844117, "dbits": 70778880 }, { "dkld": 0.0004469813778996412, "dbits": 117964800 } ] }, { "idx": 229, "layers": [ "model.layers.38.mlp.experts.0.gate_proj", "model.layers.38.mlp.experts.1.gate_proj", "model.layers.38.mlp.experts.2.gate_proj", "model.layers.38.mlp.experts.3.gate_proj", "model.layers.38.mlp.experts.4.gate_proj", "model.layers.38.mlp.experts.5.gate_proj", "model.layers.38.mlp.experts.6.gate_proj", "model.layers.38.mlp.experts.7.gate_proj", "model.layers.38.mlp.experts.8.gate_proj", "model.layers.38.mlp.experts.9.gate_proj", "model.layers.38.mlp.experts.10.gate_proj", "model.layers.38.mlp.experts.11.gate_proj", "model.layers.38.mlp.experts.12.gate_proj", "model.layers.38.mlp.experts.13.gate_proj", "model.layers.38.mlp.experts.14.gate_proj", "model.layers.38.mlp.experts.15.gate_proj", "model.layers.38.mlp.experts.16.gate_proj", "model.layers.38.mlp.experts.17.gate_proj", "model.layers.38.mlp.experts.18.gate_proj", "model.layers.38.mlp.experts.19.gate_proj", "model.layers.38.mlp.experts.20.gate_proj", "model.layers.38.mlp.experts.21.gate_proj", "model.layers.38.mlp.experts.22.gate_proj", "model.layers.38.mlp.experts.23.gate_proj", "model.layers.38.mlp.experts.24.gate_proj", "model.layers.38.mlp.experts.25.gate_proj", "model.layers.38.mlp.experts.26.gate_proj", "model.layers.38.mlp.experts.27.gate_proj", "model.layers.38.mlp.experts.28.gate_proj", "model.layers.38.mlp.experts.29.gate_proj", "model.layers.38.mlp.experts.30.gate_proj", "model.layers.38.mlp.experts.31.gate_proj", "model.layers.38.mlp.experts.32.gate_proj", "model.layers.38.mlp.experts.33.gate_proj", "model.layers.38.mlp.experts.34.gate_proj", "model.layers.38.mlp.experts.35.gate_proj", "model.layers.38.mlp.experts.36.gate_proj", "model.layers.38.mlp.experts.37.gate_proj", "model.layers.38.mlp.experts.38.gate_proj", "model.layers.38.mlp.experts.39.gate_proj", "model.layers.38.mlp.experts.40.gate_proj", "model.layers.38.mlp.experts.41.gate_proj", "model.layers.38.mlp.experts.42.gate_proj", "model.layers.38.mlp.experts.43.gate_proj", "model.layers.38.mlp.experts.44.gate_proj", "model.layers.38.mlp.experts.45.gate_proj", "model.layers.38.mlp.experts.46.gate_proj", "model.layers.38.mlp.experts.47.gate_proj", "model.layers.38.mlp.experts.48.gate_proj", "model.layers.38.mlp.experts.49.gate_proj", "model.layers.38.mlp.experts.50.gate_proj", "model.layers.38.mlp.experts.51.gate_proj", "model.layers.38.mlp.experts.52.gate_proj", "model.layers.38.mlp.experts.53.gate_proj", "model.layers.38.mlp.experts.54.gate_proj", "model.layers.38.mlp.experts.55.gate_proj", "model.layers.38.mlp.experts.56.gate_proj", "model.layers.38.mlp.experts.57.gate_proj", "model.layers.38.mlp.experts.58.gate_proj", "model.layers.38.mlp.experts.59.gate_proj", "model.layers.38.mlp.experts.60.gate_proj", "model.layers.38.mlp.experts.61.gate_proj", "model.layers.38.mlp.experts.62.gate_proj", "model.layers.38.mlp.experts.63.gate_proj", "model.layers.38.mlp.experts.64.gate_proj", "model.layers.38.mlp.experts.65.gate_proj", "model.layers.38.mlp.experts.66.gate_proj", "model.layers.38.mlp.experts.67.gate_proj", "model.layers.38.mlp.experts.68.gate_proj", "model.layers.38.mlp.experts.69.gate_proj", "model.layers.38.mlp.experts.70.gate_proj", "model.layers.38.mlp.experts.71.gate_proj", "model.layers.38.mlp.experts.72.gate_proj", "model.layers.38.mlp.experts.73.gate_proj", "model.layers.38.mlp.experts.74.gate_proj", "model.layers.38.mlp.experts.75.gate_proj", "model.layers.38.mlp.experts.76.gate_proj", "model.layers.38.mlp.experts.77.gate_proj", "model.layers.38.mlp.experts.78.gate_proj", "model.layers.38.mlp.experts.79.gate_proj", "model.layers.38.mlp.experts.80.gate_proj", "model.layers.38.mlp.experts.81.gate_proj", "model.layers.38.mlp.experts.82.gate_proj", "model.layers.38.mlp.experts.83.gate_proj", "model.layers.38.mlp.experts.84.gate_proj", "model.layers.38.mlp.experts.85.gate_proj", "model.layers.38.mlp.experts.86.gate_proj", "model.layers.38.mlp.experts.87.gate_proj", "model.layers.38.mlp.experts.88.gate_proj", "model.layers.38.mlp.experts.89.gate_proj", "model.layers.38.mlp.experts.90.gate_proj", "model.layers.38.mlp.experts.91.gate_proj", "model.layers.38.mlp.experts.92.gate_proj", "model.layers.38.mlp.experts.93.gate_proj", "model.layers.38.mlp.experts.94.gate_proj", "model.layers.38.mlp.experts.95.gate_proj", "model.layers.38.mlp.experts.96.gate_proj", "model.layers.38.mlp.experts.97.gate_proj", "model.layers.38.mlp.experts.98.gate_proj", "model.layers.38.mlp.experts.99.gate_proj", "model.layers.38.mlp.experts.100.gate_proj", "model.layers.38.mlp.experts.101.gate_proj", "model.layers.38.mlp.experts.102.gate_proj", "model.layers.38.mlp.experts.103.gate_proj", "model.layers.38.mlp.experts.104.gate_proj", "model.layers.38.mlp.experts.105.gate_proj", "model.layers.38.mlp.experts.106.gate_proj", "model.layers.38.mlp.experts.107.gate_proj", "model.layers.38.mlp.experts.108.gate_proj", "model.layers.38.mlp.experts.109.gate_proj", "model.layers.38.mlp.experts.110.gate_proj", "model.layers.38.mlp.experts.111.gate_proj", "model.layers.38.mlp.experts.112.gate_proj", "model.layers.38.mlp.experts.113.gate_proj", "model.layers.38.mlp.experts.114.gate_proj", "model.layers.38.mlp.experts.115.gate_proj", "model.layers.38.mlp.experts.116.gate_proj", "model.layers.38.mlp.experts.117.gate_proj", "model.layers.38.mlp.experts.118.gate_proj", "model.layers.38.mlp.experts.119.gate_proj", "model.layers.38.mlp.experts.120.gate_proj", "model.layers.38.mlp.experts.121.gate_proj", "model.layers.38.mlp.experts.122.gate_proj", "model.layers.38.mlp.experts.123.gate_proj", "model.layers.38.mlp.experts.124.gate_proj", "model.layers.38.mlp.experts.125.gate_proj", "model.layers.38.mlp.experts.126.gate_proj", "model.layers.38.mlp.experts.127.gate_proj", "model.layers.38.mlp.experts.128.gate_proj", "model.layers.38.mlp.experts.129.gate_proj", "model.layers.38.mlp.experts.130.gate_proj", "model.layers.38.mlp.experts.131.gate_proj", "model.layers.38.mlp.experts.132.gate_proj", "model.layers.38.mlp.experts.133.gate_proj", "model.layers.38.mlp.experts.134.gate_proj", "model.layers.38.mlp.experts.135.gate_proj", "model.layers.38.mlp.experts.136.gate_proj", "model.layers.38.mlp.experts.137.gate_proj", "model.layers.38.mlp.experts.138.gate_proj", "model.layers.38.mlp.experts.139.gate_proj", "model.layers.38.mlp.experts.140.gate_proj", "model.layers.38.mlp.experts.141.gate_proj", "model.layers.38.mlp.experts.142.gate_proj", "model.layers.38.mlp.experts.143.gate_proj", "model.layers.38.mlp.experts.144.gate_proj", "model.layers.38.mlp.experts.145.gate_proj", "model.layers.38.mlp.experts.146.gate_proj", "model.layers.38.mlp.experts.147.gate_proj", "model.layers.38.mlp.experts.148.gate_proj", "model.layers.38.mlp.experts.149.gate_proj", "model.layers.38.mlp.experts.150.gate_proj", "model.layers.38.mlp.experts.151.gate_proj", "model.layers.38.mlp.experts.152.gate_proj", "model.layers.38.mlp.experts.153.gate_proj", "model.layers.38.mlp.experts.154.gate_proj", "model.layers.38.mlp.experts.155.gate_proj", "model.layers.38.mlp.experts.156.gate_proj", "model.layers.38.mlp.experts.157.gate_proj", "model.layers.38.mlp.experts.158.gate_proj", "model.layers.38.mlp.experts.159.gate_proj", "model.layers.38.mlp.experts.0.up_proj", "model.layers.38.mlp.experts.1.up_proj", "model.layers.38.mlp.experts.2.up_proj", "model.layers.38.mlp.experts.3.up_proj", "model.layers.38.mlp.experts.4.up_proj", "model.layers.38.mlp.experts.5.up_proj", "model.layers.38.mlp.experts.6.up_proj", "model.layers.38.mlp.experts.7.up_proj", "model.layers.38.mlp.experts.8.up_proj", "model.layers.38.mlp.experts.9.up_proj", "model.layers.38.mlp.experts.10.up_proj", "model.layers.38.mlp.experts.11.up_proj", "model.layers.38.mlp.experts.12.up_proj", "model.layers.38.mlp.experts.13.up_proj", "model.layers.38.mlp.experts.14.up_proj", "model.layers.38.mlp.experts.15.up_proj", "model.layers.38.mlp.experts.16.up_proj", "model.layers.38.mlp.experts.17.up_proj", "model.layers.38.mlp.experts.18.up_proj", "model.layers.38.mlp.experts.19.up_proj", "model.layers.38.mlp.experts.20.up_proj", "model.layers.38.mlp.experts.21.up_proj", "model.layers.38.mlp.experts.22.up_proj", "model.layers.38.mlp.experts.23.up_proj", "model.layers.38.mlp.experts.24.up_proj", "model.layers.38.mlp.experts.25.up_proj", "model.layers.38.mlp.experts.26.up_proj", "model.layers.38.mlp.experts.27.up_proj", "model.layers.38.mlp.experts.28.up_proj", "model.layers.38.mlp.experts.29.up_proj", "model.layers.38.mlp.experts.30.up_proj", "model.layers.38.mlp.experts.31.up_proj", "model.layers.38.mlp.experts.32.up_proj", "model.layers.38.mlp.experts.33.up_proj", "model.layers.38.mlp.experts.34.up_proj", "model.layers.38.mlp.experts.35.up_proj", "model.layers.38.mlp.experts.36.up_proj", "model.layers.38.mlp.experts.37.up_proj", "model.layers.38.mlp.experts.38.up_proj", "model.layers.38.mlp.experts.39.up_proj", "model.layers.38.mlp.experts.40.up_proj", "model.layers.38.mlp.experts.41.up_proj", "model.layers.38.mlp.experts.42.up_proj", "model.layers.38.mlp.experts.43.up_proj", "model.layers.38.mlp.experts.44.up_proj", "model.layers.38.mlp.experts.45.up_proj", "model.layers.38.mlp.experts.46.up_proj", "model.layers.38.mlp.experts.47.up_proj", "model.layers.38.mlp.experts.48.up_proj", "model.layers.38.mlp.experts.49.up_proj", "model.layers.38.mlp.experts.50.up_proj", "model.layers.38.mlp.experts.51.up_proj", "model.layers.38.mlp.experts.52.up_proj", "model.layers.38.mlp.experts.53.up_proj", "model.layers.38.mlp.experts.54.up_proj", "model.layers.38.mlp.experts.55.up_proj", "model.layers.38.mlp.experts.56.up_proj", "model.layers.38.mlp.experts.57.up_proj", "model.layers.38.mlp.experts.58.up_proj", "model.layers.38.mlp.experts.59.up_proj", "model.layers.38.mlp.experts.60.up_proj", "model.layers.38.mlp.experts.61.up_proj", "model.layers.38.mlp.experts.62.up_proj", "model.layers.38.mlp.experts.63.up_proj", "model.layers.38.mlp.experts.64.up_proj", "model.layers.38.mlp.experts.65.up_proj", "model.layers.38.mlp.experts.66.up_proj", "model.layers.38.mlp.experts.67.up_proj", "model.layers.38.mlp.experts.68.up_proj", "model.layers.38.mlp.experts.69.up_proj", "model.layers.38.mlp.experts.70.up_proj", "model.layers.38.mlp.experts.71.up_proj", "model.layers.38.mlp.experts.72.up_proj", "model.layers.38.mlp.experts.73.up_proj", "model.layers.38.mlp.experts.74.up_proj", "model.layers.38.mlp.experts.75.up_proj", "model.layers.38.mlp.experts.76.up_proj", "model.layers.38.mlp.experts.77.up_proj", "model.layers.38.mlp.experts.78.up_proj", "model.layers.38.mlp.experts.79.up_proj", "model.layers.38.mlp.experts.80.up_proj", "model.layers.38.mlp.experts.81.up_proj", "model.layers.38.mlp.experts.82.up_proj", "model.layers.38.mlp.experts.83.up_proj", "model.layers.38.mlp.experts.84.up_proj", "model.layers.38.mlp.experts.85.up_proj", "model.layers.38.mlp.experts.86.up_proj", "model.layers.38.mlp.experts.87.up_proj", "model.layers.38.mlp.experts.88.up_proj", "model.layers.38.mlp.experts.89.up_proj", "model.layers.38.mlp.experts.90.up_proj", "model.layers.38.mlp.experts.91.up_proj", "model.layers.38.mlp.experts.92.up_proj", "model.layers.38.mlp.experts.93.up_proj", "model.layers.38.mlp.experts.94.up_proj", "model.layers.38.mlp.experts.95.up_proj", "model.layers.38.mlp.experts.96.up_proj", "model.layers.38.mlp.experts.97.up_proj", "model.layers.38.mlp.experts.98.up_proj", "model.layers.38.mlp.experts.99.up_proj", "model.layers.38.mlp.experts.100.up_proj", "model.layers.38.mlp.experts.101.up_proj", "model.layers.38.mlp.experts.102.up_proj", "model.layers.38.mlp.experts.103.up_proj", "model.layers.38.mlp.experts.104.up_proj", "model.layers.38.mlp.experts.105.up_proj", "model.layers.38.mlp.experts.106.up_proj", "model.layers.38.mlp.experts.107.up_proj", "model.layers.38.mlp.experts.108.up_proj", "model.layers.38.mlp.experts.109.up_proj", "model.layers.38.mlp.experts.110.up_proj", "model.layers.38.mlp.experts.111.up_proj", "model.layers.38.mlp.experts.112.up_proj", "model.layers.38.mlp.experts.113.up_proj", "model.layers.38.mlp.experts.114.up_proj", "model.layers.38.mlp.experts.115.up_proj", "model.layers.38.mlp.experts.116.up_proj", "model.layers.38.mlp.experts.117.up_proj", "model.layers.38.mlp.experts.118.up_proj", "model.layers.38.mlp.experts.119.up_proj", "model.layers.38.mlp.experts.120.up_proj", "model.layers.38.mlp.experts.121.up_proj", "model.layers.38.mlp.experts.122.up_proj", "model.layers.38.mlp.experts.123.up_proj", "model.layers.38.mlp.experts.124.up_proj", "model.layers.38.mlp.experts.125.up_proj", "model.layers.38.mlp.experts.126.up_proj", "model.layers.38.mlp.experts.127.up_proj", "model.layers.38.mlp.experts.128.up_proj", "model.layers.38.mlp.experts.129.up_proj", "model.layers.38.mlp.experts.130.up_proj", "model.layers.38.mlp.experts.131.up_proj", "model.layers.38.mlp.experts.132.up_proj", "model.layers.38.mlp.experts.133.up_proj", "model.layers.38.mlp.experts.134.up_proj", "model.layers.38.mlp.experts.135.up_proj", "model.layers.38.mlp.experts.136.up_proj", "model.layers.38.mlp.experts.137.up_proj", "model.layers.38.mlp.experts.138.up_proj", "model.layers.38.mlp.experts.139.up_proj", "model.layers.38.mlp.experts.140.up_proj", "model.layers.38.mlp.experts.141.up_proj", "model.layers.38.mlp.experts.142.up_proj", "model.layers.38.mlp.experts.143.up_proj", "model.layers.38.mlp.experts.144.up_proj", "model.layers.38.mlp.experts.145.up_proj", "model.layers.38.mlp.experts.146.up_proj", "model.layers.38.mlp.experts.147.up_proj", "model.layers.38.mlp.experts.148.up_proj", "model.layers.38.mlp.experts.149.up_proj", "model.layers.38.mlp.experts.150.up_proj", "model.layers.38.mlp.experts.151.up_proj", "model.layers.38.mlp.experts.152.up_proj", "model.layers.38.mlp.experts.153.up_proj", "model.layers.38.mlp.experts.154.up_proj", "model.layers.38.mlp.experts.155.up_proj", "model.layers.38.mlp.experts.156.up_proj", "model.layers.38.mlp.experts.157.up_proj", "model.layers.38.mlp.experts.158.up_proj", "model.layers.38.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00039911502972245216, "dbits": 2516582400 }, { "dkld": -0.0006131736561656026, "dbits": 5033164800 }, { "dkld": -0.0006256952881813077, "dbits": 7549747200 }, { "dkld": -0.0005977191030979156, "dbits": 12582912000 } ] }, { "idx": 230, "layers": [ "model.layers.38.mlp.experts.0.down_proj", "model.layers.38.mlp.experts.1.down_proj", "model.layers.38.mlp.experts.2.down_proj", "model.layers.38.mlp.experts.3.down_proj", "model.layers.38.mlp.experts.4.down_proj", "model.layers.38.mlp.experts.5.down_proj", "model.layers.38.mlp.experts.6.down_proj", "model.layers.38.mlp.experts.7.down_proj", "model.layers.38.mlp.experts.8.down_proj", "model.layers.38.mlp.experts.9.down_proj", "model.layers.38.mlp.experts.10.down_proj", "model.layers.38.mlp.experts.11.down_proj", "model.layers.38.mlp.experts.12.down_proj", "model.layers.38.mlp.experts.13.down_proj", "model.layers.38.mlp.experts.14.down_proj", "model.layers.38.mlp.experts.15.down_proj", "model.layers.38.mlp.experts.16.down_proj", "model.layers.38.mlp.experts.17.down_proj", "model.layers.38.mlp.experts.18.down_proj", "model.layers.38.mlp.experts.19.down_proj", "model.layers.38.mlp.experts.20.down_proj", "model.layers.38.mlp.experts.21.down_proj", "model.layers.38.mlp.experts.22.down_proj", "model.layers.38.mlp.experts.23.down_proj", "model.layers.38.mlp.experts.24.down_proj", "model.layers.38.mlp.experts.25.down_proj", "model.layers.38.mlp.experts.26.down_proj", "model.layers.38.mlp.experts.27.down_proj", "model.layers.38.mlp.experts.28.down_proj", "model.layers.38.mlp.experts.29.down_proj", "model.layers.38.mlp.experts.30.down_proj", "model.layers.38.mlp.experts.31.down_proj", "model.layers.38.mlp.experts.32.down_proj", "model.layers.38.mlp.experts.33.down_proj", "model.layers.38.mlp.experts.34.down_proj", "model.layers.38.mlp.experts.35.down_proj", "model.layers.38.mlp.experts.36.down_proj", "model.layers.38.mlp.experts.37.down_proj", "model.layers.38.mlp.experts.38.down_proj", "model.layers.38.mlp.experts.39.down_proj", "model.layers.38.mlp.experts.40.down_proj", "model.layers.38.mlp.experts.41.down_proj", "model.layers.38.mlp.experts.42.down_proj", "model.layers.38.mlp.experts.43.down_proj", "model.layers.38.mlp.experts.44.down_proj", "model.layers.38.mlp.experts.45.down_proj", "model.layers.38.mlp.experts.46.down_proj", "model.layers.38.mlp.experts.47.down_proj", "model.layers.38.mlp.experts.48.down_proj", "model.layers.38.mlp.experts.49.down_proj", "model.layers.38.mlp.experts.50.down_proj", "model.layers.38.mlp.experts.51.down_proj", "model.layers.38.mlp.experts.52.down_proj", "model.layers.38.mlp.experts.53.down_proj", "model.layers.38.mlp.experts.54.down_proj", "model.layers.38.mlp.experts.55.down_proj", "model.layers.38.mlp.experts.56.down_proj", "model.layers.38.mlp.experts.57.down_proj", "model.layers.38.mlp.experts.58.down_proj", "model.layers.38.mlp.experts.59.down_proj", "model.layers.38.mlp.experts.60.down_proj", "model.layers.38.mlp.experts.61.down_proj", "model.layers.38.mlp.experts.62.down_proj", "model.layers.38.mlp.experts.63.down_proj", "model.layers.38.mlp.experts.64.down_proj", "model.layers.38.mlp.experts.65.down_proj", "model.layers.38.mlp.experts.66.down_proj", "model.layers.38.mlp.experts.67.down_proj", "model.layers.38.mlp.experts.68.down_proj", "model.layers.38.mlp.experts.69.down_proj", "model.layers.38.mlp.experts.70.down_proj", "model.layers.38.mlp.experts.71.down_proj", "model.layers.38.mlp.experts.72.down_proj", "model.layers.38.mlp.experts.73.down_proj", "model.layers.38.mlp.experts.74.down_proj", "model.layers.38.mlp.experts.75.down_proj", "model.layers.38.mlp.experts.76.down_proj", "model.layers.38.mlp.experts.77.down_proj", "model.layers.38.mlp.experts.78.down_proj", "model.layers.38.mlp.experts.79.down_proj", "model.layers.38.mlp.experts.80.down_proj", "model.layers.38.mlp.experts.81.down_proj", "model.layers.38.mlp.experts.82.down_proj", "model.layers.38.mlp.experts.83.down_proj", "model.layers.38.mlp.experts.84.down_proj", "model.layers.38.mlp.experts.85.down_proj", "model.layers.38.mlp.experts.86.down_proj", "model.layers.38.mlp.experts.87.down_proj", "model.layers.38.mlp.experts.88.down_proj", "model.layers.38.mlp.experts.89.down_proj", "model.layers.38.mlp.experts.90.down_proj", "model.layers.38.mlp.experts.91.down_proj", "model.layers.38.mlp.experts.92.down_proj", "model.layers.38.mlp.experts.93.down_proj", "model.layers.38.mlp.experts.94.down_proj", "model.layers.38.mlp.experts.95.down_proj", "model.layers.38.mlp.experts.96.down_proj", "model.layers.38.mlp.experts.97.down_proj", "model.layers.38.mlp.experts.98.down_proj", "model.layers.38.mlp.experts.99.down_proj", "model.layers.38.mlp.experts.100.down_proj", "model.layers.38.mlp.experts.101.down_proj", "model.layers.38.mlp.experts.102.down_proj", "model.layers.38.mlp.experts.103.down_proj", "model.layers.38.mlp.experts.104.down_proj", "model.layers.38.mlp.experts.105.down_proj", "model.layers.38.mlp.experts.106.down_proj", "model.layers.38.mlp.experts.107.down_proj", "model.layers.38.mlp.experts.108.down_proj", "model.layers.38.mlp.experts.109.down_proj", "model.layers.38.mlp.experts.110.down_proj", "model.layers.38.mlp.experts.111.down_proj", "model.layers.38.mlp.experts.112.down_proj", "model.layers.38.mlp.experts.113.down_proj", "model.layers.38.mlp.experts.114.down_proj", "model.layers.38.mlp.experts.115.down_proj", "model.layers.38.mlp.experts.116.down_proj", "model.layers.38.mlp.experts.117.down_proj", "model.layers.38.mlp.experts.118.down_proj", "model.layers.38.mlp.experts.119.down_proj", "model.layers.38.mlp.experts.120.down_proj", "model.layers.38.mlp.experts.121.down_proj", "model.layers.38.mlp.experts.122.down_proj", "model.layers.38.mlp.experts.123.down_proj", "model.layers.38.mlp.experts.124.down_proj", "model.layers.38.mlp.experts.125.down_proj", "model.layers.38.mlp.experts.126.down_proj", "model.layers.38.mlp.experts.127.down_proj", "model.layers.38.mlp.experts.128.down_proj", "model.layers.38.mlp.experts.129.down_proj", "model.layers.38.mlp.experts.130.down_proj", "model.layers.38.mlp.experts.131.down_proj", "model.layers.38.mlp.experts.132.down_proj", "model.layers.38.mlp.experts.133.down_proj", "model.layers.38.mlp.experts.134.down_proj", "model.layers.38.mlp.experts.135.down_proj", "model.layers.38.mlp.experts.136.down_proj", "model.layers.38.mlp.experts.137.down_proj", "model.layers.38.mlp.experts.138.down_proj", "model.layers.38.mlp.experts.139.down_proj", "model.layers.38.mlp.experts.140.down_proj", "model.layers.38.mlp.experts.141.down_proj", "model.layers.38.mlp.experts.142.down_proj", "model.layers.38.mlp.experts.143.down_proj", "model.layers.38.mlp.experts.144.down_proj", "model.layers.38.mlp.experts.145.down_proj", "model.layers.38.mlp.experts.146.down_proj", "model.layers.38.mlp.experts.147.down_proj", "model.layers.38.mlp.experts.148.down_proj", "model.layers.38.mlp.experts.149.down_proj", "model.layers.38.mlp.experts.150.down_proj", "model.layers.38.mlp.experts.151.down_proj", "model.layers.38.mlp.experts.152.down_proj", "model.layers.38.mlp.experts.153.down_proj", "model.layers.38.mlp.experts.154.down_proj", "model.layers.38.mlp.experts.155.down_proj", "model.layers.38.mlp.experts.156.down_proj", "model.layers.38.mlp.experts.157.down_proj", "model.layers.38.mlp.experts.158.down_proj", "model.layers.38.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002757051959633855, "dbits": 1258291200 }, { "dkld": -0.0003402917645871639, "dbits": 2516582400 }, { "dkld": -0.00039457120001316903, "dbits": 3774873600 }, { "dkld": -0.0004309244453907096, "dbits": 6291456000 } ] }, { "idx": 231, "layers": [ "model.layers.39.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00045987237244844714, "dbits": 62914560 }, { "dkld": -0.0005174989812076203, "dbits": 125829120 }, { "dkld": -0.0005174610763788307, "dbits": 188743680 }, { "dkld": -0.0005058700218796841, "dbits": 314572800 } ] }, { "idx": 232, "layers": [ "model.layers.39.self_attn.k_proj", "model.layers.39.self_attn.v_proj" ], "candidates": [ { "dkld": 0.000500949006527654, "dbits": 10485760 }, { "dkld": -0.00046738050878047943, "dbits": 20971520 }, { "dkld": -0.0006678001023829039, "dbits": 31457280 }, { "dkld": -0.0005224491469562054, "dbits": 52428800 } ] }, { "idx": 233, "layers": [ "model.layers.39.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010040517896413859, "dbits": 62914560 }, { "dkld": -0.00037750862538815655, "dbits": 125829120 }, { "dkld": -0.0006023826077580507, "dbits": 188743680 }, { "dkld": -0.000533485133200895, "dbits": 314572800 } ] }, { "idx": 234, "layers": [ "model.layers.39.mlp.shared_experts.gate_proj", "model.layers.39.mlp.shared_experts.up_proj", "model.layers.39.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0008654562756419265, "dbits": 23592960 }, { "dkld": -0.0010056192055344637, "dbits": 47185920 }, { "dkld": -0.0010918699204921778, "dbits": 70778880 }, { "dkld": -0.0011028001084923827, "dbits": 117964800 } ] }, { "idx": 235, "layers": [ "model.layers.39.mlp.experts.0.gate_proj", "model.layers.39.mlp.experts.1.gate_proj", "model.layers.39.mlp.experts.2.gate_proj", "model.layers.39.mlp.experts.3.gate_proj", "model.layers.39.mlp.experts.4.gate_proj", "model.layers.39.mlp.experts.5.gate_proj", "model.layers.39.mlp.experts.6.gate_proj", "model.layers.39.mlp.experts.7.gate_proj", "model.layers.39.mlp.experts.8.gate_proj", "model.layers.39.mlp.experts.9.gate_proj", "model.layers.39.mlp.experts.10.gate_proj", "model.layers.39.mlp.experts.11.gate_proj", "model.layers.39.mlp.experts.12.gate_proj", "model.layers.39.mlp.experts.13.gate_proj", "model.layers.39.mlp.experts.14.gate_proj", "model.layers.39.mlp.experts.15.gate_proj", "model.layers.39.mlp.experts.16.gate_proj", "model.layers.39.mlp.experts.17.gate_proj", "model.layers.39.mlp.experts.18.gate_proj", "model.layers.39.mlp.experts.19.gate_proj", "model.layers.39.mlp.experts.20.gate_proj", "model.layers.39.mlp.experts.21.gate_proj", "model.layers.39.mlp.experts.22.gate_proj", "model.layers.39.mlp.experts.23.gate_proj", "model.layers.39.mlp.experts.24.gate_proj", "model.layers.39.mlp.experts.25.gate_proj", "model.layers.39.mlp.experts.26.gate_proj", "model.layers.39.mlp.experts.27.gate_proj", "model.layers.39.mlp.experts.28.gate_proj", "model.layers.39.mlp.experts.29.gate_proj", "model.layers.39.mlp.experts.30.gate_proj", "model.layers.39.mlp.experts.31.gate_proj", "model.layers.39.mlp.experts.32.gate_proj", "model.layers.39.mlp.experts.33.gate_proj", "model.layers.39.mlp.experts.34.gate_proj", "model.layers.39.mlp.experts.35.gate_proj", "model.layers.39.mlp.experts.36.gate_proj", "model.layers.39.mlp.experts.37.gate_proj", "model.layers.39.mlp.experts.38.gate_proj", "model.layers.39.mlp.experts.39.gate_proj", "model.layers.39.mlp.experts.40.gate_proj", "model.layers.39.mlp.experts.41.gate_proj", "model.layers.39.mlp.experts.42.gate_proj", "model.layers.39.mlp.experts.43.gate_proj", "model.layers.39.mlp.experts.44.gate_proj", "model.layers.39.mlp.experts.45.gate_proj", "model.layers.39.mlp.experts.46.gate_proj", "model.layers.39.mlp.experts.47.gate_proj", "model.layers.39.mlp.experts.48.gate_proj", "model.layers.39.mlp.experts.49.gate_proj", "model.layers.39.mlp.experts.50.gate_proj", "model.layers.39.mlp.experts.51.gate_proj", "model.layers.39.mlp.experts.52.gate_proj", "model.layers.39.mlp.experts.53.gate_proj", "model.layers.39.mlp.experts.54.gate_proj", "model.layers.39.mlp.experts.55.gate_proj", "model.layers.39.mlp.experts.56.gate_proj", "model.layers.39.mlp.experts.57.gate_proj", "model.layers.39.mlp.experts.58.gate_proj", "model.layers.39.mlp.experts.59.gate_proj", "model.layers.39.mlp.experts.60.gate_proj", "model.layers.39.mlp.experts.61.gate_proj", "model.layers.39.mlp.experts.62.gate_proj", "model.layers.39.mlp.experts.63.gate_proj", "model.layers.39.mlp.experts.64.gate_proj", "model.layers.39.mlp.experts.65.gate_proj", "model.layers.39.mlp.experts.66.gate_proj", "model.layers.39.mlp.experts.67.gate_proj", "model.layers.39.mlp.experts.68.gate_proj", "model.layers.39.mlp.experts.69.gate_proj", "model.layers.39.mlp.experts.70.gate_proj", "model.layers.39.mlp.experts.71.gate_proj", "model.layers.39.mlp.experts.72.gate_proj", "model.layers.39.mlp.experts.73.gate_proj", "model.layers.39.mlp.experts.74.gate_proj", "model.layers.39.mlp.experts.75.gate_proj", "model.layers.39.mlp.experts.76.gate_proj", "model.layers.39.mlp.experts.77.gate_proj", "model.layers.39.mlp.experts.78.gate_proj", "model.layers.39.mlp.experts.79.gate_proj", "model.layers.39.mlp.experts.80.gate_proj", "model.layers.39.mlp.experts.81.gate_proj", "model.layers.39.mlp.experts.82.gate_proj", "model.layers.39.mlp.experts.83.gate_proj", "model.layers.39.mlp.experts.84.gate_proj", "model.layers.39.mlp.experts.85.gate_proj", "model.layers.39.mlp.experts.86.gate_proj", "model.layers.39.mlp.experts.87.gate_proj", "model.layers.39.mlp.experts.88.gate_proj", "model.layers.39.mlp.experts.89.gate_proj", "model.layers.39.mlp.experts.90.gate_proj", "model.layers.39.mlp.experts.91.gate_proj", "model.layers.39.mlp.experts.92.gate_proj", "model.layers.39.mlp.experts.93.gate_proj", "model.layers.39.mlp.experts.94.gate_proj", "model.layers.39.mlp.experts.95.gate_proj", "model.layers.39.mlp.experts.96.gate_proj", "model.layers.39.mlp.experts.97.gate_proj", "model.layers.39.mlp.experts.98.gate_proj", "model.layers.39.mlp.experts.99.gate_proj", "model.layers.39.mlp.experts.100.gate_proj", "model.layers.39.mlp.experts.101.gate_proj", "model.layers.39.mlp.experts.102.gate_proj", "model.layers.39.mlp.experts.103.gate_proj", "model.layers.39.mlp.experts.104.gate_proj", "model.layers.39.mlp.experts.105.gate_proj", "model.layers.39.mlp.experts.106.gate_proj", "model.layers.39.mlp.experts.107.gate_proj", "model.layers.39.mlp.experts.108.gate_proj", "model.layers.39.mlp.experts.109.gate_proj", "model.layers.39.mlp.experts.110.gate_proj", "model.layers.39.mlp.experts.111.gate_proj", "model.layers.39.mlp.experts.112.gate_proj", "model.layers.39.mlp.experts.113.gate_proj", "model.layers.39.mlp.experts.114.gate_proj", "model.layers.39.mlp.experts.115.gate_proj", "model.layers.39.mlp.experts.116.gate_proj", "model.layers.39.mlp.experts.117.gate_proj", "model.layers.39.mlp.experts.118.gate_proj", "model.layers.39.mlp.experts.119.gate_proj", "model.layers.39.mlp.experts.120.gate_proj", "model.layers.39.mlp.experts.121.gate_proj", "model.layers.39.mlp.experts.122.gate_proj", "model.layers.39.mlp.experts.123.gate_proj", "model.layers.39.mlp.experts.124.gate_proj", "model.layers.39.mlp.experts.125.gate_proj", "model.layers.39.mlp.experts.126.gate_proj", "model.layers.39.mlp.experts.127.gate_proj", "model.layers.39.mlp.experts.128.gate_proj", "model.layers.39.mlp.experts.129.gate_proj", "model.layers.39.mlp.experts.130.gate_proj", "model.layers.39.mlp.experts.131.gate_proj", "model.layers.39.mlp.experts.132.gate_proj", "model.layers.39.mlp.experts.133.gate_proj", "model.layers.39.mlp.experts.134.gate_proj", "model.layers.39.mlp.experts.135.gate_proj", "model.layers.39.mlp.experts.136.gate_proj", "model.layers.39.mlp.experts.137.gate_proj", "model.layers.39.mlp.experts.138.gate_proj", "model.layers.39.mlp.experts.139.gate_proj", "model.layers.39.mlp.experts.140.gate_proj", "model.layers.39.mlp.experts.141.gate_proj", "model.layers.39.mlp.experts.142.gate_proj", "model.layers.39.mlp.experts.143.gate_proj", "model.layers.39.mlp.experts.144.gate_proj", "model.layers.39.mlp.experts.145.gate_proj", "model.layers.39.mlp.experts.146.gate_proj", "model.layers.39.mlp.experts.147.gate_proj", "model.layers.39.mlp.experts.148.gate_proj", "model.layers.39.mlp.experts.149.gate_proj", "model.layers.39.mlp.experts.150.gate_proj", "model.layers.39.mlp.experts.151.gate_proj", "model.layers.39.mlp.experts.152.gate_proj", "model.layers.39.mlp.experts.153.gate_proj", "model.layers.39.mlp.experts.154.gate_proj", "model.layers.39.mlp.experts.155.gate_proj", "model.layers.39.mlp.experts.156.gate_proj", "model.layers.39.mlp.experts.157.gate_proj", "model.layers.39.mlp.experts.158.gate_proj", "model.layers.39.mlp.experts.159.gate_proj", "model.layers.39.mlp.experts.0.up_proj", "model.layers.39.mlp.experts.1.up_proj", "model.layers.39.mlp.experts.2.up_proj", "model.layers.39.mlp.experts.3.up_proj", "model.layers.39.mlp.experts.4.up_proj", "model.layers.39.mlp.experts.5.up_proj", "model.layers.39.mlp.experts.6.up_proj", "model.layers.39.mlp.experts.7.up_proj", "model.layers.39.mlp.experts.8.up_proj", "model.layers.39.mlp.experts.9.up_proj", "model.layers.39.mlp.experts.10.up_proj", "model.layers.39.mlp.experts.11.up_proj", "model.layers.39.mlp.experts.12.up_proj", "model.layers.39.mlp.experts.13.up_proj", "model.layers.39.mlp.experts.14.up_proj", "model.layers.39.mlp.experts.15.up_proj", "model.layers.39.mlp.experts.16.up_proj", "model.layers.39.mlp.experts.17.up_proj", "model.layers.39.mlp.experts.18.up_proj", "model.layers.39.mlp.experts.19.up_proj", "model.layers.39.mlp.experts.20.up_proj", "model.layers.39.mlp.experts.21.up_proj", "model.layers.39.mlp.experts.22.up_proj", "model.layers.39.mlp.experts.23.up_proj", "model.layers.39.mlp.experts.24.up_proj", "model.layers.39.mlp.experts.25.up_proj", "model.layers.39.mlp.experts.26.up_proj", "model.layers.39.mlp.experts.27.up_proj", "model.layers.39.mlp.experts.28.up_proj", "model.layers.39.mlp.experts.29.up_proj", "model.layers.39.mlp.experts.30.up_proj", "model.layers.39.mlp.experts.31.up_proj", "model.layers.39.mlp.experts.32.up_proj", "model.layers.39.mlp.experts.33.up_proj", "model.layers.39.mlp.experts.34.up_proj", "model.layers.39.mlp.experts.35.up_proj", "model.layers.39.mlp.experts.36.up_proj", "model.layers.39.mlp.experts.37.up_proj", "model.layers.39.mlp.experts.38.up_proj", "model.layers.39.mlp.experts.39.up_proj", "model.layers.39.mlp.experts.40.up_proj", "model.layers.39.mlp.experts.41.up_proj", "model.layers.39.mlp.experts.42.up_proj", "model.layers.39.mlp.experts.43.up_proj", "model.layers.39.mlp.experts.44.up_proj", "model.layers.39.mlp.experts.45.up_proj", "model.layers.39.mlp.experts.46.up_proj", "model.layers.39.mlp.experts.47.up_proj", "model.layers.39.mlp.experts.48.up_proj", "model.layers.39.mlp.experts.49.up_proj", "model.layers.39.mlp.experts.50.up_proj", "model.layers.39.mlp.experts.51.up_proj", "model.layers.39.mlp.experts.52.up_proj", "model.layers.39.mlp.experts.53.up_proj", "model.layers.39.mlp.experts.54.up_proj", "model.layers.39.mlp.experts.55.up_proj", "model.layers.39.mlp.experts.56.up_proj", "model.layers.39.mlp.experts.57.up_proj", "model.layers.39.mlp.experts.58.up_proj", "model.layers.39.mlp.experts.59.up_proj", "model.layers.39.mlp.experts.60.up_proj", "model.layers.39.mlp.experts.61.up_proj", "model.layers.39.mlp.experts.62.up_proj", "model.layers.39.mlp.experts.63.up_proj", "model.layers.39.mlp.experts.64.up_proj", "model.layers.39.mlp.experts.65.up_proj", "model.layers.39.mlp.experts.66.up_proj", "model.layers.39.mlp.experts.67.up_proj", "model.layers.39.mlp.experts.68.up_proj", "model.layers.39.mlp.experts.69.up_proj", "model.layers.39.mlp.experts.70.up_proj", "model.layers.39.mlp.experts.71.up_proj", "model.layers.39.mlp.experts.72.up_proj", "model.layers.39.mlp.experts.73.up_proj", "model.layers.39.mlp.experts.74.up_proj", "model.layers.39.mlp.experts.75.up_proj", "model.layers.39.mlp.experts.76.up_proj", "model.layers.39.mlp.experts.77.up_proj", "model.layers.39.mlp.experts.78.up_proj", "model.layers.39.mlp.experts.79.up_proj", "model.layers.39.mlp.experts.80.up_proj", "model.layers.39.mlp.experts.81.up_proj", "model.layers.39.mlp.experts.82.up_proj", "model.layers.39.mlp.experts.83.up_proj", "model.layers.39.mlp.experts.84.up_proj", "model.layers.39.mlp.experts.85.up_proj", "model.layers.39.mlp.experts.86.up_proj", "model.layers.39.mlp.experts.87.up_proj", "model.layers.39.mlp.experts.88.up_proj", "model.layers.39.mlp.experts.89.up_proj", "model.layers.39.mlp.experts.90.up_proj", "model.layers.39.mlp.experts.91.up_proj", "model.layers.39.mlp.experts.92.up_proj", "model.layers.39.mlp.experts.93.up_proj", "model.layers.39.mlp.experts.94.up_proj", "model.layers.39.mlp.experts.95.up_proj", "model.layers.39.mlp.experts.96.up_proj", "model.layers.39.mlp.experts.97.up_proj", "model.layers.39.mlp.experts.98.up_proj", "model.layers.39.mlp.experts.99.up_proj", "model.layers.39.mlp.experts.100.up_proj", "model.layers.39.mlp.experts.101.up_proj", "model.layers.39.mlp.experts.102.up_proj", "model.layers.39.mlp.experts.103.up_proj", "model.layers.39.mlp.experts.104.up_proj", "model.layers.39.mlp.experts.105.up_proj", "model.layers.39.mlp.experts.106.up_proj", "model.layers.39.mlp.experts.107.up_proj", "model.layers.39.mlp.experts.108.up_proj", "model.layers.39.mlp.experts.109.up_proj", "model.layers.39.mlp.experts.110.up_proj", "model.layers.39.mlp.experts.111.up_proj", "model.layers.39.mlp.experts.112.up_proj", "model.layers.39.mlp.experts.113.up_proj", "model.layers.39.mlp.experts.114.up_proj", "model.layers.39.mlp.experts.115.up_proj", "model.layers.39.mlp.experts.116.up_proj", "model.layers.39.mlp.experts.117.up_proj", "model.layers.39.mlp.experts.118.up_proj", "model.layers.39.mlp.experts.119.up_proj", "model.layers.39.mlp.experts.120.up_proj", "model.layers.39.mlp.experts.121.up_proj", "model.layers.39.mlp.experts.122.up_proj", "model.layers.39.mlp.experts.123.up_proj", "model.layers.39.mlp.experts.124.up_proj", "model.layers.39.mlp.experts.125.up_proj", "model.layers.39.mlp.experts.126.up_proj", "model.layers.39.mlp.experts.127.up_proj", "model.layers.39.mlp.experts.128.up_proj", "model.layers.39.mlp.experts.129.up_proj", "model.layers.39.mlp.experts.130.up_proj", "model.layers.39.mlp.experts.131.up_proj", "model.layers.39.mlp.experts.132.up_proj", "model.layers.39.mlp.experts.133.up_proj", "model.layers.39.mlp.experts.134.up_proj", "model.layers.39.mlp.experts.135.up_proj", "model.layers.39.mlp.experts.136.up_proj", "model.layers.39.mlp.experts.137.up_proj", "model.layers.39.mlp.experts.138.up_proj", "model.layers.39.mlp.experts.139.up_proj", "model.layers.39.mlp.experts.140.up_proj", "model.layers.39.mlp.experts.141.up_proj", "model.layers.39.mlp.experts.142.up_proj", "model.layers.39.mlp.experts.143.up_proj", "model.layers.39.mlp.experts.144.up_proj", "model.layers.39.mlp.experts.145.up_proj", "model.layers.39.mlp.experts.146.up_proj", "model.layers.39.mlp.experts.147.up_proj", "model.layers.39.mlp.experts.148.up_proj", "model.layers.39.mlp.experts.149.up_proj", "model.layers.39.mlp.experts.150.up_proj", "model.layers.39.mlp.experts.151.up_proj", "model.layers.39.mlp.experts.152.up_proj", "model.layers.39.mlp.experts.153.up_proj", "model.layers.39.mlp.experts.154.up_proj", "model.layers.39.mlp.experts.155.up_proj", "model.layers.39.mlp.experts.156.up_proj", "model.layers.39.mlp.experts.157.up_proj", "model.layers.39.mlp.experts.158.up_proj", "model.layers.39.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0004459107294678688, "dbits": 2516582400 }, { "dkld": -0.0005747616291046226, "dbits": 5033164800 }, { "dkld": -0.0006434101611375864, "dbits": 7549747200 }, { "dkld": -0.000667092390358448, "dbits": 12582912000 } ] }, { "idx": 236, "layers": [ "model.layers.39.mlp.experts.0.down_proj", "model.layers.39.mlp.experts.1.down_proj", "model.layers.39.mlp.experts.2.down_proj", "model.layers.39.mlp.experts.3.down_proj", "model.layers.39.mlp.experts.4.down_proj", "model.layers.39.mlp.experts.5.down_proj", "model.layers.39.mlp.experts.6.down_proj", "model.layers.39.mlp.experts.7.down_proj", "model.layers.39.mlp.experts.8.down_proj", "model.layers.39.mlp.experts.9.down_proj", "model.layers.39.mlp.experts.10.down_proj", "model.layers.39.mlp.experts.11.down_proj", "model.layers.39.mlp.experts.12.down_proj", "model.layers.39.mlp.experts.13.down_proj", "model.layers.39.mlp.experts.14.down_proj", "model.layers.39.mlp.experts.15.down_proj", "model.layers.39.mlp.experts.16.down_proj", "model.layers.39.mlp.experts.17.down_proj", "model.layers.39.mlp.experts.18.down_proj", "model.layers.39.mlp.experts.19.down_proj", "model.layers.39.mlp.experts.20.down_proj", "model.layers.39.mlp.experts.21.down_proj", "model.layers.39.mlp.experts.22.down_proj", "model.layers.39.mlp.experts.23.down_proj", "model.layers.39.mlp.experts.24.down_proj", "model.layers.39.mlp.experts.25.down_proj", "model.layers.39.mlp.experts.26.down_proj", "model.layers.39.mlp.experts.27.down_proj", "model.layers.39.mlp.experts.28.down_proj", "model.layers.39.mlp.experts.29.down_proj", "model.layers.39.mlp.experts.30.down_proj", "model.layers.39.mlp.experts.31.down_proj", "model.layers.39.mlp.experts.32.down_proj", "model.layers.39.mlp.experts.33.down_proj", "model.layers.39.mlp.experts.34.down_proj", "model.layers.39.mlp.experts.35.down_proj", "model.layers.39.mlp.experts.36.down_proj", "model.layers.39.mlp.experts.37.down_proj", "model.layers.39.mlp.experts.38.down_proj", "model.layers.39.mlp.experts.39.down_proj", "model.layers.39.mlp.experts.40.down_proj", "model.layers.39.mlp.experts.41.down_proj", "model.layers.39.mlp.experts.42.down_proj", "model.layers.39.mlp.experts.43.down_proj", "model.layers.39.mlp.experts.44.down_proj", "model.layers.39.mlp.experts.45.down_proj", "model.layers.39.mlp.experts.46.down_proj", "model.layers.39.mlp.experts.47.down_proj", "model.layers.39.mlp.experts.48.down_proj", "model.layers.39.mlp.experts.49.down_proj", "model.layers.39.mlp.experts.50.down_proj", "model.layers.39.mlp.experts.51.down_proj", "model.layers.39.mlp.experts.52.down_proj", "model.layers.39.mlp.experts.53.down_proj", "model.layers.39.mlp.experts.54.down_proj", "model.layers.39.mlp.experts.55.down_proj", "model.layers.39.mlp.experts.56.down_proj", "model.layers.39.mlp.experts.57.down_proj", "model.layers.39.mlp.experts.58.down_proj", "model.layers.39.mlp.experts.59.down_proj", "model.layers.39.mlp.experts.60.down_proj", "model.layers.39.mlp.experts.61.down_proj", "model.layers.39.mlp.experts.62.down_proj", "model.layers.39.mlp.experts.63.down_proj", "model.layers.39.mlp.experts.64.down_proj", "model.layers.39.mlp.experts.65.down_proj", "model.layers.39.mlp.experts.66.down_proj", "model.layers.39.mlp.experts.67.down_proj", "model.layers.39.mlp.experts.68.down_proj", "model.layers.39.mlp.experts.69.down_proj", "model.layers.39.mlp.experts.70.down_proj", "model.layers.39.mlp.experts.71.down_proj", "model.layers.39.mlp.experts.72.down_proj", "model.layers.39.mlp.experts.73.down_proj", "model.layers.39.mlp.experts.74.down_proj", "model.layers.39.mlp.experts.75.down_proj", "model.layers.39.mlp.experts.76.down_proj", "model.layers.39.mlp.experts.77.down_proj", "model.layers.39.mlp.experts.78.down_proj", "model.layers.39.mlp.experts.79.down_proj", "model.layers.39.mlp.experts.80.down_proj", "model.layers.39.mlp.experts.81.down_proj", "model.layers.39.mlp.experts.82.down_proj", "model.layers.39.mlp.experts.83.down_proj", "model.layers.39.mlp.experts.84.down_proj", "model.layers.39.mlp.experts.85.down_proj", "model.layers.39.mlp.experts.86.down_proj", "model.layers.39.mlp.experts.87.down_proj", "model.layers.39.mlp.experts.88.down_proj", "model.layers.39.mlp.experts.89.down_proj", "model.layers.39.mlp.experts.90.down_proj", "model.layers.39.mlp.experts.91.down_proj", "model.layers.39.mlp.experts.92.down_proj", "model.layers.39.mlp.experts.93.down_proj", "model.layers.39.mlp.experts.94.down_proj", "model.layers.39.mlp.experts.95.down_proj", "model.layers.39.mlp.experts.96.down_proj", "model.layers.39.mlp.experts.97.down_proj", "model.layers.39.mlp.experts.98.down_proj", "model.layers.39.mlp.experts.99.down_proj", "model.layers.39.mlp.experts.100.down_proj", "model.layers.39.mlp.experts.101.down_proj", "model.layers.39.mlp.experts.102.down_proj", "model.layers.39.mlp.experts.103.down_proj", "model.layers.39.mlp.experts.104.down_proj", "model.layers.39.mlp.experts.105.down_proj", "model.layers.39.mlp.experts.106.down_proj", "model.layers.39.mlp.experts.107.down_proj", "model.layers.39.mlp.experts.108.down_proj", "model.layers.39.mlp.experts.109.down_proj", "model.layers.39.mlp.experts.110.down_proj", "model.layers.39.mlp.experts.111.down_proj", "model.layers.39.mlp.experts.112.down_proj", "model.layers.39.mlp.experts.113.down_proj", "model.layers.39.mlp.experts.114.down_proj", "model.layers.39.mlp.experts.115.down_proj", "model.layers.39.mlp.experts.116.down_proj", "model.layers.39.mlp.experts.117.down_proj", "model.layers.39.mlp.experts.118.down_proj", "model.layers.39.mlp.experts.119.down_proj", "model.layers.39.mlp.experts.120.down_proj", "model.layers.39.mlp.experts.121.down_proj", "model.layers.39.mlp.experts.122.down_proj", "model.layers.39.mlp.experts.123.down_proj", "model.layers.39.mlp.experts.124.down_proj", "model.layers.39.mlp.experts.125.down_proj", "model.layers.39.mlp.experts.126.down_proj", "model.layers.39.mlp.experts.127.down_proj", "model.layers.39.mlp.experts.128.down_proj", "model.layers.39.mlp.experts.129.down_proj", "model.layers.39.mlp.experts.130.down_proj", "model.layers.39.mlp.experts.131.down_proj", "model.layers.39.mlp.experts.132.down_proj", "model.layers.39.mlp.experts.133.down_proj", "model.layers.39.mlp.experts.134.down_proj", "model.layers.39.mlp.experts.135.down_proj", "model.layers.39.mlp.experts.136.down_proj", "model.layers.39.mlp.experts.137.down_proj", "model.layers.39.mlp.experts.138.down_proj", "model.layers.39.mlp.experts.139.down_proj", "model.layers.39.mlp.experts.140.down_proj", "model.layers.39.mlp.experts.141.down_proj", "model.layers.39.mlp.experts.142.down_proj", "model.layers.39.mlp.experts.143.down_proj", "model.layers.39.mlp.experts.144.down_proj", "model.layers.39.mlp.experts.145.down_proj", "model.layers.39.mlp.experts.146.down_proj", "model.layers.39.mlp.experts.147.down_proj", "model.layers.39.mlp.experts.148.down_proj", "model.layers.39.mlp.experts.149.down_proj", "model.layers.39.mlp.experts.150.down_proj", "model.layers.39.mlp.experts.151.down_proj", "model.layers.39.mlp.experts.152.down_proj", "model.layers.39.mlp.experts.153.down_proj", "model.layers.39.mlp.experts.154.down_proj", "model.layers.39.mlp.experts.155.down_proj", "model.layers.39.mlp.experts.156.down_proj", "model.layers.39.mlp.experts.157.down_proj", "model.layers.39.mlp.experts.158.down_proj", "model.layers.39.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00012647612020373622, "dbits": 1258291200 }, { "dkld": -0.0002773785963654518, "dbits": 2516582400 }, { "dkld": -0.00030912030488253195, "dbits": 3774873600 }, { "dkld": -0.00025822781026363373, "dbits": 6291456000 } ] }, { "idx": 237, "layers": [ "model.layers.40.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0005062971264123972, "dbits": 62914560 }, { "dkld": -0.00034074513241649784, "dbits": 125829120 }, { "dkld": -0.000381896831095227, "dbits": 188743680 }, { "dkld": -0.00031934306025505066, "dbits": 314572800 } ] }, { "idx": 238, "layers": [ "model.layers.40.self_attn.k_proj", "model.layers.40.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0008863955736160278, "dbits": 10485760 }, { "dkld": -0.0013430379331111908, "dbits": 20971520 }, { "dkld": -0.0009463824331760462, "dbits": 31457280 }, { "dkld": -0.0010756082832813346, "dbits": 52428800 } ] }, { "idx": 239, "layers": [ "model.layers.40.self_attn.o_proj" ], "candidates": [ { "dkld": 9.82427969574845e-05, "dbits": 62914560 }, { "dkld": -0.0002883719280362129, "dbits": 125829120 }, { "dkld": -0.0003503551706671715, "dbits": 188743680 }, { "dkld": -0.0003769028931856211, "dbits": 314572800 } ] }, { "idx": 240, "layers": [ "model.layers.40.mlp.shared_experts.gate_proj", "model.layers.40.mlp.shared_experts.up_proj", "model.layers.40.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005063112825155341, "dbits": 23592960 }, { "dkld": -0.0008984216488897856, "dbits": 47185920 }, { "dkld": -0.0008843015879392707, "dbits": 70778880 }, { "dkld": -0.0008671177551150405, "dbits": 117964800 } ] }, { "idx": 241, "layers": [ "model.layers.40.mlp.experts.0.gate_proj", "model.layers.40.mlp.experts.1.gate_proj", "model.layers.40.mlp.experts.2.gate_proj", "model.layers.40.mlp.experts.3.gate_proj", "model.layers.40.mlp.experts.4.gate_proj", "model.layers.40.mlp.experts.5.gate_proj", "model.layers.40.mlp.experts.6.gate_proj", "model.layers.40.mlp.experts.7.gate_proj", "model.layers.40.mlp.experts.8.gate_proj", "model.layers.40.mlp.experts.9.gate_proj", "model.layers.40.mlp.experts.10.gate_proj", "model.layers.40.mlp.experts.11.gate_proj", "model.layers.40.mlp.experts.12.gate_proj", "model.layers.40.mlp.experts.13.gate_proj", "model.layers.40.mlp.experts.14.gate_proj", "model.layers.40.mlp.experts.15.gate_proj", "model.layers.40.mlp.experts.16.gate_proj", "model.layers.40.mlp.experts.17.gate_proj", "model.layers.40.mlp.experts.18.gate_proj", "model.layers.40.mlp.experts.19.gate_proj", "model.layers.40.mlp.experts.20.gate_proj", "model.layers.40.mlp.experts.21.gate_proj", "model.layers.40.mlp.experts.22.gate_proj", "model.layers.40.mlp.experts.23.gate_proj", "model.layers.40.mlp.experts.24.gate_proj", "model.layers.40.mlp.experts.25.gate_proj", "model.layers.40.mlp.experts.26.gate_proj", "model.layers.40.mlp.experts.27.gate_proj", "model.layers.40.mlp.experts.28.gate_proj", "model.layers.40.mlp.experts.29.gate_proj", "model.layers.40.mlp.experts.30.gate_proj", "model.layers.40.mlp.experts.31.gate_proj", "model.layers.40.mlp.experts.32.gate_proj", "model.layers.40.mlp.experts.33.gate_proj", "model.layers.40.mlp.experts.34.gate_proj", "model.layers.40.mlp.experts.35.gate_proj", "model.layers.40.mlp.experts.36.gate_proj", "model.layers.40.mlp.experts.37.gate_proj", "model.layers.40.mlp.experts.38.gate_proj", "model.layers.40.mlp.experts.39.gate_proj", "model.layers.40.mlp.experts.40.gate_proj", "model.layers.40.mlp.experts.41.gate_proj", "model.layers.40.mlp.experts.42.gate_proj", "model.layers.40.mlp.experts.43.gate_proj", "model.layers.40.mlp.experts.44.gate_proj", "model.layers.40.mlp.experts.45.gate_proj", "model.layers.40.mlp.experts.46.gate_proj", "model.layers.40.mlp.experts.47.gate_proj", "model.layers.40.mlp.experts.48.gate_proj", "model.layers.40.mlp.experts.49.gate_proj", "model.layers.40.mlp.experts.50.gate_proj", "model.layers.40.mlp.experts.51.gate_proj", "model.layers.40.mlp.experts.52.gate_proj", "model.layers.40.mlp.experts.53.gate_proj", "model.layers.40.mlp.experts.54.gate_proj", "model.layers.40.mlp.experts.55.gate_proj", "model.layers.40.mlp.experts.56.gate_proj", "model.layers.40.mlp.experts.57.gate_proj", "model.layers.40.mlp.experts.58.gate_proj", "model.layers.40.mlp.experts.59.gate_proj", "model.layers.40.mlp.experts.60.gate_proj", "model.layers.40.mlp.experts.61.gate_proj", "model.layers.40.mlp.experts.62.gate_proj", "model.layers.40.mlp.experts.63.gate_proj", "model.layers.40.mlp.experts.64.gate_proj", "model.layers.40.mlp.experts.65.gate_proj", "model.layers.40.mlp.experts.66.gate_proj", "model.layers.40.mlp.experts.67.gate_proj", "model.layers.40.mlp.experts.68.gate_proj", "model.layers.40.mlp.experts.69.gate_proj", "model.layers.40.mlp.experts.70.gate_proj", "model.layers.40.mlp.experts.71.gate_proj", "model.layers.40.mlp.experts.72.gate_proj", "model.layers.40.mlp.experts.73.gate_proj", "model.layers.40.mlp.experts.74.gate_proj", "model.layers.40.mlp.experts.75.gate_proj", "model.layers.40.mlp.experts.76.gate_proj", "model.layers.40.mlp.experts.77.gate_proj", "model.layers.40.mlp.experts.78.gate_proj", "model.layers.40.mlp.experts.79.gate_proj", "model.layers.40.mlp.experts.80.gate_proj", "model.layers.40.mlp.experts.81.gate_proj", "model.layers.40.mlp.experts.82.gate_proj", "model.layers.40.mlp.experts.83.gate_proj", "model.layers.40.mlp.experts.84.gate_proj", "model.layers.40.mlp.experts.85.gate_proj", "model.layers.40.mlp.experts.86.gate_proj", "model.layers.40.mlp.experts.87.gate_proj", "model.layers.40.mlp.experts.88.gate_proj", "model.layers.40.mlp.experts.89.gate_proj", "model.layers.40.mlp.experts.90.gate_proj", "model.layers.40.mlp.experts.91.gate_proj", "model.layers.40.mlp.experts.92.gate_proj", "model.layers.40.mlp.experts.93.gate_proj", "model.layers.40.mlp.experts.94.gate_proj", "model.layers.40.mlp.experts.95.gate_proj", "model.layers.40.mlp.experts.96.gate_proj", "model.layers.40.mlp.experts.97.gate_proj", "model.layers.40.mlp.experts.98.gate_proj", "model.layers.40.mlp.experts.99.gate_proj", "model.layers.40.mlp.experts.100.gate_proj", "model.layers.40.mlp.experts.101.gate_proj", "model.layers.40.mlp.experts.102.gate_proj", "model.layers.40.mlp.experts.103.gate_proj", "model.layers.40.mlp.experts.104.gate_proj", "model.layers.40.mlp.experts.105.gate_proj", "model.layers.40.mlp.experts.106.gate_proj", "model.layers.40.mlp.experts.107.gate_proj", "model.layers.40.mlp.experts.108.gate_proj", "model.layers.40.mlp.experts.109.gate_proj", "model.layers.40.mlp.experts.110.gate_proj", "model.layers.40.mlp.experts.111.gate_proj", "model.layers.40.mlp.experts.112.gate_proj", "model.layers.40.mlp.experts.113.gate_proj", "model.layers.40.mlp.experts.114.gate_proj", "model.layers.40.mlp.experts.115.gate_proj", "model.layers.40.mlp.experts.116.gate_proj", "model.layers.40.mlp.experts.117.gate_proj", "model.layers.40.mlp.experts.118.gate_proj", "model.layers.40.mlp.experts.119.gate_proj", "model.layers.40.mlp.experts.120.gate_proj", "model.layers.40.mlp.experts.121.gate_proj", "model.layers.40.mlp.experts.122.gate_proj", "model.layers.40.mlp.experts.123.gate_proj", "model.layers.40.mlp.experts.124.gate_proj", "model.layers.40.mlp.experts.125.gate_proj", "model.layers.40.mlp.experts.126.gate_proj", "model.layers.40.mlp.experts.127.gate_proj", "model.layers.40.mlp.experts.128.gate_proj", "model.layers.40.mlp.experts.129.gate_proj", "model.layers.40.mlp.experts.130.gate_proj", "model.layers.40.mlp.experts.131.gate_proj", "model.layers.40.mlp.experts.132.gate_proj", "model.layers.40.mlp.experts.133.gate_proj", "model.layers.40.mlp.experts.134.gate_proj", "model.layers.40.mlp.experts.135.gate_proj", "model.layers.40.mlp.experts.136.gate_proj", "model.layers.40.mlp.experts.137.gate_proj", "model.layers.40.mlp.experts.138.gate_proj", "model.layers.40.mlp.experts.139.gate_proj", "model.layers.40.mlp.experts.140.gate_proj", "model.layers.40.mlp.experts.141.gate_proj", "model.layers.40.mlp.experts.142.gate_proj", "model.layers.40.mlp.experts.143.gate_proj", "model.layers.40.mlp.experts.144.gate_proj", "model.layers.40.mlp.experts.145.gate_proj", "model.layers.40.mlp.experts.146.gate_proj", "model.layers.40.mlp.experts.147.gate_proj", "model.layers.40.mlp.experts.148.gate_proj", "model.layers.40.mlp.experts.149.gate_proj", "model.layers.40.mlp.experts.150.gate_proj", "model.layers.40.mlp.experts.151.gate_proj", "model.layers.40.mlp.experts.152.gate_proj", "model.layers.40.mlp.experts.153.gate_proj", "model.layers.40.mlp.experts.154.gate_proj", "model.layers.40.mlp.experts.155.gate_proj", "model.layers.40.mlp.experts.156.gate_proj", "model.layers.40.mlp.experts.157.gate_proj", "model.layers.40.mlp.experts.158.gate_proj", "model.layers.40.mlp.experts.159.gate_proj", "model.layers.40.mlp.experts.0.up_proj", "model.layers.40.mlp.experts.1.up_proj", "model.layers.40.mlp.experts.2.up_proj", "model.layers.40.mlp.experts.3.up_proj", "model.layers.40.mlp.experts.4.up_proj", "model.layers.40.mlp.experts.5.up_proj", "model.layers.40.mlp.experts.6.up_proj", "model.layers.40.mlp.experts.7.up_proj", "model.layers.40.mlp.experts.8.up_proj", "model.layers.40.mlp.experts.9.up_proj", "model.layers.40.mlp.experts.10.up_proj", "model.layers.40.mlp.experts.11.up_proj", "model.layers.40.mlp.experts.12.up_proj", "model.layers.40.mlp.experts.13.up_proj", "model.layers.40.mlp.experts.14.up_proj", "model.layers.40.mlp.experts.15.up_proj", "model.layers.40.mlp.experts.16.up_proj", "model.layers.40.mlp.experts.17.up_proj", "model.layers.40.mlp.experts.18.up_proj", "model.layers.40.mlp.experts.19.up_proj", "model.layers.40.mlp.experts.20.up_proj", "model.layers.40.mlp.experts.21.up_proj", "model.layers.40.mlp.experts.22.up_proj", "model.layers.40.mlp.experts.23.up_proj", "model.layers.40.mlp.experts.24.up_proj", "model.layers.40.mlp.experts.25.up_proj", "model.layers.40.mlp.experts.26.up_proj", "model.layers.40.mlp.experts.27.up_proj", "model.layers.40.mlp.experts.28.up_proj", "model.layers.40.mlp.experts.29.up_proj", "model.layers.40.mlp.experts.30.up_proj", "model.layers.40.mlp.experts.31.up_proj", "model.layers.40.mlp.experts.32.up_proj", "model.layers.40.mlp.experts.33.up_proj", "model.layers.40.mlp.experts.34.up_proj", "model.layers.40.mlp.experts.35.up_proj", "model.layers.40.mlp.experts.36.up_proj", "model.layers.40.mlp.experts.37.up_proj", "model.layers.40.mlp.experts.38.up_proj", "model.layers.40.mlp.experts.39.up_proj", "model.layers.40.mlp.experts.40.up_proj", "model.layers.40.mlp.experts.41.up_proj", "model.layers.40.mlp.experts.42.up_proj", "model.layers.40.mlp.experts.43.up_proj", "model.layers.40.mlp.experts.44.up_proj", "model.layers.40.mlp.experts.45.up_proj", "model.layers.40.mlp.experts.46.up_proj", "model.layers.40.mlp.experts.47.up_proj", "model.layers.40.mlp.experts.48.up_proj", "model.layers.40.mlp.experts.49.up_proj", "model.layers.40.mlp.experts.50.up_proj", "model.layers.40.mlp.experts.51.up_proj", "model.layers.40.mlp.experts.52.up_proj", "model.layers.40.mlp.experts.53.up_proj", "model.layers.40.mlp.experts.54.up_proj", "model.layers.40.mlp.experts.55.up_proj", "model.layers.40.mlp.experts.56.up_proj", "model.layers.40.mlp.experts.57.up_proj", "model.layers.40.mlp.experts.58.up_proj", "model.layers.40.mlp.experts.59.up_proj", "model.layers.40.mlp.experts.60.up_proj", "model.layers.40.mlp.experts.61.up_proj", "model.layers.40.mlp.experts.62.up_proj", "model.layers.40.mlp.experts.63.up_proj", "model.layers.40.mlp.experts.64.up_proj", "model.layers.40.mlp.experts.65.up_proj", "model.layers.40.mlp.experts.66.up_proj", "model.layers.40.mlp.experts.67.up_proj", "model.layers.40.mlp.experts.68.up_proj", "model.layers.40.mlp.experts.69.up_proj", "model.layers.40.mlp.experts.70.up_proj", "model.layers.40.mlp.experts.71.up_proj", "model.layers.40.mlp.experts.72.up_proj", "model.layers.40.mlp.experts.73.up_proj", "model.layers.40.mlp.experts.74.up_proj", "model.layers.40.mlp.experts.75.up_proj", "model.layers.40.mlp.experts.76.up_proj", "model.layers.40.mlp.experts.77.up_proj", "model.layers.40.mlp.experts.78.up_proj", "model.layers.40.mlp.experts.79.up_proj", "model.layers.40.mlp.experts.80.up_proj", "model.layers.40.mlp.experts.81.up_proj", "model.layers.40.mlp.experts.82.up_proj", "model.layers.40.mlp.experts.83.up_proj", "model.layers.40.mlp.experts.84.up_proj", "model.layers.40.mlp.experts.85.up_proj", "model.layers.40.mlp.experts.86.up_proj", "model.layers.40.mlp.experts.87.up_proj", "model.layers.40.mlp.experts.88.up_proj", "model.layers.40.mlp.experts.89.up_proj", "model.layers.40.mlp.experts.90.up_proj", "model.layers.40.mlp.experts.91.up_proj", "model.layers.40.mlp.experts.92.up_proj", "model.layers.40.mlp.experts.93.up_proj", "model.layers.40.mlp.experts.94.up_proj", "model.layers.40.mlp.experts.95.up_proj", "model.layers.40.mlp.experts.96.up_proj", "model.layers.40.mlp.experts.97.up_proj", "model.layers.40.mlp.experts.98.up_proj", "model.layers.40.mlp.experts.99.up_proj", "model.layers.40.mlp.experts.100.up_proj", "model.layers.40.mlp.experts.101.up_proj", "model.layers.40.mlp.experts.102.up_proj", "model.layers.40.mlp.experts.103.up_proj", "model.layers.40.mlp.experts.104.up_proj", "model.layers.40.mlp.experts.105.up_proj", "model.layers.40.mlp.experts.106.up_proj", "model.layers.40.mlp.experts.107.up_proj", "model.layers.40.mlp.experts.108.up_proj", "model.layers.40.mlp.experts.109.up_proj", "model.layers.40.mlp.experts.110.up_proj", "model.layers.40.mlp.experts.111.up_proj", "model.layers.40.mlp.experts.112.up_proj", "model.layers.40.mlp.experts.113.up_proj", "model.layers.40.mlp.experts.114.up_proj", "model.layers.40.mlp.experts.115.up_proj", "model.layers.40.mlp.experts.116.up_proj", "model.layers.40.mlp.experts.117.up_proj", "model.layers.40.mlp.experts.118.up_proj", "model.layers.40.mlp.experts.119.up_proj", "model.layers.40.mlp.experts.120.up_proj", "model.layers.40.mlp.experts.121.up_proj", "model.layers.40.mlp.experts.122.up_proj", "model.layers.40.mlp.experts.123.up_proj", "model.layers.40.mlp.experts.124.up_proj", "model.layers.40.mlp.experts.125.up_proj", "model.layers.40.mlp.experts.126.up_proj", "model.layers.40.mlp.experts.127.up_proj", "model.layers.40.mlp.experts.128.up_proj", "model.layers.40.mlp.experts.129.up_proj", "model.layers.40.mlp.experts.130.up_proj", "model.layers.40.mlp.experts.131.up_proj", "model.layers.40.mlp.experts.132.up_proj", "model.layers.40.mlp.experts.133.up_proj", "model.layers.40.mlp.experts.134.up_proj", "model.layers.40.mlp.experts.135.up_proj", "model.layers.40.mlp.experts.136.up_proj", "model.layers.40.mlp.experts.137.up_proj", "model.layers.40.mlp.experts.138.up_proj", "model.layers.40.mlp.experts.139.up_proj", "model.layers.40.mlp.experts.140.up_proj", "model.layers.40.mlp.experts.141.up_proj", "model.layers.40.mlp.experts.142.up_proj", "model.layers.40.mlp.experts.143.up_proj", "model.layers.40.mlp.experts.144.up_proj", "model.layers.40.mlp.experts.145.up_proj", "model.layers.40.mlp.experts.146.up_proj", "model.layers.40.mlp.experts.147.up_proj", "model.layers.40.mlp.experts.148.up_proj", "model.layers.40.mlp.experts.149.up_proj", "model.layers.40.mlp.experts.150.up_proj", "model.layers.40.mlp.experts.151.up_proj", "model.layers.40.mlp.experts.152.up_proj", "model.layers.40.mlp.experts.153.up_proj", "model.layers.40.mlp.experts.154.up_proj", "model.layers.40.mlp.experts.155.up_proj", "model.layers.40.mlp.experts.156.up_proj", "model.layers.40.mlp.experts.157.up_proj", "model.layers.40.mlp.experts.158.up_proj", "model.layers.40.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0005510923452675398, "dbits": 2516582400 }, { "dkld": -0.0006513481959700695, "dbits": 5033164800 }, { "dkld": -0.0005945719778537861, "dbits": 7549747200 }, { "dkld": -0.0005785561166703757, "dbits": 12582912000 } ] }, { "idx": 242, "layers": [ "model.layers.40.mlp.experts.0.down_proj", "model.layers.40.mlp.experts.1.down_proj", "model.layers.40.mlp.experts.2.down_proj", "model.layers.40.mlp.experts.3.down_proj", "model.layers.40.mlp.experts.4.down_proj", "model.layers.40.mlp.experts.5.down_proj", "model.layers.40.mlp.experts.6.down_proj", "model.layers.40.mlp.experts.7.down_proj", "model.layers.40.mlp.experts.8.down_proj", "model.layers.40.mlp.experts.9.down_proj", "model.layers.40.mlp.experts.10.down_proj", "model.layers.40.mlp.experts.11.down_proj", "model.layers.40.mlp.experts.12.down_proj", "model.layers.40.mlp.experts.13.down_proj", "model.layers.40.mlp.experts.14.down_proj", "model.layers.40.mlp.experts.15.down_proj", "model.layers.40.mlp.experts.16.down_proj", "model.layers.40.mlp.experts.17.down_proj", "model.layers.40.mlp.experts.18.down_proj", "model.layers.40.mlp.experts.19.down_proj", "model.layers.40.mlp.experts.20.down_proj", "model.layers.40.mlp.experts.21.down_proj", "model.layers.40.mlp.experts.22.down_proj", "model.layers.40.mlp.experts.23.down_proj", "model.layers.40.mlp.experts.24.down_proj", "model.layers.40.mlp.experts.25.down_proj", "model.layers.40.mlp.experts.26.down_proj", "model.layers.40.mlp.experts.27.down_proj", "model.layers.40.mlp.experts.28.down_proj", "model.layers.40.mlp.experts.29.down_proj", "model.layers.40.mlp.experts.30.down_proj", "model.layers.40.mlp.experts.31.down_proj", "model.layers.40.mlp.experts.32.down_proj", "model.layers.40.mlp.experts.33.down_proj", "model.layers.40.mlp.experts.34.down_proj", "model.layers.40.mlp.experts.35.down_proj", "model.layers.40.mlp.experts.36.down_proj", "model.layers.40.mlp.experts.37.down_proj", "model.layers.40.mlp.experts.38.down_proj", "model.layers.40.mlp.experts.39.down_proj", "model.layers.40.mlp.experts.40.down_proj", "model.layers.40.mlp.experts.41.down_proj", "model.layers.40.mlp.experts.42.down_proj", "model.layers.40.mlp.experts.43.down_proj", "model.layers.40.mlp.experts.44.down_proj", "model.layers.40.mlp.experts.45.down_proj", "model.layers.40.mlp.experts.46.down_proj", "model.layers.40.mlp.experts.47.down_proj", "model.layers.40.mlp.experts.48.down_proj", "model.layers.40.mlp.experts.49.down_proj", "model.layers.40.mlp.experts.50.down_proj", "model.layers.40.mlp.experts.51.down_proj", "model.layers.40.mlp.experts.52.down_proj", "model.layers.40.mlp.experts.53.down_proj", "model.layers.40.mlp.experts.54.down_proj", "model.layers.40.mlp.experts.55.down_proj", "model.layers.40.mlp.experts.56.down_proj", "model.layers.40.mlp.experts.57.down_proj", "model.layers.40.mlp.experts.58.down_proj", "model.layers.40.mlp.experts.59.down_proj", "model.layers.40.mlp.experts.60.down_proj", "model.layers.40.mlp.experts.61.down_proj", "model.layers.40.mlp.experts.62.down_proj", "model.layers.40.mlp.experts.63.down_proj", "model.layers.40.mlp.experts.64.down_proj", "model.layers.40.mlp.experts.65.down_proj", "model.layers.40.mlp.experts.66.down_proj", "model.layers.40.mlp.experts.67.down_proj", "model.layers.40.mlp.experts.68.down_proj", "model.layers.40.mlp.experts.69.down_proj", "model.layers.40.mlp.experts.70.down_proj", "model.layers.40.mlp.experts.71.down_proj", "model.layers.40.mlp.experts.72.down_proj", "model.layers.40.mlp.experts.73.down_proj", "model.layers.40.mlp.experts.74.down_proj", "model.layers.40.mlp.experts.75.down_proj", "model.layers.40.mlp.experts.76.down_proj", "model.layers.40.mlp.experts.77.down_proj", "model.layers.40.mlp.experts.78.down_proj", "model.layers.40.mlp.experts.79.down_proj", "model.layers.40.mlp.experts.80.down_proj", "model.layers.40.mlp.experts.81.down_proj", "model.layers.40.mlp.experts.82.down_proj", "model.layers.40.mlp.experts.83.down_proj", "model.layers.40.mlp.experts.84.down_proj", "model.layers.40.mlp.experts.85.down_proj", "model.layers.40.mlp.experts.86.down_proj", "model.layers.40.mlp.experts.87.down_proj", "model.layers.40.mlp.experts.88.down_proj", "model.layers.40.mlp.experts.89.down_proj", "model.layers.40.mlp.experts.90.down_proj", "model.layers.40.mlp.experts.91.down_proj", "model.layers.40.mlp.experts.92.down_proj", "model.layers.40.mlp.experts.93.down_proj", "model.layers.40.mlp.experts.94.down_proj", "model.layers.40.mlp.experts.95.down_proj", "model.layers.40.mlp.experts.96.down_proj", "model.layers.40.mlp.experts.97.down_proj", "model.layers.40.mlp.experts.98.down_proj", "model.layers.40.mlp.experts.99.down_proj", "model.layers.40.mlp.experts.100.down_proj", "model.layers.40.mlp.experts.101.down_proj", "model.layers.40.mlp.experts.102.down_proj", "model.layers.40.mlp.experts.103.down_proj", "model.layers.40.mlp.experts.104.down_proj", "model.layers.40.mlp.experts.105.down_proj", "model.layers.40.mlp.experts.106.down_proj", "model.layers.40.mlp.experts.107.down_proj", "model.layers.40.mlp.experts.108.down_proj", "model.layers.40.mlp.experts.109.down_proj", "model.layers.40.mlp.experts.110.down_proj", "model.layers.40.mlp.experts.111.down_proj", "model.layers.40.mlp.experts.112.down_proj", "model.layers.40.mlp.experts.113.down_proj", "model.layers.40.mlp.experts.114.down_proj", "model.layers.40.mlp.experts.115.down_proj", "model.layers.40.mlp.experts.116.down_proj", "model.layers.40.mlp.experts.117.down_proj", "model.layers.40.mlp.experts.118.down_proj", "model.layers.40.mlp.experts.119.down_proj", "model.layers.40.mlp.experts.120.down_proj", "model.layers.40.mlp.experts.121.down_proj", "model.layers.40.mlp.experts.122.down_proj", "model.layers.40.mlp.experts.123.down_proj", "model.layers.40.mlp.experts.124.down_proj", "model.layers.40.mlp.experts.125.down_proj", "model.layers.40.mlp.experts.126.down_proj", "model.layers.40.mlp.experts.127.down_proj", "model.layers.40.mlp.experts.128.down_proj", "model.layers.40.mlp.experts.129.down_proj", "model.layers.40.mlp.experts.130.down_proj", "model.layers.40.mlp.experts.131.down_proj", "model.layers.40.mlp.experts.132.down_proj", "model.layers.40.mlp.experts.133.down_proj", "model.layers.40.mlp.experts.134.down_proj", "model.layers.40.mlp.experts.135.down_proj", "model.layers.40.mlp.experts.136.down_proj", "model.layers.40.mlp.experts.137.down_proj", "model.layers.40.mlp.experts.138.down_proj", "model.layers.40.mlp.experts.139.down_proj", "model.layers.40.mlp.experts.140.down_proj", "model.layers.40.mlp.experts.141.down_proj", "model.layers.40.mlp.experts.142.down_proj", "model.layers.40.mlp.experts.143.down_proj", "model.layers.40.mlp.experts.144.down_proj", "model.layers.40.mlp.experts.145.down_proj", "model.layers.40.mlp.experts.146.down_proj", "model.layers.40.mlp.experts.147.down_proj", "model.layers.40.mlp.experts.148.down_proj", "model.layers.40.mlp.experts.149.down_proj", "model.layers.40.mlp.experts.150.down_proj", "model.layers.40.mlp.experts.151.down_proj", "model.layers.40.mlp.experts.152.down_proj", "model.layers.40.mlp.experts.153.down_proj", "model.layers.40.mlp.experts.154.down_proj", "model.layers.40.mlp.experts.155.down_proj", "model.layers.40.mlp.experts.156.down_proj", "model.layers.40.mlp.experts.157.down_proj", "model.layers.40.mlp.experts.158.down_proj", "model.layers.40.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00036488287150859833, "dbits": 1258291200 }, { "dkld": -0.0005805401131510762, "dbits": 2516582400 }, { "dkld": -0.0005546147003769875, "dbits": 3774873600 }, { "dkld": -0.0005677251145243672, "dbits": 6291456000 } ] }, { "idx": 243, "layers": [ "model.layers.41.self_attn.q_proj" ], "candidates": [ { "dkld": 0.000617603585124013, "dbits": 62914560 }, { "dkld": 0.0006311688572168267, "dbits": 125829120 }, { "dkld": 0.000643853098154068, "dbits": 188743680 }, { "dkld": 0.0006713587790727615, "dbits": 314572800 } ] }, { "idx": 244, "layers": [ "model.layers.41.self_attn.k_proj", "model.layers.41.self_attn.v_proj" ], "candidates": [ { "dkld": -0.001099107787013054, "dbits": 10485760 }, { "dkld": -0.000792243517935276, "dbits": 20971520 }, { "dkld": -0.0010010850615799483, "dbits": 31457280 }, { "dkld": -0.0009982030838728007, "dbits": 52428800 } ] }, { "idx": 245, "layers": [ "model.layers.41.self_attn.o_proj" ], "candidates": [ { "dkld": 0.000585016980767239, "dbits": 62914560 }, { "dkld": 0.0007584679871797534, "dbits": 125829120 }, { "dkld": 0.0007413984276354313, "dbits": 188743680 }, { "dkld": 0.0006505465134978294, "dbits": 314572800 } ] }, { "idx": 246, "layers": [ "model.layers.41.mlp.shared_experts.gate_proj", "model.layers.41.mlp.shared_experts.up_proj", "model.layers.41.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005610095337033327, "dbits": 23592960 }, { "dkld": -0.00014287251979112903, "dbits": 47185920 }, { "dkld": -0.00039469972252846597, "dbits": 70778880 }, { "dkld": -0.00031420998275280554, "dbits": 117964800 } ] }, { "idx": 247, "layers": [ "model.layers.41.mlp.experts.0.gate_proj", "model.layers.41.mlp.experts.1.gate_proj", "model.layers.41.mlp.experts.2.gate_proj", "model.layers.41.mlp.experts.3.gate_proj", "model.layers.41.mlp.experts.4.gate_proj", "model.layers.41.mlp.experts.5.gate_proj", "model.layers.41.mlp.experts.6.gate_proj", "model.layers.41.mlp.experts.7.gate_proj", "model.layers.41.mlp.experts.8.gate_proj", "model.layers.41.mlp.experts.9.gate_proj", "model.layers.41.mlp.experts.10.gate_proj", "model.layers.41.mlp.experts.11.gate_proj", "model.layers.41.mlp.experts.12.gate_proj", "model.layers.41.mlp.experts.13.gate_proj", "model.layers.41.mlp.experts.14.gate_proj", "model.layers.41.mlp.experts.15.gate_proj", "model.layers.41.mlp.experts.16.gate_proj", "model.layers.41.mlp.experts.17.gate_proj", "model.layers.41.mlp.experts.18.gate_proj", "model.layers.41.mlp.experts.19.gate_proj", "model.layers.41.mlp.experts.20.gate_proj", "model.layers.41.mlp.experts.21.gate_proj", "model.layers.41.mlp.experts.22.gate_proj", "model.layers.41.mlp.experts.23.gate_proj", "model.layers.41.mlp.experts.24.gate_proj", "model.layers.41.mlp.experts.25.gate_proj", "model.layers.41.mlp.experts.26.gate_proj", "model.layers.41.mlp.experts.27.gate_proj", "model.layers.41.mlp.experts.28.gate_proj", "model.layers.41.mlp.experts.29.gate_proj", "model.layers.41.mlp.experts.30.gate_proj", "model.layers.41.mlp.experts.31.gate_proj", "model.layers.41.mlp.experts.32.gate_proj", "model.layers.41.mlp.experts.33.gate_proj", "model.layers.41.mlp.experts.34.gate_proj", "model.layers.41.mlp.experts.35.gate_proj", "model.layers.41.mlp.experts.36.gate_proj", "model.layers.41.mlp.experts.37.gate_proj", "model.layers.41.mlp.experts.38.gate_proj", "model.layers.41.mlp.experts.39.gate_proj", "model.layers.41.mlp.experts.40.gate_proj", "model.layers.41.mlp.experts.41.gate_proj", "model.layers.41.mlp.experts.42.gate_proj", "model.layers.41.mlp.experts.43.gate_proj", "model.layers.41.mlp.experts.44.gate_proj", "model.layers.41.mlp.experts.45.gate_proj", "model.layers.41.mlp.experts.46.gate_proj", "model.layers.41.mlp.experts.47.gate_proj", "model.layers.41.mlp.experts.48.gate_proj", "model.layers.41.mlp.experts.49.gate_proj", "model.layers.41.mlp.experts.50.gate_proj", "model.layers.41.mlp.experts.51.gate_proj", "model.layers.41.mlp.experts.52.gate_proj", "model.layers.41.mlp.experts.53.gate_proj", "model.layers.41.mlp.experts.54.gate_proj", "model.layers.41.mlp.experts.55.gate_proj", "model.layers.41.mlp.experts.56.gate_proj", "model.layers.41.mlp.experts.57.gate_proj", "model.layers.41.mlp.experts.58.gate_proj", "model.layers.41.mlp.experts.59.gate_proj", "model.layers.41.mlp.experts.60.gate_proj", "model.layers.41.mlp.experts.61.gate_proj", "model.layers.41.mlp.experts.62.gate_proj", "model.layers.41.mlp.experts.63.gate_proj", "model.layers.41.mlp.experts.64.gate_proj", "model.layers.41.mlp.experts.65.gate_proj", "model.layers.41.mlp.experts.66.gate_proj", "model.layers.41.mlp.experts.67.gate_proj", "model.layers.41.mlp.experts.68.gate_proj", "model.layers.41.mlp.experts.69.gate_proj", "model.layers.41.mlp.experts.70.gate_proj", "model.layers.41.mlp.experts.71.gate_proj", "model.layers.41.mlp.experts.72.gate_proj", "model.layers.41.mlp.experts.73.gate_proj", "model.layers.41.mlp.experts.74.gate_proj", "model.layers.41.mlp.experts.75.gate_proj", "model.layers.41.mlp.experts.76.gate_proj", "model.layers.41.mlp.experts.77.gate_proj", "model.layers.41.mlp.experts.78.gate_proj", "model.layers.41.mlp.experts.79.gate_proj", "model.layers.41.mlp.experts.80.gate_proj", "model.layers.41.mlp.experts.81.gate_proj", "model.layers.41.mlp.experts.82.gate_proj", "model.layers.41.mlp.experts.83.gate_proj", "model.layers.41.mlp.experts.84.gate_proj", "model.layers.41.mlp.experts.85.gate_proj", "model.layers.41.mlp.experts.86.gate_proj", "model.layers.41.mlp.experts.87.gate_proj", "model.layers.41.mlp.experts.88.gate_proj", "model.layers.41.mlp.experts.89.gate_proj", "model.layers.41.mlp.experts.90.gate_proj", "model.layers.41.mlp.experts.91.gate_proj", "model.layers.41.mlp.experts.92.gate_proj", "model.layers.41.mlp.experts.93.gate_proj", "model.layers.41.mlp.experts.94.gate_proj", "model.layers.41.mlp.experts.95.gate_proj", "model.layers.41.mlp.experts.96.gate_proj", "model.layers.41.mlp.experts.97.gate_proj", "model.layers.41.mlp.experts.98.gate_proj", "model.layers.41.mlp.experts.99.gate_proj", "model.layers.41.mlp.experts.100.gate_proj", "model.layers.41.mlp.experts.101.gate_proj", "model.layers.41.mlp.experts.102.gate_proj", "model.layers.41.mlp.experts.103.gate_proj", "model.layers.41.mlp.experts.104.gate_proj", "model.layers.41.mlp.experts.105.gate_proj", "model.layers.41.mlp.experts.106.gate_proj", "model.layers.41.mlp.experts.107.gate_proj", "model.layers.41.mlp.experts.108.gate_proj", "model.layers.41.mlp.experts.109.gate_proj", "model.layers.41.mlp.experts.110.gate_proj", "model.layers.41.mlp.experts.111.gate_proj", "model.layers.41.mlp.experts.112.gate_proj", "model.layers.41.mlp.experts.113.gate_proj", "model.layers.41.mlp.experts.114.gate_proj", "model.layers.41.mlp.experts.115.gate_proj", "model.layers.41.mlp.experts.116.gate_proj", "model.layers.41.mlp.experts.117.gate_proj", "model.layers.41.mlp.experts.118.gate_proj", "model.layers.41.mlp.experts.119.gate_proj", "model.layers.41.mlp.experts.120.gate_proj", "model.layers.41.mlp.experts.121.gate_proj", "model.layers.41.mlp.experts.122.gate_proj", "model.layers.41.mlp.experts.123.gate_proj", "model.layers.41.mlp.experts.124.gate_proj", "model.layers.41.mlp.experts.125.gate_proj", "model.layers.41.mlp.experts.126.gate_proj", "model.layers.41.mlp.experts.127.gate_proj", "model.layers.41.mlp.experts.128.gate_proj", "model.layers.41.mlp.experts.129.gate_proj", "model.layers.41.mlp.experts.130.gate_proj", "model.layers.41.mlp.experts.131.gate_proj", "model.layers.41.mlp.experts.132.gate_proj", "model.layers.41.mlp.experts.133.gate_proj", "model.layers.41.mlp.experts.134.gate_proj", "model.layers.41.mlp.experts.135.gate_proj", "model.layers.41.mlp.experts.136.gate_proj", "model.layers.41.mlp.experts.137.gate_proj", "model.layers.41.mlp.experts.138.gate_proj", "model.layers.41.mlp.experts.139.gate_proj", "model.layers.41.mlp.experts.140.gate_proj", "model.layers.41.mlp.experts.141.gate_proj", "model.layers.41.mlp.experts.142.gate_proj", "model.layers.41.mlp.experts.143.gate_proj", "model.layers.41.mlp.experts.144.gate_proj", "model.layers.41.mlp.experts.145.gate_proj", "model.layers.41.mlp.experts.146.gate_proj", "model.layers.41.mlp.experts.147.gate_proj", "model.layers.41.mlp.experts.148.gate_proj", "model.layers.41.mlp.experts.149.gate_proj", "model.layers.41.mlp.experts.150.gate_proj", "model.layers.41.mlp.experts.151.gate_proj", "model.layers.41.mlp.experts.152.gate_proj", "model.layers.41.mlp.experts.153.gate_proj", "model.layers.41.mlp.experts.154.gate_proj", "model.layers.41.mlp.experts.155.gate_proj", "model.layers.41.mlp.experts.156.gate_proj", "model.layers.41.mlp.experts.157.gate_proj", "model.layers.41.mlp.experts.158.gate_proj", "model.layers.41.mlp.experts.159.gate_proj", "model.layers.41.mlp.experts.0.up_proj", "model.layers.41.mlp.experts.1.up_proj", "model.layers.41.mlp.experts.2.up_proj", "model.layers.41.mlp.experts.3.up_proj", "model.layers.41.mlp.experts.4.up_proj", "model.layers.41.mlp.experts.5.up_proj", "model.layers.41.mlp.experts.6.up_proj", "model.layers.41.mlp.experts.7.up_proj", "model.layers.41.mlp.experts.8.up_proj", "model.layers.41.mlp.experts.9.up_proj", "model.layers.41.mlp.experts.10.up_proj", "model.layers.41.mlp.experts.11.up_proj", "model.layers.41.mlp.experts.12.up_proj", "model.layers.41.mlp.experts.13.up_proj", "model.layers.41.mlp.experts.14.up_proj", "model.layers.41.mlp.experts.15.up_proj", "model.layers.41.mlp.experts.16.up_proj", "model.layers.41.mlp.experts.17.up_proj", "model.layers.41.mlp.experts.18.up_proj", "model.layers.41.mlp.experts.19.up_proj", "model.layers.41.mlp.experts.20.up_proj", "model.layers.41.mlp.experts.21.up_proj", "model.layers.41.mlp.experts.22.up_proj", "model.layers.41.mlp.experts.23.up_proj", "model.layers.41.mlp.experts.24.up_proj", "model.layers.41.mlp.experts.25.up_proj", "model.layers.41.mlp.experts.26.up_proj", "model.layers.41.mlp.experts.27.up_proj", "model.layers.41.mlp.experts.28.up_proj", "model.layers.41.mlp.experts.29.up_proj", "model.layers.41.mlp.experts.30.up_proj", "model.layers.41.mlp.experts.31.up_proj", "model.layers.41.mlp.experts.32.up_proj", "model.layers.41.mlp.experts.33.up_proj", "model.layers.41.mlp.experts.34.up_proj", "model.layers.41.mlp.experts.35.up_proj", "model.layers.41.mlp.experts.36.up_proj", "model.layers.41.mlp.experts.37.up_proj", "model.layers.41.mlp.experts.38.up_proj", "model.layers.41.mlp.experts.39.up_proj", "model.layers.41.mlp.experts.40.up_proj", "model.layers.41.mlp.experts.41.up_proj", "model.layers.41.mlp.experts.42.up_proj", "model.layers.41.mlp.experts.43.up_proj", "model.layers.41.mlp.experts.44.up_proj", "model.layers.41.mlp.experts.45.up_proj", "model.layers.41.mlp.experts.46.up_proj", "model.layers.41.mlp.experts.47.up_proj", "model.layers.41.mlp.experts.48.up_proj", "model.layers.41.mlp.experts.49.up_proj", "model.layers.41.mlp.experts.50.up_proj", "model.layers.41.mlp.experts.51.up_proj", "model.layers.41.mlp.experts.52.up_proj", "model.layers.41.mlp.experts.53.up_proj", "model.layers.41.mlp.experts.54.up_proj", "model.layers.41.mlp.experts.55.up_proj", "model.layers.41.mlp.experts.56.up_proj", "model.layers.41.mlp.experts.57.up_proj", "model.layers.41.mlp.experts.58.up_proj", "model.layers.41.mlp.experts.59.up_proj", "model.layers.41.mlp.experts.60.up_proj", "model.layers.41.mlp.experts.61.up_proj", "model.layers.41.mlp.experts.62.up_proj", "model.layers.41.mlp.experts.63.up_proj", "model.layers.41.mlp.experts.64.up_proj", "model.layers.41.mlp.experts.65.up_proj", "model.layers.41.mlp.experts.66.up_proj", "model.layers.41.mlp.experts.67.up_proj", "model.layers.41.mlp.experts.68.up_proj", "model.layers.41.mlp.experts.69.up_proj", "model.layers.41.mlp.experts.70.up_proj", "model.layers.41.mlp.experts.71.up_proj", "model.layers.41.mlp.experts.72.up_proj", "model.layers.41.mlp.experts.73.up_proj", "model.layers.41.mlp.experts.74.up_proj", "model.layers.41.mlp.experts.75.up_proj", "model.layers.41.mlp.experts.76.up_proj", "model.layers.41.mlp.experts.77.up_proj", "model.layers.41.mlp.experts.78.up_proj", "model.layers.41.mlp.experts.79.up_proj", "model.layers.41.mlp.experts.80.up_proj", "model.layers.41.mlp.experts.81.up_proj", "model.layers.41.mlp.experts.82.up_proj", "model.layers.41.mlp.experts.83.up_proj", "model.layers.41.mlp.experts.84.up_proj", "model.layers.41.mlp.experts.85.up_proj", "model.layers.41.mlp.experts.86.up_proj", "model.layers.41.mlp.experts.87.up_proj", "model.layers.41.mlp.experts.88.up_proj", "model.layers.41.mlp.experts.89.up_proj", "model.layers.41.mlp.experts.90.up_proj", "model.layers.41.mlp.experts.91.up_proj", "model.layers.41.mlp.experts.92.up_proj", "model.layers.41.mlp.experts.93.up_proj", "model.layers.41.mlp.experts.94.up_proj", "model.layers.41.mlp.experts.95.up_proj", "model.layers.41.mlp.experts.96.up_proj", "model.layers.41.mlp.experts.97.up_proj", "model.layers.41.mlp.experts.98.up_proj", "model.layers.41.mlp.experts.99.up_proj", "model.layers.41.mlp.experts.100.up_proj", "model.layers.41.mlp.experts.101.up_proj", "model.layers.41.mlp.experts.102.up_proj", "model.layers.41.mlp.experts.103.up_proj", "model.layers.41.mlp.experts.104.up_proj", "model.layers.41.mlp.experts.105.up_proj", "model.layers.41.mlp.experts.106.up_proj", "model.layers.41.mlp.experts.107.up_proj", "model.layers.41.mlp.experts.108.up_proj", "model.layers.41.mlp.experts.109.up_proj", "model.layers.41.mlp.experts.110.up_proj", "model.layers.41.mlp.experts.111.up_proj", "model.layers.41.mlp.experts.112.up_proj", "model.layers.41.mlp.experts.113.up_proj", "model.layers.41.mlp.experts.114.up_proj", "model.layers.41.mlp.experts.115.up_proj", "model.layers.41.mlp.experts.116.up_proj", "model.layers.41.mlp.experts.117.up_proj", "model.layers.41.mlp.experts.118.up_proj", "model.layers.41.mlp.experts.119.up_proj", "model.layers.41.mlp.experts.120.up_proj", "model.layers.41.mlp.experts.121.up_proj", "model.layers.41.mlp.experts.122.up_proj", "model.layers.41.mlp.experts.123.up_proj", "model.layers.41.mlp.experts.124.up_proj", "model.layers.41.mlp.experts.125.up_proj", "model.layers.41.mlp.experts.126.up_proj", "model.layers.41.mlp.experts.127.up_proj", "model.layers.41.mlp.experts.128.up_proj", "model.layers.41.mlp.experts.129.up_proj", "model.layers.41.mlp.experts.130.up_proj", "model.layers.41.mlp.experts.131.up_proj", "model.layers.41.mlp.experts.132.up_proj", "model.layers.41.mlp.experts.133.up_proj", "model.layers.41.mlp.experts.134.up_proj", "model.layers.41.mlp.experts.135.up_proj", "model.layers.41.mlp.experts.136.up_proj", "model.layers.41.mlp.experts.137.up_proj", "model.layers.41.mlp.experts.138.up_proj", "model.layers.41.mlp.experts.139.up_proj", "model.layers.41.mlp.experts.140.up_proj", "model.layers.41.mlp.experts.141.up_proj", "model.layers.41.mlp.experts.142.up_proj", "model.layers.41.mlp.experts.143.up_proj", "model.layers.41.mlp.experts.144.up_proj", "model.layers.41.mlp.experts.145.up_proj", "model.layers.41.mlp.experts.146.up_proj", "model.layers.41.mlp.experts.147.up_proj", "model.layers.41.mlp.experts.148.up_proj", "model.layers.41.mlp.experts.149.up_proj", "model.layers.41.mlp.experts.150.up_proj", "model.layers.41.mlp.experts.151.up_proj", "model.layers.41.mlp.experts.152.up_proj", "model.layers.41.mlp.experts.153.up_proj", "model.layers.41.mlp.experts.154.up_proj", "model.layers.41.mlp.experts.155.up_proj", "model.layers.41.mlp.experts.156.up_proj", "model.layers.41.mlp.experts.157.up_proj", "model.layers.41.mlp.experts.158.up_proj", "model.layers.41.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0007411091588437613, "dbits": 2516582400 }, { "dkld": -0.0009087816812098082, "dbits": 5033164800 }, { "dkld": -0.0008003478869795799, "dbits": 7549747200 }, { "dkld": -0.0007512837648391835, "dbits": 12582912000 } ] }, { "idx": 248, "layers": [ "model.layers.41.mlp.experts.0.down_proj", "model.layers.41.mlp.experts.1.down_proj", "model.layers.41.mlp.experts.2.down_proj", "model.layers.41.mlp.experts.3.down_proj", "model.layers.41.mlp.experts.4.down_proj", "model.layers.41.mlp.experts.5.down_proj", "model.layers.41.mlp.experts.6.down_proj", "model.layers.41.mlp.experts.7.down_proj", "model.layers.41.mlp.experts.8.down_proj", "model.layers.41.mlp.experts.9.down_proj", "model.layers.41.mlp.experts.10.down_proj", "model.layers.41.mlp.experts.11.down_proj", "model.layers.41.mlp.experts.12.down_proj", "model.layers.41.mlp.experts.13.down_proj", "model.layers.41.mlp.experts.14.down_proj", "model.layers.41.mlp.experts.15.down_proj", "model.layers.41.mlp.experts.16.down_proj", "model.layers.41.mlp.experts.17.down_proj", "model.layers.41.mlp.experts.18.down_proj", "model.layers.41.mlp.experts.19.down_proj", "model.layers.41.mlp.experts.20.down_proj", "model.layers.41.mlp.experts.21.down_proj", "model.layers.41.mlp.experts.22.down_proj", "model.layers.41.mlp.experts.23.down_proj", "model.layers.41.mlp.experts.24.down_proj", "model.layers.41.mlp.experts.25.down_proj", "model.layers.41.mlp.experts.26.down_proj", "model.layers.41.mlp.experts.27.down_proj", "model.layers.41.mlp.experts.28.down_proj", "model.layers.41.mlp.experts.29.down_proj", "model.layers.41.mlp.experts.30.down_proj", "model.layers.41.mlp.experts.31.down_proj", "model.layers.41.mlp.experts.32.down_proj", "model.layers.41.mlp.experts.33.down_proj", "model.layers.41.mlp.experts.34.down_proj", "model.layers.41.mlp.experts.35.down_proj", "model.layers.41.mlp.experts.36.down_proj", "model.layers.41.mlp.experts.37.down_proj", "model.layers.41.mlp.experts.38.down_proj", "model.layers.41.mlp.experts.39.down_proj", "model.layers.41.mlp.experts.40.down_proj", "model.layers.41.mlp.experts.41.down_proj", "model.layers.41.mlp.experts.42.down_proj", "model.layers.41.mlp.experts.43.down_proj", "model.layers.41.mlp.experts.44.down_proj", "model.layers.41.mlp.experts.45.down_proj", "model.layers.41.mlp.experts.46.down_proj", "model.layers.41.mlp.experts.47.down_proj", "model.layers.41.mlp.experts.48.down_proj", "model.layers.41.mlp.experts.49.down_proj", "model.layers.41.mlp.experts.50.down_proj", "model.layers.41.mlp.experts.51.down_proj", "model.layers.41.mlp.experts.52.down_proj", "model.layers.41.mlp.experts.53.down_proj", "model.layers.41.mlp.experts.54.down_proj", "model.layers.41.mlp.experts.55.down_proj", "model.layers.41.mlp.experts.56.down_proj", "model.layers.41.mlp.experts.57.down_proj", "model.layers.41.mlp.experts.58.down_proj", "model.layers.41.mlp.experts.59.down_proj", "model.layers.41.mlp.experts.60.down_proj", "model.layers.41.mlp.experts.61.down_proj", "model.layers.41.mlp.experts.62.down_proj", "model.layers.41.mlp.experts.63.down_proj", "model.layers.41.mlp.experts.64.down_proj", "model.layers.41.mlp.experts.65.down_proj", "model.layers.41.mlp.experts.66.down_proj", "model.layers.41.mlp.experts.67.down_proj", "model.layers.41.mlp.experts.68.down_proj", "model.layers.41.mlp.experts.69.down_proj", "model.layers.41.mlp.experts.70.down_proj", "model.layers.41.mlp.experts.71.down_proj", "model.layers.41.mlp.experts.72.down_proj", "model.layers.41.mlp.experts.73.down_proj", "model.layers.41.mlp.experts.74.down_proj", "model.layers.41.mlp.experts.75.down_proj", "model.layers.41.mlp.experts.76.down_proj", "model.layers.41.mlp.experts.77.down_proj", "model.layers.41.mlp.experts.78.down_proj", "model.layers.41.mlp.experts.79.down_proj", "model.layers.41.mlp.experts.80.down_proj", "model.layers.41.mlp.experts.81.down_proj", "model.layers.41.mlp.experts.82.down_proj", "model.layers.41.mlp.experts.83.down_proj", "model.layers.41.mlp.experts.84.down_proj", "model.layers.41.mlp.experts.85.down_proj", "model.layers.41.mlp.experts.86.down_proj", "model.layers.41.mlp.experts.87.down_proj", "model.layers.41.mlp.experts.88.down_proj", "model.layers.41.mlp.experts.89.down_proj", "model.layers.41.mlp.experts.90.down_proj", "model.layers.41.mlp.experts.91.down_proj", "model.layers.41.mlp.experts.92.down_proj", "model.layers.41.mlp.experts.93.down_proj", "model.layers.41.mlp.experts.94.down_proj", "model.layers.41.mlp.experts.95.down_proj", "model.layers.41.mlp.experts.96.down_proj", "model.layers.41.mlp.experts.97.down_proj", "model.layers.41.mlp.experts.98.down_proj", "model.layers.41.mlp.experts.99.down_proj", "model.layers.41.mlp.experts.100.down_proj", "model.layers.41.mlp.experts.101.down_proj", "model.layers.41.mlp.experts.102.down_proj", "model.layers.41.mlp.experts.103.down_proj", "model.layers.41.mlp.experts.104.down_proj", "model.layers.41.mlp.experts.105.down_proj", "model.layers.41.mlp.experts.106.down_proj", "model.layers.41.mlp.experts.107.down_proj", "model.layers.41.mlp.experts.108.down_proj", "model.layers.41.mlp.experts.109.down_proj", "model.layers.41.mlp.experts.110.down_proj", "model.layers.41.mlp.experts.111.down_proj", "model.layers.41.mlp.experts.112.down_proj", "model.layers.41.mlp.experts.113.down_proj", "model.layers.41.mlp.experts.114.down_proj", "model.layers.41.mlp.experts.115.down_proj", "model.layers.41.mlp.experts.116.down_proj", "model.layers.41.mlp.experts.117.down_proj", "model.layers.41.mlp.experts.118.down_proj", "model.layers.41.mlp.experts.119.down_proj", "model.layers.41.mlp.experts.120.down_proj", "model.layers.41.mlp.experts.121.down_proj", "model.layers.41.mlp.experts.122.down_proj", "model.layers.41.mlp.experts.123.down_proj", "model.layers.41.mlp.experts.124.down_proj", "model.layers.41.mlp.experts.125.down_proj", "model.layers.41.mlp.experts.126.down_proj", "model.layers.41.mlp.experts.127.down_proj", "model.layers.41.mlp.experts.128.down_proj", "model.layers.41.mlp.experts.129.down_proj", "model.layers.41.mlp.experts.130.down_proj", "model.layers.41.mlp.experts.131.down_proj", "model.layers.41.mlp.experts.132.down_proj", "model.layers.41.mlp.experts.133.down_proj", "model.layers.41.mlp.experts.134.down_proj", "model.layers.41.mlp.experts.135.down_proj", "model.layers.41.mlp.experts.136.down_proj", "model.layers.41.mlp.experts.137.down_proj", "model.layers.41.mlp.experts.138.down_proj", "model.layers.41.mlp.experts.139.down_proj", "model.layers.41.mlp.experts.140.down_proj", "model.layers.41.mlp.experts.141.down_proj", "model.layers.41.mlp.experts.142.down_proj", "model.layers.41.mlp.experts.143.down_proj", "model.layers.41.mlp.experts.144.down_proj", "model.layers.41.mlp.experts.145.down_proj", "model.layers.41.mlp.experts.146.down_proj", "model.layers.41.mlp.experts.147.down_proj", "model.layers.41.mlp.experts.148.down_proj", "model.layers.41.mlp.experts.149.down_proj", "model.layers.41.mlp.experts.150.down_proj", "model.layers.41.mlp.experts.151.down_proj", "model.layers.41.mlp.experts.152.down_proj", "model.layers.41.mlp.experts.153.down_proj", "model.layers.41.mlp.experts.154.down_proj", "model.layers.41.mlp.experts.155.down_proj", "model.layers.41.mlp.experts.156.down_proj", "model.layers.41.mlp.experts.157.down_proj", "model.layers.41.mlp.experts.158.down_proj", "model.layers.41.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00020975545048713962, "dbits": 1258291200 }, { "dkld": -0.0002054031938314438, "dbits": 2516582400 }, { "dkld": -0.00011838991194963733, "dbits": 3774873600 }, { "dkld": -0.00015998017042875845, "dbits": 6291456000 } ] }, { "idx": 249, "layers": [ "model.layers.42.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0003453691489994637, "dbits": 62914560 }, { "dkld": -0.00034209303557873605, "dbits": 125829120 }, { "dkld": -0.0003657594323158292, "dbits": 188743680 }, { "dkld": -0.00037030056118965426, "dbits": 314572800 } ] }, { "idx": 250, "layers": [ "model.layers.42.self_attn.k_proj", "model.layers.42.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00023239813745021265, "dbits": 10485760 }, { "dkld": -0.00048492783680559315, "dbits": 20971520 }, { "dkld": -0.0005543493665754878, "dbits": 31457280 }, { "dkld": -0.0005288410931825749, "dbits": 52428800 } ] }, { "idx": 251, "layers": [ "model.layers.42.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0009684733115136734, "dbits": 62914560 }, { "dkld": -0.0007618832401931341, "dbits": 125829120 }, { "dkld": -0.0007656428962946049, "dbits": 188743680 }, { "dkld": -0.0007687916979193798, "dbits": 314572800 } ] }, { "idx": 252, "layers": [ "model.layers.42.mlp.shared_experts.gate_proj", "model.layers.42.mlp.shared_experts.up_proj", "model.layers.42.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0007234267890453422, "dbits": 23592960 }, { "dkld": -0.0007857883349061123, "dbits": 47185920 }, { "dkld": -0.0006418433040380478, "dbits": 70778880 }, { "dkld": -0.0006452073343098247, "dbits": 117964800 } ] }, { "idx": 253, "layers": [ "model.layers.42.mlp.experts.0.gate_proj", "model.layers.42.mlp.experts.1.gate_proj", "model.layers.42.mlp.experts.2.gate_proj", "model.layers.42.mlp.experts.3.gate_proj", "model.layers.42.mlp.experts.4.gate_proj", "model.layers.42.mlp.experts.5.gate_proj", "model.layers.42.mlp.experts.6.gate_proj", "model.layers.42.mlp.experts.7.gate_proj", "model.layers.42.mlp.experts.8.gate_proj", "model.layers.42.mlp.experts.9.gate_proj", "model.layers.42.mlp.experts.10.gate_proj", "model.layers.42.mlp.experts.11.gate_proj", "model.layers.42.mlp.experts.12.gate_proj", "model.layers.42.mlp.experts.13.gate_proj", "model.layers.42.mlp.experts.14.gate_proj", "model.layers.42.mlp.experts.15.gate_proj", "model.layers.42.mlp.experts.16.gate_proj", "model.layers.42.mlp.experts.17.gate_proj", "model.layers.42.mlp.experts.18.gate_proj", "model.layers.42.mlp.experts.19.gate_proj", "model.layers.42.mlp.experts.20.gate_proj", "model.layers.42.mlp.experts.21.gate_proj", "model.layers.42.mlp.experts.22.gate_proj", "model.layers.42.mlp.experts.23.gate_proj", "model.layers.42.mlp.experts.24.gate_proj", "model.layers.42.mlp.experts.25.gate_proj", "model.layers.42.mlp.experts.26.gate_proj", "model.layers.42.mlp.experts.27.gate_proj", "model.layers.42.mlp.experts.28.gate_proj", "model.layers.42.mlp.experts.29.gate_proj", "model.layers.42.mlp.experts.30.gate_proj", "model.layers.42.mlp.experts.31.gate_proj", "model.layers.42.mlp.experts.32.gate_proj", "model.layers.42.mlp.experts.33.gate_proj", "model.layers.42.mlp.experts.34.gate_proj", "model.layers.42.mlp.experts.35.gate_proj", "model.layers.42.mlp.experts.36.gate_proj", "model.layers.42.mlp.experts.37.gate_proj", "model.layers.42.mlp.experts.38.gate_proj", "model.layers.42.mlp.experts.39.gate_proj", "model.layers.42.mlp.experts.40.gate_proj", "model.layers.42.mlp.experts.41.gate_proj", "model.layers.42.mlp.experts.42.gate_proj", "model.layers.42.mlp.experts.43.gate_proj", "model.layers.42.mlp.experts.44.gate_proj", "model.layers.42.mlp.experts.45.gate_proj", "model.layers.42.mlp.experts.46.gate_proj", "model.layers.42.mlp.experts.47.gate_proj", "model.layers.42.mlp.experts.48.gate_proj", "model.layers.42.mlp.experts.49.gate_proj", "model.layers.42.mlp.experts.50.gate_proj", "model.layers.42.mlp.experts.51.gate_proj", "model.layers.42.mlp.experts.52.gate_proj", "model.layers.42.mlp.experts.53.gate_proj", "model.layers.42.mlp.experts.54.gate_proj", "model.layers.42.mlp.experts.55.gate_proj", "model.layers.42.mlp.experts.56.gate_proj", "model.layers.42.mlp.experts.57.gate_proj", "model.layers.42.mlp.experts.58.gate_proj", "model.layers.42.mlp.experts.59.gate_proj", "model.layers.42.mlp.experts.60.gate_proj", "model.layers.42.mlp.experts.61.gate_proj", "model.layers.42.mlp.experts.62.gate_proj", "model.layers.42.mlp.experts.63.gate_proj", "model.layers.42.mlp.experts.64.gate_proj", "model.layers.42.mlp.experts.65.gate_proj", "model.layers.42.mlp.experts.66.gate_proj", "model.layers.42.mlp.experts.67.gate_proj", "model.layers.42.mlp.experts.68.gate_proj", "model.layers.42.mlp.experts.69.gate_proj", "model.layers.42.mlp.experts.70.gate_proj", "model.layers.42.mlp.experts.71.gate_proj", "model.layers.42.mlp.experts.72.gate_proj", "model.layers.42.mlp.experts.73.gate_proj", "model.layers.42.mlp.experts.74.gate_proj", "model.layers.42.mlp.experts.75.gate_proj", "model.layers.42.mlp.experts.76.gate_proj", "model.layers.42.mlp.experts.77.gate_proj", "model.layers.42.mlp.experts.78.gate_proj", "model.layers.42.mlp.experts.79.gate_proj", "model.layers.42.mlp.experts.80.gate_proj", "model.layers.42.mlp.experts.81.gate_proj", "model.layers.42.mlp.experts.82.gate_proj", "model.layers.42.mlp.experts.83.gate_proj", "model.layers.42.mlp.experts.84.gate_proj", "model.layers.42.mlp.experts.85.gate_proj", "model.layers.42.mlp.experts.86.gate_proj", "model.layers.42.mlp.experts.87.gate_proj", "model.layers.42.mlp.experts.88.gate_proj", "model.layers.42.mlp.experts.89.gate_proj", "model.layers.42.mlp.experts.90.gate_proj", "model.layers.42.mlp.experts.91.gate_proj", "model.layers.42.mlp.experts.92.gate_proj", "model.layers.42.mlp.experts.93.gate_proj", "model.layers.42.mlp.experts.94.gate_proj", "model.layers.42.mlp.experts.95.gate_proj", "model.layers.42.mlp.experts.96.gate_proj", "model.layers.42.mlp.experts.97.gate_proj", "model.layers.42.mlp.experts.98.gate_proj", "model.layers.42.mlp.experts.99.gate_proj", "model.layers.42.mlp.experts.100.gate_proj", "model.layers.42.mlp.experts.101.gate_proj", "model.layers.42.mlp.experts.102.gate_proj", "model.layers.42.mlp.experts.103.gate_proj", "model.layers.42.mlp.experts.104.gate_proj", "model.layers.42.mlp.experts.105.gate_proj", "model.layers.42.mlp.experts.106.gate_proj", "model.layers.42.mlp.experts.107.gate_proj", "model.layers.42.mlp.experts.108.gate_proj", "model.layers.42.mlp.experts.109.gate_proj", "model.layers.42.mlp.experts.110.gate_proj", "model.layers.42.mlp.experts.111.gate_proj", "model.layers.42.mlp.experts.112.gate_proj", "model.layers.42.mlp.experts.113.gate_proj", "model.layers.42.mlp.experts.114.gate_proj", "model.layers.42.mlp.experts.115.gate_proj", "model.layers.42.mlp.experts.116.gate_proj", "model.layers.42.mlp.experts.117.gate_proj", "model.layers.42.mlp.experts.118.gate_proj", "model.layers.42.mlp.experts.119.gate_proj", "model.layers.42.mlp.experts.120.gate_proj", "model.layers.42.mlp.experts.121.gate_proj", "model.layers.42.mlp.experts.122.gate_proj", "model.layers.42.mlp.experts.123.gate_proj", "model.layers.42.mlp.experts.124.gate_proj", "model.layers.42.mlp.experts.125.gate_proj", "model.layers.42.mlp.experts.126.gate_proj", "model.layers.42.mlp.experts.127.gate_proj", "model.layers.42.mlp.experts.128.gate_proj", "model.layers.42.mlp.experts.129.gate_proj", "model.layers.42.mlp.experts.130.gate_proj", "model.layers.42.mlp.experts.131.gate_proj", "model.layers.42.mlp.experts.132.gate_proj", "model.layers.42.mlp.experts.133.gate_proj", "model.layers.42.mlp.experts.134.gate_proj", "model.layers.42.mlp.experts.135.gate_proj", "model.layers.42.mlp.experts.136.gate_proj", "model.layers.42.mlp.experts.137.gate_proj", "model.layers.42.mlp.experts.138.gate_proj", "model.layers.42.mlp.experts.139.gate_proj", "model.layers.42.mlp.experts.140.gate_proj", "model.layers.42.mlp.experts.141.gate_proj", "model.layers.42.mlp.experts.142.gate_proj", "model.layers.42.mlp.experts.143.gate_proj", "model.layers.42.mlp.experts.144.gate_proj", "model.layers.42.mlp.experts.145.gate_proj", "model.layers.42.mlp.experts.146.gate_proj", "model.layers.42.mlp.experts.147.gate_proj", "model.layers.42.mlp.experts.148.gate_proj", "model.layers.42.mlp.experts.149.gate_proj", "model.layers.42.mlp.experts.150.gate_proj", "model.layers.42.mlp.experts.151.gate_proj", "model.layers.42.mlp.experts.152.gate_proj", "model.layers.42.mlp.experts.153.gate_proj", "model.layers.42.mlp.experts.154.gate_proj", "model.layers.42.mlp.experts.155.gate_proj", "model.layers.42.mlp.experts.156.gate_proj", "model.layers.42.mlp.experts.157.gate_proj", "model.layers.42.mlp.experts.158.gate_proj", "model.layers.42.mlp.experts.159.gate_proj", "model.layers.42.mlp.experts.0.up_proj", "model.layers.42.mlp.experts.1.up_proj", "model.layers.42.mlp.experts.2.up_proj", "model.layers.42.mlp.experts.3.up_proj", "model.layers.42.mlp.experts.4.up_proj", "model.layers.42.mlp.experts.5.up_proj", "model.layers.42.mlp.experts.6.up_proj", "model.layers.42.mlp.experts.7.up_proj", "model.layers.42.mlp.experts.8.up_proj", "model.layers.42.mlp.experts.9.up_proj", "model.layers.42.mlp.experts.10.up_proj", "model.layers.42.mlp.experts.11.up_proj", "model.layers.42.mlp.experts.12.up_proj", "model.layers.42.mlp.experts.13.up_proj", "model.layers.42.mlp.experts.14.up_proj", "model.layers.42.mlp.experts.15.up_proj", "model.layers.42.mlp.experts.16.up_proj", "model.layers.42.mlp.experts.17.up_proj", "model.layers.42.mlp.experts.18.up_proj", "model.layers.42.mlp.experts.19.up_proj", "model.layers.42.mlp.experts.20.up_proj", "model.layers.42.mlp.experts.21.up_proj", "model.layers.42.mlp.experts.22.up_proj", "model.layers.42.mlp.experts.23.up_proj", "model.layers.42.mlp.experts.24.up_proj", "model.layers.42.mlp.experts.25.up_proj", "model.layers.42.mlp.experts.26.up_proj", "model.layers.42.mlp.experts.27.up_proj", "model.layers.42.mlp.experts.28.up_proj", "model.layers.42.mlp.experts.29.up_proj", "model.layers.42.mlp.experts.30.up_proj", "model.layers.42.mlp.experts.31.up_proj", "model.layers.42.mlp.experts.32.up_proj", "model.layers.42.mlp.experts.33.up_proj", "model.layers.42.mlp.experts.34.up_proj", "model.layers.42.mlp.experts.35.up_proj", "model.layers.42.mlp.experts.36.up_proj", "model.layers.42.mlp.experts.37.up_proj", "model.layers.42.mlp.experts.38.up_proj", "model.layers.42.mlp.experts.39.up_proj", "model.layers.42.mlp.experts.40.up_proj", "model.layers.42.mlp.experts.41.up_proj", "model.layers.42.mlp.experts.42.up_proj", "model.layers.42.mlp.experts.43.up_proj", "model.layers.42.mlp.experts.44.up_proj", "model.layers.42.mlp.experts.45.up_proj", "model.layers.42.mlp.experts.46.up_proj", "model.layers.42.mlp.experts.47.up_proj", "model.layers.42.mlp.experts.48.up_proj", "model.layers.42.mlp.experts.49.up_proj", "model.layers.42.mlp.experts.50.up_proj", "model.layers.42.mlp.experts.51.up_proj", "model.layers.42.mlp.experts.52.up_proj", "model.layers.42.mlp.experts.53.up_proj", "model.layers.42.mlp.experts.54.up_proj", "model.layers.42.mlp.experts.55.up_proj", "model.layers.42.mlp.experts.56.up_proj", "model.layers.42.mlp.experts.57.up_proj", "model.layers.42.mlp.experts.58.up_proj", "model.layers.42.mlp.experts.59.up_proj", "model.layers.42.mlp.experts.60.up_proj", "model.layers.42.mlp.experts.61.up_proj", "model.layers.42.mlp.experts.62.up_proj", "model.layers.42.mlp.experts.63.up_proj", "model.layers.42.mlp.experts.64.up_proj", "model.layers.42.mlp.experts.65.up_proj", "model.layers.42.mlp.experts.66.up_proj", "model.layers.42.mlp.experts.67.up_proj", "model.layers.42.mlp.experts.68.up_proj", "model.layers.42.mlp.experts.69.up_proj", "model.layers.42.mlp.experts.70.up_proj", "model.layers.42.mlp.experts.71.up_proj", "model.layers.42.mlp.experts.72.up_proj", "model.layers.42.mlp.experts.73.up_proj", "model.layers.42.mlp.experts.74.up_proj", "model.layers.42.mlp.experts.75.up_proj", "model.layers.42.mlp.experts.76.up_proj", "model.layers.42.mlp.experts.77.up_proj", "model.layers.42.mlp.experts.78.up_proj", "model.layers.42.mlp.experts.79.up_proj", "model.layers.42.mlp.experts.80.up_proj", "model.layers.42.mlp.experts.81.up_proj", "model.layers.42.mlp.experts.82.up_proj", "model.layers.42.mlp.experts.83.up_proj", "model.layers.42.mlp.experts.84.up_proj", "model.layers.42.mlp.experts.85.up_proj", "model.layers.42.mlp.experts.86.up_proj", "model.layers.42.mlp.experts.87.up_proj", "model.layers.42.mlp.experts.88.up_proj", "model.layers.42.mlp.experts.89.up_proj", "model.layers.42.mlp.experts.90.up_proj", "model.layers.42.mlp.experts.91.up_proj", "model.layers.42.mlp.experts.92.up_proj", "model.layers.42.mlp.experts.93.up_proj", "model.layers.42.mlp.experts.94.up_proj", "model.layers.42.mlp.experts.95.up_proj", "model.layers.42.mlp.experts.96.up_proj", "model.layers.42.mlp.experts.97.up_proj", "model.layers.42.mlp.experts.98.up_proj", "model.layers.42.mlp.experts.99.up_proj", "model.layers.42.mlp.experts.100.up_proj", "model.layers.42.mlp.experts.101.up_proj", "model.layers.42.mlp.experts.102.up_proj", "model.layers.42.mlp.experts.103.up_proj", "model.layers.42.mlp.experts.104.up_proj", "model.layers.42.mlp.experts.105.up_proj", "model.layers.42.mlp.experts.106.up_proj", "model.layers.42.mlp.experts.107.up_proj", "model.layers.42.mlp.experts.108.up_proj", "model.layers.42.mlp.experts.109.up_proj", "model.layers.42.mlp.experts.110.up_proj", "model.layers.42.mlp.experts.111.up_proj", "model.layers.42.mlp.experts.112.up_proj", "model.layers.42.mlp.experts.113.up_proj", "model.layers.42.mlp.experts.114.up_proj", "model.layers.42.mlp.experts.115.up_proj", "model.layers.42.mlp.experts.116.up_proj", "model.layers.42.mlp.experts.117.up_proj", "model.layers.42.mlp.experts.118.up_proj", "model.layers.42.mlp.experts.119.up_proj", "model.layers.42.mlp.experts.120.up_proj", "model.layers.42.mlp.experts.121.up_proj", "model.layers.42.mlp.experts.122.up_proj", "model.layers.42.mlp.experts.123.up_proj", "model.layers.42.mlp.experts.124.up_proj", "model.layers.42.mlp.experts.125.up_proj", "model.layers.42.mlp.experts.126.up_proj", "model.layers.42.mlp.experts.127.up_proj", "model.layers.42.mlp.experts.128.up_proj", "model.layers.42.mlp.experts.129.up_proj", "model.layers.42.mlp.experts.130.up_proj", "model.layers.42.mlp.experts.131.up_proj", "model.layers.42.mlp.experts.132.up_proj", "model.layers.42.mlp.experts.133.up_proj", "model.layers.42.mlp.experts.134.up_proj", "model.layers.42.mlp.experts.135.up_proj", "model.layers.42.mlp.experts.136.up_proj", "model.layers.42.mlp.experts.137.up_proj", "model.layers.42.mlp.experts.138.up_proj", "model.layers.42.mlp.experts.139.up_proj", "model.layers.42.mlp.experts.140.up_proj", "model.layers.42.mlp.experts.141.up_proj", "model.layers.42.mlp.experts.142.up_proj", "model.layers.42.mlp.experts.143.up_proj", "model.layers.42.mlp.experts.144.up_proj", "model.layers.42.mlp.experts.145.up_proj", "model.layers.42.mlp.experts.146.up_proj", "model.layers.42.mlp.experts.147.up_proj", "model.layers.42.mlp.experts.148.up_proj", "model.layers.42.mlp.experts.149.up_proj", "model.layers.42.mlp.experts.150.up_proj", "model.layers.42.mlp.experts.151.up_proj", "model.layers.42.mlp.experts.152.up_proj", "model.layers.42.mlp.experts.153.up_proj", "model.layers.42.mlp.experts.154.up_proj", "model.layers.42.mlp.experts.155.up_proj", "model.layers.42.mlp.experts.156.up_proj", "model.layers.42.mlp.experts.157.up_proj", "model.layers.42.mlp.experts.158.up_proj", "model.layers.42.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00038281586021185476, "dbits": 2516582400 }, { "dkld": -0.00035182209685445665, "dbits": 5033164800 }, { "dkld": -0.00036617899313569346, "dbits": 7549747200 }, { "dkld": -0.00042835734784603396, "dbits": 12582912000 } ] }, { "idx": 254, "layers": [ "model.layers.42.mlp.experts.0.down_proj", "model.layers.42.mlp.experts.1.down_proj", "model.layers.42.mlp.experts.2.down_proj", "model.layers.42.mlp.experts.3.down_proj", "model.layers.42.mlp.experts.4.down_proj", "model.layers.42.mlp.experts.5.down_proj", "model.layers.42.mlp.experts.6.down_proj", "model.layers.42.mlp.experts.7.down_proj", "model.layers.42.mlp.experts.8.down_proj", "model.layers.42.mlp.experts.9.down_proj", "model.layers.42.mlp.experts.10.down_proj", "model.layers.42.mlp.experts.11.down_proj", "model.layers.42.mlp.experts.12.down_proj", "model.layers.42.mlp.experts.13.down_proj", "model.layers.42.mlp.experts.14.down_proj", "model.layers.42.mlp.experts.15.down_proj", "model.layers.42.mlp.experts.16.down_proj", "model.layers.42.mlp.experts.17.down_proj", "model.layers.42.mlp.experts.18.down_proj", "model.layers.42.mlp.experts.19.down_proj", "model.layers.42.mlp.experts.20.down_proj", "model.layers.42.mlp.experts.21.down_proj", "model.layers.42.mlp.experts.22.down_proj", "model.layers.42.mlp.experts.23.down_proj", "model.layers.42.mlp.experts.24.down_proj", "model.layers.42.mlp.experts.25.down_proj", "model.layers.42.mlp.experts.26.down_proj", "model.layers.42.mlp.experts.27.down_proj", "model.layers.42.mlp.experts.28.down_proj", "model.layers.42.mlp.experts.29.down_proj", "model.layers.42.mlp.experts.30.down_proj", "model.layers.42.mlp.experts.31.down_proj", "model.layers.42.mlp.experts.32.down_proj", "model.layers.42.mlp.experts.33.down_proj", "model.layers.42.mlp.experts.34.down_proj", "model.layers.42.mlp.experts.35.down_proj", "model.layers.42.mlp.experts.36.down_proj", "model.layers.42.mlp.experts.37.down_proj", "model.layers.42.mlp.experts.38.down_proj", "model.layers.42.mlp.experts.39.down_proj", "model.layers.42.mlp.experts.40.down_proj", "model.layers.42.mlp.experts.41.down_proj", "model.layers.42.mlp.experts.42.down_proj", "model.layers.42.mlp.experts.43.down_proj", "model.layers.42.mlp.experts.44.down_proj", "model.layers.42.mlp.experts.45.down_proj", "model.layers.42.mlp.experts.46.down_proj", "model.layers.42.mlp.experts.47.down_proj", "model.layers.42.mlp.experts.48.down_proj", "model.layers.42.mlp.experts.49.down_proj", "model.layers.42.mlp.experts.50.down_proj", "model.layers.42.mlp.experts.51.down_proj", "model.layers.42.mlp.experts.52.down_proj", "model.layers.42.mlp.experts.53.down_proj", "model.layers.42.mlp.experts.54.down_proj", "model.layers.42.mlp.experts.55.down_proj", "model.layers.42.mlp.experts.56.down_proj", "model.layers.42.mlp.experts.57.down_proj", "model.layers.42.mlp.experts.58.down_proj", "model.layers.42.mlp.experts.59.down_proj", "model.layers.42.mlp.experts.60.down_proj", "model.layers.42.mlp.experts.61.down_proj", "model.layers.42.mlp.experts.62.down_proj", "model.layers.42.mlp.experts.63.down_proj", "model.layers.42.mlp.experts.64.down_proj", "model.layers.42.mlp.experts.65.down_proj", "model.layers.42.mlp.experts.66.down_proj", "model.layers.42.mlp.experts.67.down_proj", "model.layers.42.mlp.experts.68.down_proj", "model.layers.42.mlp.experts.69.down_proj", "model.layers.42.mlp.experts.70.down_proj", "model.layers.42.mlp.experts.71.down_proj", "model.layers.42.mlp.experts.72.down_proj", "model.layers.42.mlp.experts.73.down_proj", "model.layers.42.mlp.experts.74.down_proj", "model.layers.42.mlp.experts.75.down_proj", "model.layers.42.mlp.experts.76.down_proj", "model.layers.42.mlp.experts.77.down_proj", "model.layers.42.mlp.experts.78.down_proj", "model.layers.42.mlp.experts.79.down_proj", "model.layers.42.mlp.experts.80.down_proj", "model.layers.42.mlp.experts.81.down_proj", "model.layers.42.mlp.experts.82.down_proj", "model.layers.42.mlp.experts.83.down_proj", "model.layers.42.mlp.experts.84.down_proj", "model.layers.42.mlp.experts.85.down_proj", "model.layers.42.mlp.experts.86.down_proj", "model.layers.42.mlp.experts.87.down_proj", "model.layers.42.mlp.experts.88.down_proj", "model.layers.42.mlp.experts.89.down_proj", "model.layers.42.mlp.experts.90.down_proj", "model.layers.42.mlp.experts.91.down_proj", "model.layers.42.mlp.experts.92.down_proj", "model.layers.42.mlp.experts.93.down_proj", "model.layers.42.mlp.experts.94.down_proj", "model.layers.42.mlp.experts.95.down_proj", "model.layers.42.mlp.experts.96.down_proj", "model.layers.42.mlp.experts.97.down_proj", "model.layers.42.mlp.experts.98.down_proj", "model.layers.42.mlp.experts.99.down_proj", "model.layers.42.mlp.experts.100.down_proj", "model.layers.42.mlp.experts.101.down_proj", "model.layers.42.mlp.experts.102.down_proj", "model.layers.42.mlp.experts.103.down_proj", "model.layers.42.mlp.experts.104.down_proj", "model.layers.42.mlp.experts.105.down_proj", "model.layers.42.mlp.experts.106.down_proj", "model.layers.42.mlp.experts.107.down_proj", "model.layers.42.mlp.experts.108.down_proj", "model.layers.42.mlp.experts.109.down_proj", "model.layers.42.mlp.experts.110.down_proj", "model.layers.42.mlp.experts.111.down_proj", "model.layers.42.mlp.experts.112.down_proj", "model.layers.42.mlp.experts.113.down_proj", "model.layers.42.mlp.experts.114.down_proj", "model.layers.42.mlp.experts.115.down_proj", "model.layers.42.mlp.experts.116.down_proj", "model.layers.42.mlp.experts.117.down_proj", "model.layers.42.mlp.experts.118.down_proj", "model.layers.42.mlp.experts.119.down_proj", "model.layers.42.mlp.experts.120.down_proj", "model.layers.42.mlp.experts.121.down_proj", "model.layers.42.mlp.experts.122.down_proj", "model.layers.42.mlp.experts.123.down_proj", "model.layers.42.mlp.experts.124.down_proj", "model.layers.42.mlp.experts.125.down_proj", "model.layers.42.mlp.experts.126.down_proj", "model.layers.42.mlp.experts.127.down_proj", "model.layers.42.mlp.experts.128.down_proj", "model.layers.42.mlp.experts.129.down_proj", "model.layers.42.mlp.experts.130.down_proj", "model.layers.42.mlp.experts.131.down_proj", "model.layers.42.mlp.experts.132.down_proj", "model.layers.42.mlp.experts.133.down_proj", "model.layers.42.mlp.experts.134.down_proj", "model.layers.42.mlp.experts.135.down_proj", "model.layers.42.mlp.experts.136.down_proj", "model.layers.42.mlp.experts.137.down_proj", "model.layers.42.mlp.experts.138.down_proj", "model.layers.42.mlp.experts.139.down_proj", "model.layers.42.mlp.experts.140.down_proj", "model.layers.42.mlp.experts.141.down_proj", "model.layers.42.mlp.experts.142.down_proj", "model.layers.42.mlp.experts.143.down_proj", "model.layers.42.mlp.experts.144.down_proj", "model.layers.42.mlp.experts.145.down_proj", "model.layers.42.mlp.experts.146.down_proj", "model.layers.42.mlp.experts.147.down_proj", "model.layers.42.mlp.experts.148.down_proj", "model.layers.42.mlp.experts.149.down_proj", "model.layers.42.mlp.experts.150.down_proj", "model.layers.42.mlp.experts.151.down_proj", "model.layers.42.mlp.experts.152.down_proj", "model.layers.42.mlp.experts.153.down_proj", "model.layers.42.mlp.experts.154.down_proj", "model.layers.42.mlp.experts.155.down_proj", "model.layers.42.mlp.experts.156.down_proj", "model.layers.42.mlp.experts.157.down_proj", "model.layers.42.mlp.experts.158.down_proj", "model.layers.42.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00016024447977543155, "dbits": 1258291200 }, { "dkld": -0.00028432402759791253, "dbits": 2516582400 }, { "dkld": -0.0003157790750265149, "dbits": 3774873600 }, { "dkld": -0.00028512561693788685, "dbits": 6291456000 } ] }, { "idx": 255, "layers": [ "model.layers.43.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00023118834942580102, "dbits": 62914560 }, { "dkld": -0.0002935929223895073, "dbits": 125829120 }, { "dkld": -0.00034266533330083726, "dbits": 188743680 }, { "dkld": -0.0003784009255468901, "dbits": 314572800 } ] }, { "idx": 256, "layers": [ "model.layers.43.self_attn.k_proj", "model.layers.43.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0008228180930018453, "dbits": 10485760 }, { "dkld": -0.000662013702094566, "dbits": 20971520 }, { "dkld": -0.0007518416270613781, "dbits": 31457280 }, { "dkld": -0.000796511583030235, "dbits": 52428800 } ] }, { "idx": 257, "layers": [ "model.layers.43.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003418907523155268, "dbits": 62914560 }, { "dkld": -0.0005822082050144783, "dbits": 125829120 }, { "dkld": -0.0006719188764691408, "dbits": 188743680 }, { "dkld": -0.0005583700723946206, "dbits": 314572800 } ] }, { "idx": 258, "layers": [ "model.layers.43.mlp.shared_experts.gate_proj", "model.layers.43.mlp.shared_experts.up_proj", "model.layers.43.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -5.345921963453848e-05, "dbits": 23592960 }, { "dkld": 1.3308599591255188e-06, "dbits": 47185920 }, { "dkld": -9.864009916782379e-05, "dbits": 70778880 }, { "dkld": -1.604966819286624e-05, "dbits": 117964800 } ] }, { "idx": 259, "layers": [ "model.layers.43.mlp.experts.0.gate_proj", "model.layers.43.mlp.experts.1.gate_proj", "model.layers.43.mlp.experts.2.gate_proj", "model.layers.43.mlp.experts.3.gate_proj", "model.layers.43.mlp.experts.4.gate_proj", "model.layers.43.mlp.experts.5.gate_proj", "model.layers.43.mlp.experts.6.gate_proj", "model.layers.43.mlp.experts.7.gate_proj", "model.layers.43.mlp.experts.8.gate_proj", "model.layers.43.mlp.experts.9.gate_proj", "model.layers.43.mlp.experts.10.gate_proj", "model.layers.43.mlp.experts.11.gate_proj", "model.layers.43.mlp.experts.12.gate_proj", "model.layers.43.mlp.experts.13.gate_proj", "model.layers.43.mlp.experts.14.gate_proj", "model.layers.43.mlp.experts.15.gate_proj", "model.layers.43.mlp.experts.16.gate_proj", "model.layers.43.mlp.experts.17.gate_proj", "model.layers.43.mlp.experts.18.gate_proj", "model.layers.43.mlp.experts.19.gate_proj", "model.layers.43.mlp.experts.20.gate_proj", "model.layers.43.mlp.experts.21.gate_proj", "model.layers.43.mlp.experts.22.gate_proj", "model.layers.43.mlp.experts.23.gate_proj", "model.layers.43.mlp.experts.24.gate_proj", "model.layers.43.mlp.experts.25.gate_proj", "model.layers.43.mlp.experts.26.gate_proj", "model.layers.43.mlp.experts.27.gate_proj", "model.layers.43.mlp.experts.28.gate_proj", "model.layers.43.mlp.experts.29.gate_proj", "model.layers.43.mlp.experts.30.gate_proj", "model.layers.43.mlp.experts.31.gate_proj", "model.layers.43.mlp.experts.32.gate_proj", "model.layers.43.mlp.experts.33.gate_proj", "model.layers.43.mlp.experts.34.gate_proj", "model.layers.43.mlp.experts.35.gate_proj", "model.layers.43.mlp.experts.36.gate_proj", "model.layers.43.mlp.experts.37.gate_proj", "model.layers.43.mlp.experts.38.gate_proj", "model.layers.43.mlp.experts.39.gate_proj", "model.layers.43.mlp.experts.40.gate_proj", "model.layers.43.mlp.experts.41.gate_proj", "model.layers.43.mlp.experts.42.gate_proj", "model.layers.43.mlp.experts.43.gate_proj", "model.layers.43.mlp.experts.44.gate_proj", "model.layers.43.mlp.experts.45.gate_proj", "model.layers.43.mlp.experts.46.gate_proj", "model.layers.43.mlp.experts.47.gate_proj", "model.layers.43.mlp.experts.48.gate_proj", "model.layers.43.mlp.experts.49.gate_proj", "model.layers.43.mlp.experts.50.gate_proj", "model.layers.43.mlp.experts.51.gate_proj", "model.layers.43.mlp.experts.52.gate_proj", "model.layers.43.mlp.experts.53.gate_proj", "model.layers.43.mlp.experts.54.gate_proj", "model.layers.43.mlp.experts.55.gate_proj", "model.layers.43.mlp.experts.56.gate_proj", "model.layers.43.mlp.experts.57.gate_proj", "model.layers.43.mlp.experts.58.gate_proj", "model.layers.43.mlp.experts.59.gate_proj", "model.layers.43.mlp.experts.60.gate_proj", "model.layers.43.mlp.experts.61.gate_proj", "model.layers.43.mlp.experts.62.gate_proj", "model.layers.43.mlp.experts.63.gate_proj", "model.layers.43.mlp.experts.64.gate_proj", "model.layers.43.mlp.experts.65.gate_proj", "model.layers.43.mlp.experts.66.gate_proj", "model.layers.43.mlp.experts.67.gate_proj", "model.layers.43.mlp.experts.68.gate_proj", "model.layers.43.mlp.experts.69.gate_proj", "model.layers.43.mlp.experts.70.gate_proj", "model.layers.43.mlp.experts.71.gate_proj", "model.layers.43.mlp.experts.72.gate_proj", "model.layers.43.mlp.experts.73.gate_proj", "model.layers.43.mlp.experts.74.gate_proj", "model.layers.43.mlp.experts.75.gate_proj", "model.layers.43.mlp.experts.76.gate_proj", "model.layers.43.mlp.experts.77.gate_proj", "model.layers.43.mlp.experts.78.gate_proj", "model.layers.43.mlp.experts.79.gate_proj", "model.layers.43.mlp.experts.80.gate_proj", "model.layers.43.mlp.experts.81.gate_proj", "model.layers.43.mlp.experts.82.gate_proj", "model.layers.43.mlp.experts.83.gate_proj", "model.layers.43.mlp.experts.84.gate_proj", "model.layers.43.mlp.experts.85.gate_proj", "model.layers.43.mlp.experts.86.gate_proj", "model.layers.43.mlp.experts.87.gate_proj", "model.layers.43.mlp.experts.88.gate_proj", "model.layers.43.mlp.experts.89.gate_proj", "model.layers.43.mlp.experts.90.gate_proj", "model.layers.43.mlp.experts.91.gate_proj", "model.layers.43.mlp.experts.92.gate_proj", "model.layers.43.mlp.experts.93.gate_proj", "model.layers.43.mlp.experts.94.gate_proj", "model.layers.43.mlp.experts.95.gate_proj", "model.layers.43.mlp.experts.96.gate_proj", "model.layers.43.mlp.experts.97.gate_proj", "model.layers.43.mlp.experts.98.gate_proj", "model.layers.43.mlp.experts.99.gate_proj", "model.layers.43.mlp.experts.100.gate_proj", "model.layers.43.mlp.experts.101.gate_proj", "model.layers.43.mlp.experts.102.gate_proj", "model.layers.43.mlp.experts.103.gate_proj", "model.layers.43.mlp.experts.104.gate_proj", "model.layers.43.mlp.experts.105.gate_proj", "model.layers.43.mlp.experts.106.gate_proj", "model.layers.43.mlp.experts.107.gate_proj", "model.layers.43.mlp.experts.108.gate_proj", "model.layers.43.mlp.experts.109.gate_proj", "model.layers.43.mlp.experts.110.gate_proj", "model.layers.43.mlp.experts.111.gate_proj", "model.layers.43.mlp.experts.112.gate_proj", "model.layers.43.mlp.experts.113.gate_proj", "model.layers.43.mlp.experts.114.gate_proj", "model.layers.43.mlp.experts.115.gate_proj", "model.layers.43.mlp.experts.116.gate_proj", "model.layers.43.mlp.experts.117.gate_proj", "model.layers.43.mlp.experts.118.gate_proj", "model.layers.43.mlp.experts.119.gate_proj", "model.layers.43.mlp.experts.120.gate_proj", "model.layers.43.mlp.experts.121.gate_proj", "model.layers.43.mlp.experts.122.gate_proj", "model.layers.43.mlp.experts.123.gate_proj", "model.layers.43.mlp.experts.124.gate_proj", "model.layers.43.mlp.experts.125.gate_proj", "model.layers.43.mlp.experts.126.gate_proj", "model.layers.43.mlp.experts.127.gate_proj", "model.layers.43.mlp.experts.128.gate_proj", "model.layers.43.mlp.experts.129.gate_proj", "model.layers.43.mlp.experts.130.gate_proj", "model.layers.43.mlp.experts.131.gate_proj", "model.layers.43.mlp.experts.132.gate_proj", "model.layers.43.mlp.experts.133.gate_proj", "model.layers.43.mlp.experts.134.gate_proj", "model.layers.43.mlp.experts.135.gate_proj", "model.layers.43.mlp.experts.136.gate_proj", "model.layers.43.mlp.experts.137.gate_proj", "model.layers.43.mlp.experts.138.gate_proj", "model.layers.43.mlp.experts.139.gate_proj", "model.layers.43.mlp.experts.140.gate_proj", "model.layers.43.mlp.experts.141.gate_proj", "model.layers.43.mlp.experts.142.gate_proj", "model.layers.43.mlp.experts.143.gate_proj", "model.layers.43.mlp.experts.144.gate_proj", "model.layers.43.mlp.experts.145.gate_proj", "model.layers.43.mlp.experts.146.gate_proj", "model.layers.43.mlp.experts.147.gate_proj", "model.layers.43.mlp.experts.148.gate_proj", "model.layers.43.mlp.experts.149.gate_proj", "model.layers.43.mlp.experts.150.gate_proj", "model.layers.43.mlp.experts.151.gate_proj", "model.layers.43.mlp.experts.152.gate_proj", "model.layers.43.mlp.experts.153.gate_proj", "model.layers.43.mlp.experts.154.gate_proj", "model.layers.43.mlp.experts.155.gate_proj", "model.layers.43.mlp.experts.156.gate_proj", "model.layers.43.mlp.experts.157.gate_proj", "model.layers.43.mlp.experts.158.gate_proj", "model.layers.43.mlp.experts.159.gate_proj", "model.layers.43.mlp.experts.0.up_proj", "model.layers.43.mlp.experts.1.up_proj", "model.layers.43.mlp.experts.2.up_proj", "model.layers.43.mlp.experts.3.up_proj", "model.layers.43.mlp.experts.4.up_proj", "model.layers.43.mlp.experts.5.up_proj", "model.layers.43.mlp.experts.6.up_proj", "model.layers.43.mlp.experts.7.up_proj", "model.layers.43.mlp.experts.8.up_proj", "model.layers.43.mlp.experts.9.up_proj", "model.layers.43.mlp.experts.10.up_proj", "model.layers.43.mlp.experts.11.up_proj", "model.layers.43.mlp.experts.12.up_proj", "model.layers.43.mlp.experts.13.up_proj", "model.layers.43.mlp.experts.14.up_proj", "model.layers.43.mlp.experts.15.up_proj", "model.layers.43.mlp.experts.16.up_proj", "model.layers.43.mlp.experts.17.up_proj", "model.layers.43.mlp.experts.18.up_proj", "model.layers.43.mlp.experts.19.up_proj", "model.layers.43.mlp.experts.20.up_proj", "model.layers.43.mlp.experts.21.up_proj", "model.layers.43.mlp.experts.22.up_proj", "model.layers.43.mlp.experts.23.up_proj", "model.layers.43.mlp.experts.24.up_proj", "model.layers.43.mlp.experts.25.up_proj", "model.layers.43.mlp.experts.26.up_proj", "model.layers.43.mlp.experts.27.up_proj", "model.layers.43.mlp.experts.28.up_proj", "model.layers.43.mlp.experts.29.up_proj", "model.layers.43.mlp.experts.30.up_proj", "model.layers.43.mlp.experts.31.up_proj", "model.layers.43.mlp.experts.32.up_proj", "model.layers.43.mlp.experts.33.up_proj", "model.layers.43.mlp.experts.34.up_proj", "model.layers.43.mlp.experts.35.up_proj", "model.layers.43.mlp.experts.36.up_proj", "model.layers.43.mlp.experts.37.up_proj", "model.layers.43.mlp.experts.38.up_proj", "model.layers.43.mlp.experts.39.up_proj", "model.layers.43.mlp.experts.40.up_proj", "model.layers.43.mlp.experts.41.up_proj", "model.layers.43.mlp.experts.42.up_proj", "model.layers.43.mlp.experts.43.up_proj", "model.layers.43.mlp.experts.44.up_proj", "model.layers.43.mlp.experts.45.up_proj", "model.layers.43.mlp.experts.46.up_proj", "model.layers.43.mlp.experts.47.up_proj", "model.layers.43.mlp.experts.48.up_proj", "model.layers.43.mlp.experts.49.up_proj", "model.layers.43.mlp.experts.50.up_proj", "model.layers.43.mlp.experts.51.up_proj", "model.layers.43.mlp.experts.52.up_proj", "model.layers.43.mlp.experts.53.up_proj", "model.layers.43.mlp.experts.54.up_proj", "model.layers.43.mlp.experts.55.up_proj", "model.layers.43.mlp.experts.56.up_proj", "model.layers.43.mlp.experts.57.up_proj", "model.layers.43.mlp.experts.58.up_proj", "model.layers.43.mlp.experts.59.up_proj", "model.layers.43.mlp.experts.60.up_proj", "model.layers.43.mlp.experts.61.up_proj", "model.layers.43.mlp.experts.62.up_proj", "model.layers.43.mlp.experts.63.up_proj", "model.layers.43.mlp.experts.64.up_proj", "model.layers.43.mlp.experts.65.up_proj", "model.layers.43.mlp.experts.66.up_proj", "model.layers.43.mlp.experts.67.up_proj", "model.layers.43.mlp.experts.68.up_proj", "model.layers.43.mlp.experts.69.up_proj", "model.layers.43.mlp.experts.70.up_proj", "model.layers.43.mlp.experts.71.up_proj", "model.layers.43.mlp.experts.72.up_proj", "model.layers.43.mlp.experts.73.up_proj", "model.layers.43.mlp.experts.74.up_proj", "model.layers.43.mlp.experts.75.up_proj", "model.layers.43.mlp.experts.76.up_proj", "model.layers.43.mlp.experts.77.up_proj", "model.layers.43.mlp.experts.78.up_proj", "model.layers.43.mlp.experts.79.up_proj", "model.layers.43.mlp.experts.80.up_proj", "model.layers.43.mlp.experts.81.up_proj", "model.layers.43.mlp.experts.82.up_proj", "model.layers.43.mlp.experts.83.up_proj", "model.layers.43.mlp.experts.84.up_proj", "model.layers.43.mlp.experts.85.up_proj", "model.layers.43.mlp.experts.86.up_proj", "model.layers.43.mlp.experts.87.up_proj", "model.layers.43.mlp.experts.88.up_proj", "model.layers.43.mlp.experts.89.up_proj", "model.layers.43.mlp.experts.90.up_proj", "model.layers.43.mlp.experts.91.up_proj", "model.layers.43.mlp.experts.92.up_proj", "model.layers.43.mlp.experts.93.up_proj", "model.layers.43.mlp.experts.94.up_proj", "model.layers.43.mlp.experts.95.up_proj", "model.layers.43.mlp.experts.96.up_proj", "model.layers.43.mlp.experts.97.up_proj", "model.layers.43.mlp.experts.98.up_proj", "model.layers.43.mlp.experts.99.up_proj", "model.layers.43.mlp.experts.100.up_proj", "model.layers.43.mlp.experts.101.up_proj", "model.layers.43.mlp.experts.102.up_proj", "model.layers.43.mlp.experts.103.up_proj", "model.layers.43.mlp.experts.104.up_proj", "model.layers.43.mlp.experts.105.up_proj", "model.layers.43.mlp.experts.106.up_proj", "model.layers.43.mlp.experts.107.up_proj", "model.layers.43.mlp.experts.108.up_proj", "model.layers.43.mlp.experts.109.up_proj", "model.layers.43.mlp.experts.110.up_proj", "model.layers.43.mlp.experts.111.up_proj", "model.layers.43.mlp.experts.112.up_proj", "model.layers.43.mlp.experts.113.up_proj", "model.layers.43.mlp.experts.114.up_proj", "model.layers.43.mlp.experts.115.up_proj", "model.layers.43.mlp.experts.116.up_proj", "model.layers.43.mlp.experts.117.up_proj", "model.layers.43.mlp.experts.118.up_proj", "model.layers.43.mlp.experts.119.up_proj", "model.layers.43.mlp.experts.120.up_proj", "model.layers.43.mlp.experts.121.up_proj", "model.layers.43.mlp.experts.122.up_proj", "model.layers.43.mlp.experts.123.up_proj", "model.layers.43.mlp.experts.124.up_proj", "model.layers.43.mlp.experts.125.up_proj", "model.layers.43.mlp.experts.126.up_proj", "model.layers.43.mlp.experts.127.up_proj", "model.layers.43.mlp.experts.128.up_proj", "model.layers.43.mlp.experts.129.up_proj", "model.layers.43.mlp.experts.130.up_proj", "model.layers.43.mlp.experts.131.up_proj", "model.layers.43.mlp.experts.132.up_proj", "model.layers.43.mlp.experts.133.up_proj", "model.layers.43.mlp.experts.134.up_proj", "model.layers.43.mlp.experts.135.up_proj", "model.layers.43.mlp.experts.136.up_proj", "model.layers.43.mlp.experts.137.up_proj", "model.layers.43.mlp.experts.138.up_proj", "model.layers.43.mlp.experts.139.up_proj", "model.layers.43.mlp.experts.140.up_proj", "model.layers.43.mlp.experts.141.up_proj", "model.layers.43.mlp.experts.142.up_proj", "model.layers.43.mlp.experts.143.up_proj", "model.layers.43.mlp.experts.144.up_proj", "model.layers.43.mlp.experts.145.up_proj", "model.layers.43.mlp.experts.146.up_proj", "model.layers.43.mlp.experts.147.up_proj", "model.layers.43.mlp.experts.148.up_proj", "model.layers.43.mlp.experts.149.up_proj", "model.layers.43.mlp.experts.150.up_proj", "model.layers.43.mlp.experts.151.up_proj", "model.layers.43.mlp.experts.152.up_proj", "model.layers.43.mlp.experts.153.up_proj", "model.layers.43.mlp.experts.154.up_proj", "model.layers.43.mlp.experts.155.up_proj", "model.layers.43.mlp.experts.156.up_proj", "model.layers.43.mlp.experts.157.up_proj", "model.layers.43.mlp.experts.158.up_proj", "model.layers.43.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0005431108176708332, "dbits": 2516582400 }, { "dkld": -0.0005882827565074078, "dbits": 5033164800 }, { "dkld": -0.0005743797868490302, "dbits": 7549747200 }, { "dkld": -0.0005945520475506838, "dbits": 12582912000 } ] }, { "idx": 260, "layers": [ "model.layers.43.mlp.experts.0.down_proj", "model.layers.43.mlp.experts.1.down_proj", "model.layers.43.mlp.experts.2.down_proj", "model.layers.43.mlp.experts.3.down_proj", "model.layers.43.mlp.experts.4.down_proj", "model.layers.43.mlp.experts.5.down_proj", "model.layers.43.mlp.experts.6.down_proj", "model.layers.43.mlp.experts.7.down_proj", "model.layers.43.mlp.experts.8.down_proj", "model.layers.43.mlp.experts.9.down_proj", "model.layers.43.mlp.experts.10.down_proj", "model.layers.43.mlp.experts.11.down_proj", "model.layers.43.mlp.experts.12.down_proj", "model.layers.43.mlp.experts.13.down_proj", "model.layers.43.mlp.experts.14.down_proj", "model.layers.43.mlp.experts.15.down_proj", "model.layers.43.mlp.experts.16.down_proj", "model.layers.43.mlp.experts.17.down_proj", "model.layers.43.mlp.experts.18.down_proj", "model.layers.43.mlp.experts.19.down_proj", "model.layers.43.mlp.experts.20.down_proj", "model.layers.43.mlp.experts.21.down_proj", "model.layers.43.mlp.experts.22.down_proj", "model.layers.43.mlp.experts.23.down_proj", "model.layers.43.mlp.experts.24.down_proj", "model.layers.43.mlp.experts.25.down_proj", "model.layers.43.mlp.experts.26.down_proj", "model.layers.43.mlp.experts.27.down_proj", "model.layers.43.mlp.experts.28.down_proj", "model.layers.43.mlp.experts.29.down_proj", "model.layers.43.mlp.experts.30.down_proj", "model.layers.43.mlp.experts.31.down_proj", "model.layers.43.mlp.experts.32.down_proj", "model.layers.43.mlp.experts.33.down_proj", "model.layers.43.mlp.experts.34.down_proj", "model.layers.43.mlp.experts.35.down_proj", "model.layers.43.mlp.experts.36.down_proj", "model.layers.43.mlp.experts.37.down_proj", "model.layers.43.mlp.experts.38.down_proj", "model.layers.43.mlp.experts.39.down_proj", "model.layers.43.mlp.experts.40.down_proj", "model.layers.43.mlp.experts.41.down_proj", "model.layers.43.mlp.experts.42.down_proj", "model.layers.43.mlp.experts.43.down_proj", "model.layers.43.mlp.experts.44.down_proj", "model.layers.43.mlp.experts.45.down_proj", "model.layers.43.mlp.experts.46.down_proj", "model.layers.43.mlp.experts.47.down_proj", "model.layers.43.mlp.experts.48.down_proj", "model.layers.43.mlp.experts.49.down_proj", "model.layers.43.mlp.experts.50.down_proj", "model.layers.43.mlp.experts.51.down_proj", "model.layers.43.mlp.experts.52.down_proj", "model.layers.43.mlp.experts.53.down_proj", "model.layers.43.mlp.experts.54.down_proj", "model.layers.43.mlp.experts.55.down_proj", "model.layers.43.mlp.experts.56.down_proj", "model.layers.43.mlp.experts.57.down_proj", "model.layers.43.mlp.experts.58.down_proj", "model.layers.43.mlp.experts.59.down_proj", "model.layers.43.mlp.experts.60.down_proj", "model.layers.43.mlp.experts.61.down_proj", "model.layers.43.mlp.experts.62.down_proj", "model.layers.43.mlp.experts.63.down_proj", "model.layers.43.mlp.experts.64.down_proj", "model.layers.43.mlp.experts.65.down_proj", "model.layers.43.mlp.experts.66.down_proj", "model.layers.43.mlp.experts.67.down_proj", "model.layers.43.mlp.experts.68.down_proj", "model.layers.43.mlp.experts.69.down_proj", "model.layers.43.mlp.experts.70.down_proj", "model.layers.43.mlp.experts.71.down_proj", "model.layers.43.mlp.experts.72.down_proj", "model.layers.43.mlp.experts.73.down_proj", "model.layers.43.mlp.experts.74.down_proj", "model.layers.43.mlp.experts.75.down_proj", "model.layers.43.mlp.experts.76.down_proj", "model.layers.43.mlp.experts.77.down_proj", "model.layers.43.mlp.experts.78.down_proj", "model.layers.43.mlp.experts.79.down_proj", "model.layers.43.mlp.experts.80.down_proj", "model.layers.43.mlp.experts.81.down_proj", "model.layers.43.mlp.experts.82.down_proj", "model.layers.43.mlp.experts.83.down_proj", "model.layers.43.mlp.experts.84.down_proj", "model.layers.43.mlp.experts.85.down_proj", "model.layers.43.mlp.experts.86.down_proj", "model.layers.43.mlp.experts.87.down_proj", "model.layers.43.mlp.experts.88.down_proj", "model.layers.43.mlp.experts.89.down_proj", "model.layers.43.mlp.experts.90.down_proj", "model.layers.43.mlp.experts.91.down_proj", "model.layers.43.mlp.experts.92.down_proj", "model.layers.43.mlp.experts.93.down_proj", "model.layers.43.mlp.experts.94.down_proj", "model.layers.43.mlp.experts.95.down_proj", "model.layers.43.mlp.experts.96.down_proj", "model.layers.43.mlp.experts.97.down_proj", "model.layers.43.mlp.experts.98.down_proj", "model.layers.43.mlp.experts.99.down_proj", "model.layers.43.mlp.experts.100.down_proj", "model.layers.43.mlp.experts.101.down_proj", "model.layers.43.mlp.experts.102.down_proj", "model.layers.43.mlp.experts.103.down_proj", "model.layers.43.mlp.experts.104.down_proj", "model.layers.43.mlp.experts.105.down_proj", "model.layers.43.mlp.experts.106.down_proj", "model.layers.43.mlp.experts.107.down_proj", "model.layers.43.mlp.experts.108.down_proj", "model.layers.43.mlp.experts.109.down_proj", "model.layers.43.mlp.experts.110.down_proj", "model.layers.43.mlp.experts.111.down_proj", "model.layers.43.mlp.experts.112.down_proj", "model.layers.43.mlp.experts.113.down_proj", "model.layers.43.mlp.experts.114.down_proj", "model.layers.43.mlp.experts.115.down_proj", "model.layers.43.mlp.experts.116.down_proj", "model.layers.43.mlp.experts.117.down_proj", "model.layers.43.mlp.experts.118.down_proj", "model.layers.43.mlp.experts.119.down_proj", "model.layers.43.mlp.experts.120.down_proj", "model.layers.43.mlp.experts.121.down_proj", "model.layers.43.mlp.experts.122.down_proj", "model.layers.43.mlp.experts.123.down_proj", "model.layers.43.mlp.experts.124.down_proj", "model.layers.43.mlp.experts.125.down_proj", "model.layers.43.mlp.experts.126.down_proj", "model.layers.43.mlp.experts.127.down_proj", "model.layers.43.mlp.experts.128.down_proj", "model.layers.43.mlp.experts.129.down_proj", "model.layers.43.mlp.experts.130.down_proj", "model.layers.43.mlp.experts.131.down_proj", "model.layers.43.mlp.experts.132.down_proj", "model.layers.43.mlp.experts.133.down_proj", "model.layers.43.mlp.experts.134.down_proj", "model.layers.43.mlp.experts.135.down_proj", "model.layers.43.mlp.experts.136.down_proj", "model.layers.43.mlp.experts.137.down_proj", "model.layers.43.mlp.experts.138.down_proj", "model.layers.43.mlp.experts.139.down_proj", "model.layers.43.mlp.experts.140.down_proj", "model.layers.43.mlp.experts.141.down_proj", "model.layers.43.mlp.experts.142.down_proj", "model.layers.43.mlp.experts.143.down_proj", "model.layers.43.mlp.experts.144.down_proj", "model.layers.43.mlp.experts.145.down_proj", "model.layers.43.mlp.experts.146.down_proj", "model.layers.43.mlp.experts.147.down_proj", "model.layers.43.mlp.experts.148.down_proj", "model.layers.43.mlp.experts.149.down_proj", "model.layers.43.mlp.experts.150.down_proj", "model.layers.43.mlp.experts.151.down_proj", "model.layers.43.mlp.experts.152.down_proj", "model.layers.43.mlp.experts.153.down_proj", "model.layers.43.mlp.experts.154.down_proj", "model.layers.43.mlp.experts.155.down_proj", "model.layers.43.mlp.experts.156.down_proj", "model.layers.43.mlp.experts.157.down_proj", "model.layers.43.mlp.experts.158.down_proj", "model.layers.43.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0004460261203348692, "dbits": 1258291200 }, { "dkld": -0.0003895206376910293, "dbits": 2516582400 }, { "dkld": -0.00037266463041306375, "dbits": 3774873600 }, { "dkld": -0.00039815120398999093, "dbits": 6291456000 } ] }, { "idx": 261, "layers": [ "model.layers.44.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0005585476756095886, "dbits": 62914560 }, { "dkld": -0.0003655569627881161, "dbits": 125829120 }, { "dkld": -0.0004176380112767275, "dbits": 188743680 }, { "dkld": -0.00041979439556599774, "dbits": 314572800 } ] }, { "idx": 262, "layers": [ "model.layers.44.self_attn.k_proj", "model.layers.44.self_attn.v_proj" ], "candidates": [ { "dkld": 7.977876812219342e-05, "dbits": 10485760 }, { "dkld": 0.0004929272457957212, "dbits": 20971520 }, { "dkld": 0.0004989545792341149, "dbits": 31457280 }, { "dkld": 0.0004396048374473993, "dbits": 52428800 } ] }, { "idx": 263, "layers": [ "model.layers.44.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00040725115686655877, "dbits": 62914560 }, { "dkld": -0.0007876258343458287, "dbits": 125829120 }, { "dkld": -0.0007811006158590372, "dbits": 188743680 }, { "dkld": -0.0008329184725880678, "dbits": 314572800 } ] }, { "idx": 264, "layers": [ "model.layers.44.mlp.shared_experts.gate_proj", "model.layers.44.mlp.shared_experts.up_proj", "model.layers.44.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 5.0448998808849677e-05, "dbits": 23592960 }, { "dkld": -0.00015503466129303256, "dbits": 47185920 }, { "dkld": 2.779234200715741e-05, "dbits": 70778880 }, { "dkld": -2.8376653790476714e-05, "dbits": 117964800 } ] }, { "idx": 265, "layers": [ "model.layers.44.mlp.experts.0.gate_proj", "model.layers.44.mlp.experts.1.gate_proj", "model.layers.44.mlp.experts.2.gate_proj", "model.layers.44.mlp.experts.3.gate_proj", "model.layers.44.mlp.experts.4.gate_proj", "model.layers.44.mlp.experts.5.gate_proj", "model.layers.44.mlp.experts.6.gate_proj", "model.layers.44.mlp.experts.7.gate_proj", "model.layers.44.mlp.experts.8.gate_proj", "model.layers.44.mlp.experts.9.gate_proj", "model.layers.44.mlp.experts.10.gate_proj", "model.layers.44.mlp.experts.11.gate_proj", "model.layers.44.mlp.experts.12.gate_proj", "model.layers.44.mlp.experts.13.gate_proj", "model.layers.44.mlp.experts.14.gate_proj", "model.layers.44.mlp.experts.15.gate_proj", "model.layers.44.mlp.experts.16.gate_proj", "model.layers.44.mlp.experts.17.gate_proj", "model.layers.44.mlp.experts.18.gate_proj", "model.layers.44.mlp.experts.19.gate_proj", "model.layers.44.mlp.experts.20.gate_proj", "model.layers.44.mlp.experts.21.gate_proj", "model.layers.44.mlp.experts.22.gate_proj", "model.layers.44.mlp.experts.23.gate_proj", "model.layers.44.mlp.experts.24.gate_proj", "model.layers.44.mlp.experts.25.gate_proj", "model.layers.44.mlp.experts.26.gate_proj", "model.layers.44.mlp.experts.27.gate_proj", "model.layers.44.mlp.experts.28.gate_proj", "model.layers.44.mlp.experts.29.gate_proj", "model.layers.44.mlp.experts.30.gate_proj", "model.layers.44.mlp.experts.31.gate_proj", "model.layers.44.mlp.experts.32.gate_proj", "model.layers.44.mlp.experts.33.gate_proj", "model.layers.44.mlp.experts.34.gate_proj", "model.layers.44.mlp.experts.35.gate_proj", "model.layers.44.mlp.experts.36.gate_proj", "model.layers.44.mlp.experts.37.gate_proj", "model.layers.44.mlp.experts.38.gate_proj", "model.layers.44.mlp.experts.39.gate_proj", "model.layers.44.mlp.experts.40.gate_proj", "model.layers.44.mlp.experts.41.gate_proj", "model.layers.44.mlp.experts.42.gate_proj", "model.layers.44.mlp.experts.43.gate_proj", "model.layers.44.mlp.experts.44.gate_proj", "model.layers.44.mlp.experts.45.gate_proj", "model.layers.44.mlp.experts.46.gate_proj", "model.layers.44.mlp.experts.47.gate_proj", "model.layers.44.mlp.experts.48.gate_proj", "model.layers.44.mlp.experts.49.gate_proj", "model.layers.44.mlp.experts.50.gate_proj", "model.layers.44.mlp.experts.51.gate_proj", "model.layers.44.mlp.experts.52.gate_proj", "model.layers.44.mlp.experts.53.gate_proj", "model.layers.44.mlp.experts.54.gate_proj", "model.layers.44.mlp.experts.55.gate_proj", "model.layers.44.mlp.experts.56.gate_proj", "model.layers.44.mlp.experts.57.gate_proj", "model.layers.44.mlp.experts.58.gate_proj", "model.layers.44.mlp.experts.59.gate_proj", "model.layers.44.mlp.experts.60.gate_proj", "model.layers.44.mlp.experts.61.gate_proj", "model.layers.44.mlp.experts.62.gate_proj", "model.layers.44.mlp.experts.63.gate_proj", "model.layers.44.mlp.experts.64.gate_proj", "model.layers.44.mlp.experts.65.gate_proj", "model.layers.44.mlp.experts.66.gate_proj", "model.layers.44.mlp.experts.67.gate_proj", "model.layers.44.mlp.experts.68.gate_proj", "model.layers.44.mlp.experts.69.gate_proj", "model.layers.44.mlp.experts.70.gate_proj", "model.layers.44.mlp.experts.71.gate_proj", "model.layers.44.mlp.experts.72.gate_proj", "model.layers.44.mlp.experts.73.gate_proj", "model.layers.44.mlp.experts.74.gate_proj", "model.layers.44.mlp.experts.75.gate_proj", "model.layers.44.mlp.experts.76.gate_proj", "model.layers.44.mlp.experts.77.gate_proj", "model.layers.44.mlp.experts.78.gate_proj", "model.layers.44.mlp.experts.79.gate_proj", "model.layers.44.mlp.experts.80.gate_proj", "model.layers.44.mlp.experts.81.gate_proj", "model.layers.44.mlp.experts.82.gate_proj", "model.layers.44.mlp.experts.83.gate_proj", "model.layers.44.mlp.experts.84.gate_proj", "model.layers.44.mlp.experts.85.gate_proj", "model.layers.44.mlp.experts.86.gate_proj", "model.layers.44.mlp.experts.87.gate_proj", "model.layers.44.mlp.experts.88.gate_proj", "model.layers.44.mlp.experts.89.gate_proj", "model.layers.44.mlp.experts.90.gate_proj", "model.layers.44.mlp.experts.91.gate_proj", "model.layers.44.mlp.experts.92.gate_proj", "model.layers.44.mlp.experts.93.gate_proj", "model.layers.44.mlp.experts.94.gate_proj", "model.layers.44.mlp.experts.95.gate_proj", "model.layers.44.mlp.experts.96.gate_proj", "model.layers.44.mlp.experts.97.gate_proj", "model.layers.44.mlp.experts.98.gate_proj", "model.layers.44.mlp.experts.99.gate_proj", "model.layers.44.mlp.experts.100.gate_proj", "model.layers.44.mlp.experts.101.gate_proj", "model.layers.44.mlp.experts.102.gate_proj", "model.layers.44.mlp.experts.103.gate_proj", "model.layers.44.mlp.experts.104.gate_proj", "model.layers.44.mlp.experts.105.gate_proj", "model.layers.44.mlp.experts.106.gate_proj", "model.layers.44.mlp.experts.107.gate_proj", "model.layers.44.mlp.experts.108.gate_proj", "model.layers.44.mlp.experts.109.gate_proj", "model.layers.44.mlp.experts.110.gate_proj", "model.layers.44.mlp.experts.111.gate_proj", "model.layers.44.mlp.experts.112.gate_proj", "model.layers.44.mlp.experts.113.gate_proj", "model.layers.44.mlp.experts.114.gate_proj", "model.layers.44.mlp.experts.115.gate_proj", "model.layers.44.mlp.experts.116.gate_proj", "model.layers.44.mlp.experts.117.gate_proj", "model.layers.44.mlp.experts.118.gate_proj", "model.layers.44.mlp.experts.119.gate_proj", "model.layers.44.mlp.experts.120.gate_proj", "model.layers.44.mlp.experts.121.gate_proj", "model.layers.44.mlp.experts.122.gate_proj", "model.layers.44.mlp.experts.123.gate_proj", "model.layers.44.mlp.experts.124.gate_proj", "model.layers.44.mlp.experts.125.gate_proj", "model.layers.44.mlp.experts.126.gate_proj", "model.layers.44.mlp.experts.127.gate_proj", "model.layers.44.mlp.experts.128.gate_proj", "model.layers.44.mlp.experts.129.gate_proj", "model.layers.44.mlp.experts.130.gate_proj", "model.layers.44.mlp.experts.131.gate_proj", "model.layers.44.mlp.experts.132.gate_proj", "model.layers.44.mlp.experts.133.gate_proj", "model.layers.44.mlp.experts.134.gate_proj", "model.layers.44.mlp.experts.135.gate_proj", "model.layers.44.mlp.experts.136.gate_proj", "model.layers.44.mlp.experts.137.gate_proj", "model.layers.44.mlp.experts.138.gate_proj", "model.layers.44.mlp.experts.139.gate_proj", "model.layers.44.mlp.experts.140.gate_proj", "model.layers.44.mlp.experts.141.gate_proj", "model.layers.44.mlp.experts.142.gate_proj", "model.layers.44.mlp.experts.143.gate_proj", "model.layers.44.mlp.experts.144.gate_proj", "model.layers.44.mlp.experts.145.gate_proj", "model.layers.44.mlp.experts.146.gate_proj", "model.layers.44.mlp.experts.147.gate_proj", "model.layers.44.mlp.experts.148.gate_proj", "model.layers.44.mlp.experts.149.gate_proj", "model.layers.44.mlp.experts.150.gate_proj", "model.layers.44.mlp.experts.151.gate_proj", "model.layers.44.mlp.experts.152.gate_proj", "model.layers.44.mlp.experts.153.gate_proj", "model.layers.44.mlp.experts.154.gate_proj", "model.layers.44.mlp.experts.155.gate_proj", "model.layers.44.mlp.experts.156.gate_proj", "model.layers.44.mlp.experts.157.gate_proj", "model.layers.44.mlp.experts.158.gate_proj", "model.layers.44.mlp.experts.159.gate_proj", "model.layers.44.mlp.experts.0.up_proj", "model.layers.44.mlp.experts.1.up_proj", "model.layers.44.mlp.experts.2.up_proj", "model.layers.44.mlp.experts.3.up_proj", "model.layers.44.mlp.experts.4.up_proj", "model.layers.44.mlp.experts.5.up_proj", "model.layers.44.mlp.experts.6.up_proj", "model.layers.44.mlp.experts.7.up_proj", "model.layers.44.mlp.experts.8.up_proj", "model.layers.44.mlp.experts.9.up_proj", "model.layers.44.mlp.experts.10.up_proj", "model.layers.44.mlp.experts.11.up_proj", "model.layers.44.mlp.experts.12.up_proj", "model.layers.44.mlp.experts.13.up_proj", "model.layers.44.mlp.experts.14.up_proj", "model.layers.44.mlp.experts.15.up_proj", "model.layers.44.mlp.experts.16.up_proj", "model.layers.44.mlp.experts.17.up_proj", "model.layers.44.mlp.experts.18.up_proj", "model.layers.44.mlp.experts.19.up_proj", "model.layers.44.mlp.experts.20.up_proj", "model.layers.44.mlp.experts.21.up_proj", "model.layers.44.mlp.experts.22.up_proj", "model.layers.44.mlp.experts.23.up_proj", "model.layers.44.mlp.experts.24.up_proj", "model.layers.44.mlp.experts.25.up_proj", "model.layers.44.mlp.experts.26.up_proj", "model.layers.44.mlp.experts.27.up_proj", "model.layers.44.mlp.experts.28.up_proj", "model.layers.44.mlp.experts.29.up_proj", "model.layers.44.mlp.experts.30.up_proj", "model.layers.44.mlp.experts.31.up_proj", "model.layers.44.mlp.experts.32.up_proj", "model.layers.44.mlp.experts.33.up_proj", "model.layers.44.mlp.experts.34.up_proj", "model.layers.44.mlp.experts.35.up_proj", "model.layers.44.mlp.experts.36.up_proj", "model.layers.44.mlp.experts.37.up_proj", "model.layers.44.mlp.experts.38.up_proj", "model.layers.44.mlp.experts.39.up_proj", "model.layers.44.mlp.experts.40.up_proj", "model.layers.44.mlp.experts.41.up_proj", "model.layers.44.mlp.experts.42.up_proj", "model.layers.44.mlp.experts.43.up_proj", "model.layers.44.mlp.experts.44.up_proj", "model.layers.44.mlp.experts.45.up_proj", "model.layers.44.mlp.experts.46.up_proj", "model.layers.44.mlp.experts.47.up_proj", "model.layers.44.mlp.experts.48.up_proj", "model.layers.44.mlp.experts.49.up_proj", "model.layers.44.mlp.experts.50.up_proj", "model.layers.44.mlp.experts.51.up_proj", "model.layers.44.mlp.experts.52.up_proj", "model.layers.44.mlp.experts.53.up_proj", "model.layers.44.mlp.experts.54.up_proj", "model.layers.44.mlp.experts.55.up_proj", "model.layers.44.mlp.experts.56.up_proj", "model.layers.44.mlp.experts.57.up_proj", "model.layers.44.mlp.experts.58.up_proj", "model.layers.44.mlp.experts.59.up_proj", "model.layers.44.mlp.experts.60.up_proj", "model.layers.44.mlp.experts.61.up_proj", "model.layers.44.mlp.experts.62.up_proj", "model.layers.44.mlp.experts.63.up_proj", "model.layers.44.mlp.experts.64.up_proj", "model.layers.44.mlp.experts.65.up_proj", "model.layers.44.mlp.experts.66.up_proj", "model.layers.44.mlp.experts.67.up_proj", "model.layers.44.mlp.experts.68.up_proj", "model.layers.44.mlp.experts.69.up_proj", "model.layers.44.mlp.experts.70.up_proj", "model.layers.44.mlp.experts.71.up_proj", "model.layers.44.mlp.experts.72.up_proj", "model.layers.44.mlp.experts.73.up_proj", "model.layers.44.mlp.experts.74.up_proj", "model.layers.44.mlp.experts.75.up_proj", "model.layers.44.mlp.experts.76.up_proj", "model.layers.44.mlp.experts.77.up_proj", "model.layers.44.mlp.experts.78.up_proj", "model.layers.44.mlp.experts.79.up_proj", "model.layers.44.mlp.experts.80.up_proj", "model.layers.44.mlp.experts.81.up_proj", "model.layers.44.mlp.experts.82.up_proj", "model.layers.44.mlp.experts.83.up_proj", "model.layers.44.mlp.experts.84.up_proj", "model.layers.44.mlp.experts.85.up_proj", "model.layers.44.mlp.experts.86.up_proj", "model.layers.44.mlp.experts.87.up_proj", "model.layers.44.mlp.experts.88.up_proj", "model.layers.44.mlp.experts.89.up_proj", "model.layers.44.mlp.experts.90.up_proj", "model.layers.44.mlp.experts.91.up_proj", "model.layers.44.mlp.experts.92.up_proj", "model.layers.44.mlp.experts.93.up_proj", "model.layers.44.mlp.experts.94.up_proj", "model.layers.44.mlp.experts.95.up_proj", "model.layers.44.mlp.experts.96.up_proj", "model.layers.44.mlp.experts.97.up_proj", "model.layers.44.mlp.experts.98.up_proj", "model.layers.44.mlp.experts.99.up_proj", "model.layers.44.mlp.experts.100.up_proj", "model.layers.44.mlp.experts.101.up_proj", "model.layers.44.mlp.experts.102.up_proj", "model.layers.44.mlp.experts.103.up_proj", "model.layers.44.mlp.experts.104.up_proj", "model.layers.44.mlp.experts.105.up_proj", "model.layers.44.mlp.experts.106.up_proj", "model.layers.44.mlp.experts.107.up_proj", "model.layers.44.mlp.experts.108.up_proj", "model.layers.44.mlp.experts.109.up_proj", "model.layers.44.mlp.experts.110.up_proj", "model.layers.44.mlp.experts.111.up_proj", "model.layers.44.mlp.experts.112.up_proj", "model.layers.44.mlp.experts.113.up_proj", "model.layers.44.mlp.experts.114.up_proj", "model.layers.44.mlp.experts.115.up_proj", "model.layers.44.mlp.experts.116.up_proj", "model.layers.44.mlp.experts.117.up_proj", "model.layers.44.mlp.experts.118.up_proj", "model.layers.44.mlp.experts.119.up_proj", "model.layers.44.mlp.experts.120.up_proj", "model.layers.44.mlp.experts.121.up_proj", "model.layers.44.mlp.experts.122.up_proj", "model.layers.44.mlp.experts.123.up_proj", "model.layers.44.mlp.experts.124.up_proj", "model.layers.44.mlp.experts.125.up_proj", "model.layers.44.mlp.experts.126.up_proj", "model.layers.44.mlp.experts.127.up_proj", "model.layers.44.mlp.experts.128.up_proj", "model.layers.44.mlp.experts.129.up_proj", "model.layers.44.mlp.experts.130.up_proj", "model.layers.44.mlp.experts.131.up_proj", "model.layers.44.mlp.experts.132.up_proj", "model.layers.44.mlp.experts.133.up_proj", "model.layers.44.mlp.experts.134.up_proj", "model.layers.44.mlp.experts.135.up_proj", "model.layers.44.mlp.experts.136.up_proj", "model.layers.44.mlp.experts.137.up_proj", "model.layers.44.mlp.experts.138.up_proj", "model.layers.44.mlp.experts.139.up_proj", "model.layers.44.mlp.experts.140.up_proj", "model.layers.44.mlp.experts.141.up_proj", "model.layers.44.mlp.experts.142.up_proj", "model.layers.44.mlp.experts.143.up_proj", "model.layers.44.mlp.experts.144.up_proj", "model.layers.44.mlp.experts.145.up_proj", "model.layers.44.mlp.experts.146.up_proj", "model.layers.44.mlp.experts.147.up_proj", "model.layers.44.mlp.experts.148.up_proj", "model.layers.44.mlp.experts.149.up_proj", "model.layers.44.mlp.experts.150.up_proj", "model.layers.44.mlp.experts.151.up_proj", "model.layers.44.mlp.experts.152.up_proj", "model.layers.44.mlp.experts.153.up_proj", "model.layers.44.mlp.experts.154.up_proj", "model.layers.44.mlp.experts.155.up_proj", "model.layers.44.mlp.experts.156.up_proj", "model.layers.44.mlp.experts.157.up_proj", "model.layers.44.mlp.experts.158.up_proj", "model.layers.44.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0003016212955117281, "dbits": 2516582400 }, { "dkld": -0.00033216411247849464, "dbits": 5033164800 }, { "dkld": -0.00031166598200799145, "dbits": 7549747200 }, { "dkld": -0.00035982988774777014, "dbits": 12582912000 } ] }, { "idx": 266, "layers": [ "model.layers.44.mlp.experts.0.down_proj", "model.layers.44.mlp.experts.1.down_proj", "model.layers.44.mlp.experts.2.down_proj", "model.layers.44.mlp.experts.3.down_proj", "model.layers.44.mlp.experts.4.down_proj", "model.layers.44.mlp.experts.5.down_proj", "model.layers.44.mlp.experts.6.down_proj", "model.layers.44.mlp.experts.7.down_proj", "model.layers.44.mlp.experts.8.down_proj", "model.layers.44.mlp.experts.9.down_proj", "model.layers.44.mlp.experts.10.down_proj", "model.layers.44.mlp.experts.11.down_proj", "model.layers.44.mlp.experts.12.down_proj", "model.layers.44.mlp.experts.13.down_proj", "model.layers.44.mlp.experts.14.down_proj", "model.layers.44.mlp.experts.15.down_proj", "model.layers.44.mlp.experts.16.down_proj", "model.layers.44.mlp.experts.17.down_proj", "model.layers.44.mlp.experts.18.down_proj", "model.layers.44.mlp.experts.19.down_proj", "model.layers.44.mlp.experts.20.down_proj", "model.layers.44.mlp.experts.21.down_proj", "model.layers.44.mlp.experts.22.down_proj", "model.layers.44.mlp.experts.23.down_proj", "model.layers.44.mlp.experts.24.down_proj", "model.layers.44.mlp.experts.25.down_proj", "model.layers.44.mlp.experts.26.down_proj", "model.layers.44.mlp.experts.27.down_proj", "model.layers.44.mlp.experts.28.down_proj", "model.layers.44.mlp.experts.29.down_proj", "model.layers.44.mlp.experts.30.down_proj", "model.layers.44.mlp.experts.31.down_proj", "model.layers.44.mlp.experts.32.down_proj", "model.layers.44.mlp.experts.33.down_proj", "model.layers.44.mlp.experts.34.down_proj", "model.layers.44.mlp.experts.35.down_proj", "model.layers.44.mlp.experts.36.down_proj", "model.layers.44.mlp.experts.37.down_proj", "model.layers.44.mlp.experts.38.down_proj", "model.layers.44.mlp.experts.39.down_proj", "model.layers.44.mlp.experts.40.down_proj", "model.layers.44.mlp.experts.41.down_proj", "model.layers.44.mlp.experts.42.down_proj", "model.layers.44.mlp.experts.43.down_proj", "model.layers.44.mlp.experts.44.down_proj", "model.layers.44.mlp.experts.45.down_proj", "model.layers.44.mlp.experts.46.down_proj", "model.layers.44.mlp.experts.47.down_proj", "model.layers.44.mlp.experts.48.down_proj", "model.layers.44.mlp.experts.49.down_proj", "model.layers.44.mlp.experts.50.down_proj", "model.layers.44.mlp.experts.51.down_proj", "model.layers.44.mlp.experts.52.down_proj", "model.layers.44.mlp.experts.53.down_proj", "model.layers.44.mlp.experts.54.down_proj", "model.layers.44.mlp.experts.55.down_proj", "model.layers.44.mlp.experts.56.down_proj", "model.layers.44.mlp.experts.57.down_proj", "model.layers.44.mlp.experts.58.down_proj", "model.layers.44.mlp.experts.59.down_proj", "model.layers.44.mlp.experts.60.down_proj", "model.layers.44.mlp.experts.61.down_proj", "model.layers.44.mlp.experts.62.down_proj", "model.layers.44.mlp.experts.63.down_proj", "model.layers.44.mlp.experts.64.down_proj", "model.layers.44.mlp.experts.65.down_proj", "model.layers.44.mlp.experts.66.down_proj", "model.layers.44.mlp.experts.67.down_proj", "model.layers.44.mlp.experts.68.down_proj", "model.layers.44.mlp.experts.69.down_proj", "model.layers.44.mlp.experts.70.down_proj", "model.layers.44.mlp.experts.71.down_proj", "model.layers.44.mlp.experts.72.down_proj", "model.layers.44.mlp.experts.73.down_proj", "model.layers.44.mlp.experts.74.down_proj", "model.layers.44.mlp.experts.75.down_proj", "model.layers.44.mlp.experts.76.down_proj", "model.layers.44.mlp.experts.77.down_proj", "model.layers.44.mlp.experts.78.down_proj", "model.layers.44.mlp.experts.79.down_proj", "model.layers.44.mlp.experts.80.down_proj", "model.layers.44.mlp.experts.81.down_proj", "model.layers.44.mlp.experts.82.down_proj", "model.layers.44.mlp.experts.83.down_proj", "model.layers.44.mlp.experts.84.down_proj", "model.layers.44.mlp.experts.85.down_proj", "model.layers.44.mlp.experts.86.down_proj", "model.layers.44.mlp.experts.87.down_proj", "model.layers.44.mlp.experts.88.down_proj", "model.layers.44.mlp.experts.89.down_proj", "model.layers.44.mlp.experts.90.down_proj", "model.layers.44.mlp.experts.91.down_proj", "model.layers.44.mlp.experts.92.down_proj", "model.layers.44.mlp.experts.93.down_proj", "model.layers.44.mlp.experts.94.down_proj", "model.layers.44.mlp.experts.95.down_proj", "model.layers.44.mlp.experts.96.down_proj", "model.layers.44.mlp.experts.97.down_proj", "model.layers.44.mlp.experts.98.down_proj", "model.layers.44.mlp.experts.99.down_proj", "model.layers.44.mlp.experts.100.down_proj", "model.layers.44.mlp.experts.101.down_proj", "model.layers.44.mlp.experts.102.down_proj", "model.layers.44.mlp.experts.103.down_proj", "model.layers.44.mlp.experts.104.down_proj", "model.layers.44.mlp.experts.105.down_proj", "model.layers.44.mlp.experts.106.down_proj", "model.layers.44.mlp.experts.107.down_proj", "model.layers.44.mlp.experts.108.down_proj", "model.layers.44.mlp.experts.109.down_proj", "model.layers.44.mlp.experts.110.down_proj", "model.layers.44.mlp.experts.111.down_proj", "model.layers.44.mlp.experts.112.down_proj", "model.layers.44.mlp.experts.113.down_proj", "model.layers.44.mlp.experts.114.down_proj", "model.layers.44.mlp.experts.115.down_proj", "model.layers.44.mlp.experts.116.down_proj", "model.layers.44.mlp.experts.117.down_proj", "model.layers.44.mlp.experts.118.down_proj", "model.layers.44.mlp.experts.119.down_proj", "model.layers.44.mlp.experts.120.down_proj", "model.layers.44.mlp.experts.121.down_proj", "model.layers.44.mlp.experts.122.down_proj", "model.layers.44.mlp.experts.123.down_proj", "model.layers.44.mlp.experts.124.down_proj", "model.layers.44.mlp.experts.125.down_proj", "model.layers.44.mlp.experts.126.down_proj", "model.layers.44.mlp.experts.127.down_proj", "model.layers.44.mlp.experts.128.down_proj", "model.layers.44.mlp.experts.129.down_proj", "model.layers.44.mlp.experts.130.down_proj", "model.layers.44.mlp.experts.131.down_proj", "model.layers.44.mlp.experts.132.down_proj", "model.layers.44.mlp.experts.133.down_proj", "model.layers.44.mlp.experts.134.down_proj", "model.layers.44.mlp.experts.135.down_proj", "model.layers.44.mlp.experts.136.down_proj", "model.layers.44.mlp.experts.137.down_proj", "model.layers.44.mlp.experts.138.down_proj", "model.layers.44.mlp.experts.139.down_proj", "model.layers.44.mlp.experts.140.down_proj", "model.layers.44.mlp.experts.141.down_proj", "model.layers.44.mlp.experts.142.down_proj", "model.layers.44.mlp.experts.143.down_proj", "model.layers.44.mlp.experts.144.down_proj", "model.layers.44.mlp.experts.145.down_proj", "model.layers.44.mlp.experts.146.down_proj", "model.layers.44.mlp.experts.147.down_proj", "model.layers.44.mlp.experts.148.down_proj", "model.layers.44.mlp.experts.149.down_proj", "model.layers.44.mlp.experts.150.down_proj", "model.layers.44.mlp.experts.151.down_proj", "model.layers.44.mlp.experts.152.down_proj", "model.layers.44.mlp.experts.153.down_proj", "model.layers.44.mlp.experts.154.down_proj", "model.layers.44.mlp.experts.155.down_proj", "model.layers.44.mlp.experts.156.down_proj", "model.layers.44.mlp.experts.157.down_proj", "model.layers.44.mlp.experts.158.down_proj", "model.layers.44.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0003256873227655971, "dbits": 1258291200 }, { "dkld": -0.00029610591009260334, "dbits": 2516582400 }, { "dkld": -0.0003101949580013863, "dbits": 3774873600 }, { "dkld": -0.0003300871700048502, "dbits": 6291456000 } ] }, { "idx": 267, "layers": [ "model.layers.45.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0004702902399003617, "dbits": 62914560 }, { "dkld": -0.0003463180735707311, "dbits": 125829120 }, { "dkld": -0.0003926372155547142, "dbits": 188743680 }, { "dkld": -0.0004026904702186668, "dbits": 314572800 } ] }, { "idx": 268, "layers": [ "model.layers.45.self_attn.k_proj", "model.layers.45.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00117379538714886, "dbits": 10485760 }, { "dkld": -0.0012144139036536217, "dbits": 20971520 }, { "dkld": -0.001193599961698058, "dbits": 31457280 }, { "dkld": -0.0011631535366177642, "dbits": 52428800 } ] }, { "idx": 269, "layers": [ "model.layers.45.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00020789587870240211, "dbits": 62914560 }, { "dkld": -0.00018788194283843318, "dbits": 125829120 }, { "dkld": -0.00026428196579218743, "dbits": 188743680 }, { "dkld": -0.00023554638028144836, "dbits": 314572800 } ] }, { "idx": 270, "layers": [ "model.layers.45.mlp.shared_experts.gate_proj", "model.layers.45.mlp.shared_experts.up_proj", "model.layers.45.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00032052230089903155, "dbits": 23592960 }, { "dkld": -0.00029747728258372463, "dbits": 47185920 }, { "dkld": -0.00033750757575035095, "dbits": 70778880 }, { "dkld": -0.00032113492488861084, "dbits": 117964800 } ] }, { "idx": 271, "layers": [ "model.layers.45.mlp.experts.0.gate_proj", "model.layers.45.mlp.experts.1.gate_proj", "model.layers.45.mlp.experts.2.gate_proj", "model.layers.45.mlp.experts.3.gate_proj", "model.layers.45.mlp.experts.4.gate_proj", "model.layers.45.mlp.experts.5.gate_proj", "model.layers.45.mlp.experts.6.gate_proj", "model.layers.45.mlp.experts.7.gate_proj", "model.layers.45.mlp.experts.8.gate_proj", "model.layers.45.mlp.experts.9.gate_proj", "model.layers.45.mlp.experts.10.gate_proj", "model.layers.45.mlp.experts.11.gate_proj", "model.layers.45.mlp.experts.12.gate_proj", "model.layers.45.mlp.experts.13.gate_proj", "model.layers.45.mlp.experts.14.gate_proj", "model.layers.45.mlp.experts.15.gate_proj", "model.layers.45.mlp.experts.16.gate_proj", "model.layers.45.mlp.experts.17.gate_proj", "model.layers.45.mlp.experts.18.gate_proj", "model.layers.45.mlp.experts.19.gate_proj", "model.layers.45.mlp.experts.20.gate_proj", "model.layers.45.mlp.experts.21.gate_proj", "model.layers.45.mlp.experts.22.gate_proj", "model.layers.45.mlp.experts.23.gate_proj", "model.layers.45.mlp.experts.24.gate_proj", "model.layers.45.mlp.experts.25.gate_proj", "model.layers.45.mlp.experts.26.gate_proj", "model.layers.45.mlp.experts.27.gate_proj", "model.layers.45.mlp.experts.28.gate_proj", "model.layers.45.mlp.experts.29.gate_proj", "model.layers.45.mlp.experts.30.gate_proj", "model.layers.45.mlp.experts.31.gate_proj", "model.layers.45.mlp.experts.32.gate_proj", "model.layers.45.mlp.experts.33.gate_proj", "model.layers.45.mlp.experts.34.gate_proj", "model.layers.45.mlp.experts.35.gate_proj", "model.layers.45.mlp.experts.36.gate_proj", "model.layers.45.mlp.experts.37.gate_proj", "model.layers.45.mlp.experts.38.gate_proj", "model.layers.45.mlp.experts.39.gate_proj", "model.layers.45.mlp.experts.40.gate_proj", "model.layers.45.mlp.experts.41.gate_proj", "model.layers.45.mlp.experts.42.gate_proj", "model.layers.45.mlp.experts.43.gate_proj", "model.layers.45.mlp.experts.44.gate_proj", "model.layers.45.mlp.experts.45.gate_proj", "model.layers.45.mlp.experts.46.gate_proj", "model.layers.45.mlp.experts.47.gate_proj", "model.layers.45.mlp.experts.48.gate_proj", "model.layers.45.mlp.experts.49.gate_proj", "model.layers.45.mlp.experts.50.gate_proj", "model.layers.45.mlp.experts.51.gate_proj", "model.layers.45.mlp.experts.52.gate_proj", "model.layers.45.mlp.experts.53.gate_proj", "model.layers.45.mlp.experts.54.gate_proj", "model.layers.45.mlp.experts.55.gate_proj", "model.layers.45.mlp.experts.56.gate_proj", "model.layers.45.mlp.experts.57.gate_proj", "model.layers.45.mlp.experts.58.gate_proj", "model.layers.45.mlp.experts.59.gate_proj", "model.layers.45.mlp.experts.60.gate_proj", "model.layers.45.mlp.experts.61.gate_proj", "model.layers.45.mlp.experts.62.gate_proj", "model.layers.45.mlp.experts.63.gate_proj", "model.layers.45.mlp.experts.64.gate_proj", "model.layers.45.mlp.experts.65.gate_proj", "model.layers.45.mlp.experts.66.gate_proj", "model.layers.45.mlp.experts.67.gate_proj", "model.layers.45.mlp.experts.68.gate_proj", "model.layers.45.mlp.experts.69.gate_proj", "model.layers.45.mlp.experts.70.gate_proj", "model.layers.45.mlp.experts.71.gate_proj", "model.layers.45.mlp.experts.72.gate_proj", "model.layers.45.mlp.experts.73.gate_proj", "model.layers.45.mlp.experts.74.gate_proj", "model.layers.45.mlp.experts.75.gate_proj", "model.layers.45.mlp.experts.76.gate_proj", "model.layers.45.mlp.experts.77.gate_proj", "model.layers.45.mlp.experts.78.gate_proj", "model.layers.45.mlp.experts.79.gate_proj", "model.layers.45.mlp.experts.80.gate_proj", "model.layers.45.mlp.experts.81.gate_proj", "model.layers.45.mlp.experts.82.gate_proj", "model.layers.45.mlp.experts.83.gate_proj", "model.layers.45.mlp.experts.84.gate_proj", "model.layers.45.mlp.experts.85.gate_proj", "model.layers.45.mlp.experts.86.gate_proj", "model.layers.45.mlp.experts.87.gate_proj", "model.layers.45.mlp.experts.88.gate_proj", "model.layers.45.mlp.experts.89.gate_proj", "model.layers.45.mlp.experts.90.gate_proj", "model.layers.45.mlp.experts.91.gate_proj", "model.layers.45.mlp.experts.92.gate_proj", "model.layers.45.mlp.experts.93.gate_proj", "model.layers.45.mlp.experts.94.gate_proj", "model.layers.45.mlp.experts.95.gate_proj", "model.layers.45.mlp.experts.96.gate_proj", "model.layers.45.mlp.experts.97.gate_proj", "model.layers.45.mlp.experts.98.gate_proj", "model.layers.45.mlp.experts.99.gate_proj", "model.layers.45.mlp.experts.100.gate_proj", "model.layers.45.mlp.experts.101.gate_proj", "model.layers.45.mlp.experts.102.gate_proj", "model.layers.45.mlp.experts.103.gate_proj", "model.layers.45.mlp.experts.104.gate_proj", "model.layers.45.mlp.experts.105.gate_proj", "model.layers.45.mlp.experts.106.gate_proj", "model.layers.45.mlp.experts.107.gate_proj", "model.layers.45.mlp.experts.108.gate_proj", "model.layers.45.mlp.experts.109.gate_proj", "model.layers.45.mlp.experts.110.gate_proj", "model.layers.45.mlp.experts.111.gate_proj", "model.layers.45.mlp.experts.112.gate_proj", "model.layers.45.mlp.experts.113.gate_proj", "model.layers.45.mlp.experts.114.gate_proj", "model.layers.45.mlp.experts.115.gate_proj", "model.layers.45.mlp.experts.116.gate_proj", "model.layers.45.mlp.experts.117.gate_proj", "model.layers.45.mlp.experts.118.gate_proj", "model.layers.45.mlp.experts.119.gate_proj", "model.layers.45.mlp.experts.120.gate_proj", "model.layers.45.mlp.experts.121.gate_proj", "model.layers.45.mlp.experts.122.gate_proj", "model.layers.45.mlp.experts.123.gate_proj", "model.layers.45.mlp.experts.124.gate_proj", "model.layers.45.mlp.experts.125.gate_proj", "model.layers.45.mlp.experts.126.gate_proj", "model.layers.45.mlp.experts.127.gate_proj", "model.layers.45.mlp.experts.128.gate_proj", "model.layers.45.mlp.experts.129.gate_proj", "model.layers.45.mlp.experts.130.gate_proj", "model.layers.45.mlp.experts.131.gate_proj", "model.layers.45.mlp.experts.132.gate_proj", "model.layers.45.mlp.experts.133.gate_proj", "model.layers.45.mlp.experts.134.gate_proj", "model.layers.45.mlp.experts.135.gate_proj", "model.layers.45.mlp.experts.136.gate_proj", "model.layers.45.mlp.experts.137.gate_proj", "model.layers.45.mlp.experts.138.gate_proj", "model.layers.45.mlp.experts.139.gate_proj", "model.layers.45.mlp.experts.140.gate_proj", "model.layers.45.mlp.experts.141.gate_proj", "model.layers.45.mlp.experts.142.gate_proj", "model.layers.45.mlp.experts.143.gate_proj", "model.layers.45.mlp.experts.144.gate_proj", "model.layers.45.mlp.experts.145.gate_proj", "model.layers.45.mlp.experts.146.gate_proj", "model.layers.45.mlp.experts.147.gate_proj", "model.layers.45.mlp.experts.148.gate_proj", "model.layers.45.mlp.experts.149.gate_proj", "model.layers.45.mlp.experts.150.gate_proj", "model.layers.45.mlp.experts.151.gate_proj", "model.layers.45.mlp.experts.152.gate_proj", "model.layers.45.mlp.experts.153.gate_proj", "model.layers.45.mlp.experts.154.gate_proj", "model.layers.45.mlp.experts.155.gate_proj", "model.layers.45.mlp.experts.156.gate_proj", "model.layers.45.mlp.experts.157.gate_proj", "model.layers.45.mlp.experts.158.gate_proj", "model.layers.45.mlp.experts.159.gate_proj", "model.layers.45.mlp.experts.0.up_proj", "model.layers.45.mlp.experts.1.up_proj", "model.layers.45.mlp.experts.2.up_proj", "model.layers.45.mlp.experts.3.up_proj", "model.layers.45.mlp.experts.4.up_proj", "model.layers.45.mlp.experts.5.up_proj", "model.layers.45.mlp.experts.6.up_proj", "model.layers.45.mlp.experts.7.up_proj", "model.layers.45.mlp.experts.8.up_proj", "model.layers.45.mlp.experts.9.up_proj", "model.layers.45.mlp.experts.10.up_proj", "model.layers.45.mlp.experts.11.up_proj", "model.layers.45.mlp.experts.12.up_proj", "model.layers.45.mlp.experts.13.up_proj", "model.layers.45.mlp.experts.14.up_proj", "model.layers.45.mlp.experts.15.up_proj", "model.layers.45.mlp.experts.16.up_proj", "model.layers.45.mlp.experts.17.up_proj", "model.layers.45.mlp.experts.18.up_proj", "model.layers.45.mlp.experts.19.up_proj", "model.layers.45.mlp.experts.20.up_proj", "model.layers.45.mlp.experts.21.up_proj", "model.layers.45.mlp.experts.22.up_proj", "model.layers.45.mlp.experts.23.up_proj", "model.layers.45.mlp.experts.24.up_proj", "model.layers.45.mlp.experts.25.up_proj", "model.layers.45.mlp.experts.26.up_proj", "model.layers.45.mlp.experts.27.up_proj", "model.layers.45.mlp.experts.28.up_proj", "model.layers.45.mlp.experts.29.up_proj", "model.layers.45.mlp.experts.30.up_proj", "model.layers.45.mlp.experts.31.up_proj", "model.layers.45.mlp.experts.32.up_proj", "model.layers.45.mlp.experts.33.up_proj", "model.layers.45.mlp.experts.34.up_proj", "model.layers.45.mlp.experts.35.up_proj", "model.layers.45.mlp.experts.36.up_proj", "model.layers.45.mlp.experts.37.up_proj", "model.layers.45.mlp.experts.38.up_proj", "model.layers.45.mlp.experts.39.up_proj", "model.layers.45.mlp.experts.40.up_proj", "model.layers.45.mlp.experts.41.up_proj", "model.layers.45.mlp.experts.42.up_proj", "model.layers.45.mlp.experts.43.up_proj", "model.layers.45.mlp.experts.44.up_proj", "model.layers.45.mlp.experts.45.up_proj", "model.layers.45.mlp.experts.46.up_proj", "model.layers.45.mlp.experts.47.up_proj", "model.layers.45.mlp.experts.48.up_proj", "model.layers.45.mlp.experts.49.up_proj", "model.layers.45.mlp.experts.50.up_proj", "model.layers.45.mlp.experts.51.up_proj", "model.layers.45.mlp.experts.52.up_proj", "model.layers.45.mlp.experts.53.up_proj", "model.layers.45.mlp.experts.54.up_proj", "model.layers.45.mlp.experts.55.up_proj", "model.layers.45.mlp.experts.56.up_proj", "model.layers.45.mlp.experts.57.up_proj", "model.layers.45.mlp.experts.58.up_proj", "model.layers.45.mlp.experts.59.up_proj", "model.layers.45.mlp.experts.60.up_proj", "model.layers.45.mlp.experts.61.up_proj", "model.layers.45.mlp.experts.62.up_proj", "model.layers.45.mlp.experts.63.up_proj", "model.layers.45.mlp.experts.64.up_proj", "model.layers.45.mlp.experts.65.up_proj", "model.layers.45.mlp.experts.66.up_proj", "model.layers.45.mlp.experts.67.up_proj", "model.layers.45.mlp.experts.68.up_proj", "model.layers.45.mlp.experts.69.up_proj", "model.layers.45.mlp.experts.70.up_proj", "model.layers.45.mlp.experts.71.up_proj", "model.layers.45.mlp.experts.72.up_proj", "model.layers.45.mlp.experts.73.up_proj", "model.layers.45.mlp.experts.74.up_proj", "model.layers.45.mlp.experts.75.up_proj", "model.layers.45.mlp.experts.76.up_proj", "model.layers.45.mlp.experts.77.up_proj", "model.layers.45.mlp.experts.78.up_proj", "model.layers.45.mlp.experts.79.up_proj", "model.layers.45.mlp.experts.80.up_proj", "model.layers.45.mlp.experts.81.up_proj", "model.layers.45.mlp.experts.82.up_proj", "model.layers.45.mlp.experts.83.up_proj", "model.layers.45.mlp.experts.84.up_proj", "model.layers.45.mlp.experts.85.up_proj", "model.layers.45.mlp.experts.86.up_proj", "model.layers.45.mlp.experts.87.up_proj", "model.layers.45.mlp.experts.88.up_proj", "model.layers.45.mlp.experts.89.up_proj", "model.layers.45.mlp.experts.90.up_proj", "model.layers.45.mlp.experts.91.up_proj", "model.layers.45.mlp.experts.92.up_proj", "model.layers.45.mlp.experts.93.up_proj", "model.layers.45.mlp.experts.94.up_proj", "model.layers.45.mlp.experts.95.up_proj", "model.layers.45.mlp.experts.96.up_proj", "model.layers.45.mlp.experts.97.up_proj", "model.layers.45.mlp.experts.98.up_proj", "model.layers.45.mlp.experts.99.up_proj", "model.layers.45.mlp.experts.100.up_proj", "model.layers.45.mlp.experts.101.up_proj", "model.layers.45.mlp.experts.102.up_proj", "model.layers.45.mlp.experts.103.up_proj", "model.layers.45.mlp.experts.104.up_proj", "model.layers.45.mlp.experts.105.up_proj", "model.layers.45.mlp.experts.106.up_proj", "model.layers.45.mlp.experts.107.up_proj", "model.layers.45.mlp.experts.108.up_proj", "model.layers.45.mlp.experts.109.up_proj", "model.layers.45.mlp.experts.110.up_proj", "model.layers.45.mlp.experts.111.up_proj", "model.layers.45.mlp.experts.112.up_proj", "model.layers.45.mlp.experts.113.up_proj", "model.layers.45.mlp.experts.114.up_proj", "model.layers.45.mlp.experts.115.up_proj", "model.layers.45.mlp.experts.116.up_proj", "model.layers.45.mlp.experts.117.up_proj", "model.layers.45.mlp.experts.118.up_proj", "model.layers.45.mlp.experts.119.up_proj", "model.layers.45.mlp.experts.120.up_proj", "model.layers.45.mlp.experts.121.up_proj", "model.layers.45.mlp.experts.122.up_proj", "model.layers.45.mlp.experts.123.up_proj", "model.layers.45.mlp.experts.124.up_proj", "model.layers.45.mlp.experts.125.up_proj", "model.layers.45.mlp.experts.126.up_proj", "model.layers.45.mlp.experts.127.up_proj", "model.layers.45.mlp.experts.128.up_proj", "model.layers.45.mlp.experts.129.up_proj", "model.layers.45.mlp.experts.130.up_proj", "model.layers.45.mlp.experts.131.up_proj", "model.layers.45.mlp.experts.132.up_proj", "model.layers.45.mlp.experts.133.up_proj", "model.layers.45.mlp.experts.134.up_proj", "model.layers.45.mlp.experts.135.up_proj", "model.layers.45.mlp.experts.136.up_proj", "model.layers.45.mlp.experts.137.up_proj", "model.layers.45.mlp.experts.138.up_proj", "model.layers.45.mlp.experts.139.up_proj", "model.layers.45.mlp.experts.140.up_proj", "model.layers.45.mlp.experts.141.up_proj", "model.layers.45.mlp.experts.142.up_proj", "model.layers.45.mlp.experts.143.up_proj", "model.layers.45.mlp.experts.144.up_proj", "model.layers.45.mlp.experts.145.up_proj", "model.layers.45.mlp.experts.146.up_proj", "model.layers.45.mlp.experts.147.up_proj", "model.layers.45.mlp.experts.148.up_proj", "model.layers.45.mlp.experts.149.up_proj", "model.layers.45.mlp.experts.150.up_proj", "model.layers.45.mlp.experts.151.up_proj", "model.layers.45.mlp.experts.152.up_proj", "model.layers.45.mlp.experts.153.up_proj", "model.layers.45.mlp.experts.154.up_proj", "model.layers.45.mlp.experts.155.up_proj", "model.layers.45.mlp.experts.156.up_proj", "model.layers.45.mlp.experts.157.up_proj", "model.layers.45.mlp.experts.158.up_proj", "model.layers.45.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0004920841194689385, "dbits": 2516582400 }, { "dkld": -0.00046388600021601833, "dbits": 5033164800 }, { "dkld": -0.00041874237358570376, "dbits": 7549747200 }, { "dkld": -0.00048401858657598495, "dbits": 12582912000 } ] }, { "idx": 272, "layers": [ "model.layers.45.mlp.experts.0.down_proj", "model.layers.45.mlp.experts.1.down_proj", "model.layers.45.mlp.experts.2.down_proj", "model.layers.45.mlp.experts.3.down_proj", "model.layers.45.mlp.experts.4.down_proj", "model.layers.45.mlp.experts.5.down_proj", "model.layers.45.mlp.experts.6.down_proj", "model.layers.45.mlp.experts.7.down_proj", "model.layers.45.mlp.experts.8.down_proj", "model.layers.45.mlp.experts.9.down_proj", "model.layers.45.mlp.experts.10.down_proj", "model.layers.45.mlp.experts.11.down_proj", "model.layers.45.mlp.experts.12.down_proj", "model.layers.45.mlp.experts.13.down_proj", "model.layers.45.mlp.experts.14.down_proj", "model.layers.45.mlp.experts.15.down_proj", "model.layers.45.mlp.experts.16.down_proj", "model.layers.45.mlp.experts.17.down_proj", "model.layers.45.mlp.experts.18.down_proj", "model.layers.45.mlp.experts.19.down_proj", "model.layers.45.mlp.experts.20.down_proj", "model.layers.45.mlp.experts.21.down_proj", "model.layers.45.mlp.experts.22.down_proj", "model.layers.45.mlp.experts.23.down_proj", "model.layers.45.mlp.experts.24.down_proj", "model.layers.45.mlp.experts.25.down_proj", "model.layers.45.mlp.experts.26.down_proj", "model.layers.45.mlp.experts.27.down_proj", "model.layers.45.mlp.experts.28.down_proj", "model.layers.45.mlp.experts.29.down_proj", "model.layers.45.mlp.experts.30.down_proj", "model.layers.45.mlp.experts.31.down_proj", "model.layers.45.mlp.experts.32.down_proj", "model.layers.45.mlp.experts.33.down_proj", "model.layers.45.mlp.experts.34.down_proj", "model.layers.45.mlp.experts.35.down_proj", "model.layers.45.mlp.experts.36.down_proj", "model.layers.45.mlp.experts.37.down_proj", "model.layers.45.mlp.experts.38.down_proj", "model.layers.45.mlp.experts.39.down_proj", "model.layers.45.mlp.experts.40.down_proj", "model.layers.45.mlp.experts.41.down_proj", "model.layers.45.mlp.experts.42.down_proj", "model.layers.45.mlp.experts.43.down_proj", "model.layers.45.mlp.experts.44.down_proj", "model.layers.45.mlp.experts.45.down_proj", "model.layers.45.mlp.experts.46.down_proj", "model.layers.45.mlp.experts.47.down_proj", "model.layers.45.mlp.experts.48.down_proj", "model.layers.45.mlp.experts.49.down_proj", "model.layers.45.mlp.experts.50.down_proj", "model.layers.45.mlp.experts.51.down_proj", "model.layers.45.mlp.experts.52.down_proj", "model.layers.45.mlp.experts.53.down_proj", "model.layers.45.mlp.experts.54.down_proj", "model.layers.45.mlp.experts.55.down_proj", "model.layers.45.mlp.experts.56.down_proj", "model.layers.45.mlp.experts.57.down_proj", "model.layers.45.mlp.experts.58.down_proj", "model.layers.45.mlp.experts.59.down_proj", "model.layers.45.mlp.experts.60.down_proj", "model.layers.45.mlp.experts.61.down_proj", "model.layers.45.mlp.experts.62.down_proj", "model.layers.45.mlp.experts.63.down_proj", "model.layers.45.mlp.experts.64.down_proj", "model.layers.45.mlp.experts.65.down_proj", "model.layers.45.mlp.experts.66.down_proj", "model.layers.45.mlp.experts.67.down_proj", "model.layers.45.mlp.experts.68.down_proj", "model.layers.45.mlp.experts.69.down_proj", "model.layers.45.mlp.experts.70.down_proj", "model.layers.45.mlp.experts.71.down_proj", "model.layers.45.mlp.experts.72.down_proj", "model.layers.45.mlp.experts.73.down_proj", "model.layers.45.mlp.experts.74.down_proj", "model.layers.45.mlp.experts.75.down_proj", "model.layers.45.mlp.experts.76.down_proj", "model.layers.45.mlp.experts.77.down_proj", "model.layers.45.mlp.experts.78.down_proj", "model.layers.45.mlp.experts.79.down_proj", "model.layers.45.mlp.experts.80.down_proj", "model.layers.45.mlp.experts.81.down_proj", "model.layers.45.mlp.experts.82.down_proj", "model.layers.45.mlp.experts.83.down_proj", "model.layers.45.mlp.experts.84.down_proj", "model.layers.45.mlp.experts.85.down_proj", "model.layers.45.mlp.experts.86.down_proj", "model.layers.45.mlp.experts.87.down_proj", "model.layers.45.mlp.experts.88.down_proj", "model.layers.45.mlp.experts.89.down_proj", "model.layers.45.mlp.experts.90.down_proj", "model.layers.45.mlp.experts.91.down_proj", "model.layers.45.mlp.experts.92.down_proj", "model.layers.45.mlp.experts.93.down_proj", "model.layers.45.mlp.experts.94.down_proj", "model.layers.45.mlp.experts.95.down_proj", "model.layers.45.mlp.experts.96.down_proj", "model.layers.45.mlp.experts.97.down_proj", "model.layers.45.mlp.experts.98.down_proj", "model.layers.45.mlp.experts.99.down_proj", "model.layers.45.mlp.experts.100.down_proj", "model.layers.45.mlp.experts.101.down_proj", "model.layers.45.mlp.experts.102.down_proj", "model.layers.45.mlp.experts.103.down_proj", "model.layers.45.mlp.experts.104.down_proj", "model.layers.45.mlp.experts.105.down_proj", "model.layers.45.mlp.experts.106.down_proj", "model.layers.45.mlp.experts.107.down_proj", "model.layers.45.mlp.experts.108.down_proj", "model.layers.45.mlp.experts.109.down_proj", "model.layers.45.mlp.experts.110.down_proj", "model.layers.45.mlp.experts.111.down_proj", "model.layers.45.mlp.experts.112.down_proj", "model.layers.45.mlp.experts.113.down_proj", "model.layers.45.mlp.experts.114.down_proj", "model.layers.45.mlp.experts.115.down_proj", "model.layers.45.mlp.experts.116.down_proj", "model.layers.45.mlp.experts.117.down_proj", "model.layers.45.mlp.experts.118.down_proj", "model.layers.45.mlp.experts.119.down_proj", "model.layers.45.mlp.experts.120.down_proj", "model.layers.45.mlp.experts.121.down_proj", "model.layers.45.mlp.experts.122.down_proj", "model.layers.45.mlp.experts.123.down_proj", "model.layers.45.mlp.experts.124.down_proj", "model.layers.45.mlp.experts.125.down_proj", "model.layers.45.mlp.experts.126.down_proj", "model.layers.45.mlp.experts.127.down_proj", "model.layers.45.mlp.experts.128.down_proj", "model.layers.45.mlp.experts.129.down_proj", "model.layers.45.mlp.experts.130.down_proj", "model.layers.45.mlp.experts.131.down_proj", "model.layers.45.mlp.experts.132.down_proj", "model.layers.45.mlp.experts.133.down_proj", "model.layers.45.mlp.experts.134.down_proj", "model.layers.45.mlp.experts.135.down_proj", "model.layers.45.mlp.experts.136.down_proj", "model.layers.45.mlp.experts.137.down_proj", "model.layers.45.mlp.experts.138.down_proj", "model.layers.45.mlp.experts.139.down_proj", "model.layers.45.mlp.experts.140.down_proj", "model.layers.45.mlp.experts.141.down_proj", "model.layers.45.mlp.experts.142.down_proj", "model.layers.45.mlp.experts.143.down_proj", "model.layers.45.mlp.experts.144.down_proj", "model.layers.45.mlp.experts.145.down_proj", "model.layers.45.mlp.experts.146.down_proj", "model.layers.45.mlp.experts.147.down_proj", "model.layers.45.mlp.experts.148.down_proj", "model.layers.45.mlp.experts.149.down_proj", "model.layers.45.mlp.experts.150.down_proj", "model.layers.45.mlp.experts.151.down_proj", "model.layers.45.mlp.experts.152.down_proj", "model.layers.45.mlp.experts.153.down_proj", "model.layers.45.mlp.experts.154.down_proj", "model.layers.45.mlp.experts.155.down_proj", "model.layers.45.mlp.experts.156.down_proj", "model.layers.45.mlp.experts.157.down_proj", "model.layers.45.mlp.experts.158.down_proj", "model.layers.45.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00016219038516283868, "dbits": 1258291200 }, { "dkld": -0.0002290559932589531, "dbits": 2516582400 }, { "dkld": -0.00025685848668217937, "dbits": 3774873600 }, { "dkld": -0.00022404044866562722, "dbits": 6291456000 } ] }, { "idx": 273, "layers": [ "model.layers.46.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00018346244469286122, "dbits": 62914560 }, { "dkld": -0.000603708252310764, "dbits": 125829120 }, { "dkld": -0.0004961388185620308, "dbits": 188743680 }, { "dkld": -0.0004992809146642796, "dbits": 314572800 } ] }, { "idx": 274, "layers": [ "model.layers.46.self_attn.k_proj", "model.layers.46.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00029814038425685085, "dbits": 10485760 }, { "dkld": -0.00036063548177481253, "dbits": 20971520 }, { "dkld": -0.0003830243833363167, "dbits": 31457280 }, { "dkld": -0.0003739733248949162, "dbits": 52428800 } ] }, { "idx": 275, "layers": [ "model.layers.46.self_attn.o_proj" ], "candidates": [ { "dkld": -1.089405268431265e-05, "dbits": 62914560 }, { "dkld": -0.00010841563344002325, "dbits": 125829120 }, { "dkld": -4.3482892215257474e-05, "dbits": 188743680 }, { "dkld": -2.815695479513325e-05, "dbits": 314572800 } ] }, { "idx": 276, "layers": [ "model.layers.46.mlp.shared_experts.gate_proj", "model.layers.46.mlp.shared_experts.up_proj", "model.layers.46.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 1.1834781616926193e-05, "dbits": 23592960 }, { "dkld": -0.00011832891032100001, "dbits": 47185920 }, { "dkld": -0.0002325948327779881, "dbits": 70778880 }, { "dkld": -0.00022239359095693745, "dbits": 117964800 } ] }, { "idx": 277, "layers": [ "model.layers.46.mlp.experts.0.gate_proj", "model.layers.46.mlp.experts.1.gate_proj", "model.layers.46.mlp.experts.2.gate_proj", "model.layers.46.mlp.experts.3.gate_proj", "model.layers.46.mlp.experts.4.gate_proj", "model.layers.46.mlp.experts.5.gate_proj", "model.layers.46.mlp.experts.6.gate_proj", "model.layers.46.mlp.experts.7.gate_proj", "model.layers.46.mlp.experts.8.gate_proj", "model.layers.46.mlp.experts.9.gate_proj", "model.layers.46.mlp.experts.10.gate_proj", "model.layers.46.mlp.experts.11.gate_proj", "model.layers.46.mlp.experts.12.gate_proj", "model.layers.46.mlp.experts.13.gate_proj", "model.layers.46.mlp.experts.14.gate_proj", "model.layers.46.mlp.experts.15.gate_proj", "model.layers.46.mlp.experts.16.gate_proj", "model.layers.46.mlp.experts.17.gate_proj", "model.layers.46.mlp.experts.18.gate_proj", "model.layers.46.mlp.experts.19.gate_proj", "model.layers.46.mlp.experts.20.gate_proj", "model.layers.46.mlp.experts.21.gate_proj", "model.layers.46.mlp.experts.22.gate_proj", "model.layers.46.mlp.experts.23.gate_proj", "model.layers.46.mlp.experts.24.gate_proj", "model.layers.46.mlp.experts.25.gate_proj", "model.layers.46.mlp.experts.26.gate_proj", "model.layers.46.mlp.experts.27.gate_proj", "model.layers.46.mlp.experts.28.gate_proj", "model.layers.46.mlp.experts.29.gate_proj", "model.layers.46.mlp.experts.30.gate_proj", "model.layers.46.mlp.experts.31.gate_proj", "model.layers.46.mlp.experts.32.gate_proj", "model.layers.46.mlp.experts.33.gate_proj", "model.layers.46.mlp.experts.34.gate_proj", "model.layers.46.mlp.experts.35.gate_proj", "model.layers.46.mlp.experts.36.gate_proj", "model.layers.46.mlp.experts.37.gate_proj", "model.layers.46.mlp.experts.38.gate_proj", "model.layers.46.mlp.experts.39.gate_proj", "model.layers.46.mlp.experts.40.gate_proj", "model.layers.46.mlp.experts.41.gate_proj", "model.layers.46.mlp.experts.42.gate_proj", "model.layers.46.mlp.experts.43.gate_proj", "model.layers.46.mlp.experts.44.gate_proj", "model.layers.46.mlp.experts.45.gate_proj", "model.layers.46.mlp.experts.46.gate_proj", "model.layers.46.mlp.experts.47.gate_proj", "model.layers.46.mlp.experts.48.gate_proj", "model.layers.46.mlp.experts.49.gate_proj", "model.layers.46.mlp.experts.50.gate_proj", "model.layers.46.mlp.experts.51.gate_proj", "model.layers.46.mlp.experts.52.gate_proj", "model.layers.46.mlp.experts.53.gate_proj", "model.layers.46.mlp.experts.54.gate_proj", "model.layers.46.mlp.experts.55.gate_proj", "model.layers.46.mlp.experts.56.gate_proj", "model.layers.46.mlp.experts.57.gate_proj", "model.layers.46.mlp.experts.58.gate_proj", "model.layers.46.mlp.experts.59.gate_proj", "model.layers.46.mlp.experts.60.gate_proj", "model.layers.46.mlp.experts.61.gate_proj", "model.layers.46.mlp.experts.62.gate_proj", "model.layers.46.mlp.experts.63.gate_proj", "model.layers.46.mlp.experts.64.gate_proj", "model.layers.46.mlp.experts.65.gate_proj", "model.layers.46.mlp.experts.66.gate_proj", "model.layers.46.mlp.experts.67.gate_proj", "model.layers.46.mlp.experts.68.gate_proj", "model.layers.46.mlp.experts.69.gate_proj", "model.layers.46.mlp.experts.70.gate_proj", "model.layers.46.mlp.experts.71.gate_proj", "model.layers.46.mlp.experts.72.gate_proj", "model.layers.46.mlp.experts.73.gate_proj", "model.layers.46.mlp.experts.74.gate_proj", "model.layers.46.mlp.experts.75.gate_proj", "model.layers.46.mlp.experts.76.gate_proj", "model.layers.46.mlp.experts.77.gate_proj", "model.layers.46.mlp.experts.78.gate_proj", "model.layers.46.mlp.experts.79.gate_proj", "model.layers.46.mlp.experts.80.gate_proj", "model.layers.46.mlp.experts.81.gate_proj", "model.layers.46.mlp.experts.82.gate_proj", "model.layers.46.mlp.experts.83.gate_proj", "model.layers.46.mlp.experts.84.gate_proj", "model.layers.46.mlp.experts.85.gate_proj", "model.layers.46.mlp.experts.86.gate_proj", "model.layers.46.mlp.experts.87.gate_proj", "model.layers.46.mlp.experts.88.gate_proj", "model.layers.46.mlp.experts.89.gate_proj", "model.layers.46.mlp.experts.90.gate_proj", "model.layers.46.mlp.experts.91.gate_proj", "model.layers.46.mlp.experts.92.gate_proj", "model.layers.46.mlp.experts.93.gate_proj", "model.layers.46.mlp.experts.94.gate_proj", "model.layers.46.mlp.experts.95.gate_proj", "model.layers.46.mlp.experts.96.gate_proj", "model.layers.46.mlp.experts.97.gate_proj", "model.layers.46.mlp.experts.98.gate_proj", "model.layers.46.mlp.experts.99.gate_proj", "model.layers.46.mlp.experts.100.gate_proj", "model.layers.46.mlp.experts.101.gate_proj", "model.layers.46.mlp.experts.102.gate_proj", "model.layers.46.mlp.experts.103.gate_proj", "model.layers.46.mlp.experts.104.gate_proj", "model.layers.46.mlp.experts.105.gate_proj", "model.layers.46.mlp.experts.106.gate_proj", "model.layers.46.mlp.experts.107.gate_proj", "model.layers.46.mlp.experts.108.gate_proj", "model.layers.46.mlp.experts.109.gate_proj", "model.layers.46.mlp.experts.110.gate_proj", "model.layers.46.mlp.experts.111.gate_proj", "model.layers.46.mlp.experts.112.gate_proj", "model.layers.46.mlp.experts.113.gate_proj", "model.layers.46.mlp.experts.114.gate_proj", "model.layers.46.mlp.experts.115.gate_proj", "model.layers.46.mlp.experts.116.gate_proj", "model.layers.46.mlp.experts.117.gate_proj", "model.layers.46.mlp.experts.118.gate_proj", "model.layers.46.mlp.experts.119.gate_proj", "model.layers.46.mlp.experts.120.gate_proj", "model.layers.46.mlp.experts.121.gate_proj", "model.layers.46.mlp.experts.122.gate_proj", "model.layers.46.mlp.experts.123.gate_proj", "model.layers.46.mlp.experts.124.gate_proj", "model.layers.46.mlp.experts.125.gate_proj", "model.layers.46.mlp.experts.126.gate_proj", "model.layers.46.mlp.experts.127.gate_proj", "model.layers.46.mlp.experts.128.gate_proj", "model.layers.46.mlp.experts.129.gate_proj", "model.layers.46.mlp.experts.130.gate_proj", "model.layers.46.mlp.experts.131.gate_proj", "model.layers.46.mlp.experts.132.gate_proj", "model.layers.46.mlp.experts.133.gate_proj", "model.layers.46.mlp.experts.134.gate_proj", "model.layers.46.mlp.experts.135.gate_proj", "model.layers.46.mlp.experts.136.gate_proj", "model.layers.46.mlp.experts.137.gate_proj", "model.layers.46.mlp.experts.138.gate_proj", "model.layers.46.mlp.experts.139.gate_proj", "model.layers.46.mlp.experts.140.gate_proj", "model.layers.46.mlp.experts.141.gate_proj", "model.layers.46.mlp.experts.142.gate_proj", "model.layers.46.mlp.experts.143.gate_proj", "model.layers.46.mlp.experts.144.gate_proj", "model.layers.46.mlp.experts.145.gate_proj", "model.layers.46.mlp.experts.146.gate_proj", "model.layers.46.mlp.experts.147.gate_proj", "model.layers.46.mlp.experts.148.gate_proj", "model.layers.46.mlp.experts.149.gate_proj", "model.layers.46.mlp.experts.150.gate_proj", "model.layers.46.mlp.experts.151.gate_proj", "model.layers.46.mlp.experts.152.gate_proj", "model.layers.46.mlp.experts.153.gate_proj", "model.layers.46.mlp.experts.154.gate_proj", "model.layers.46.mlp.experts.155.gate_proj", "model.layers.46.mlp.experts.156.gate_proj", "model.layers.46.mlp.experts.157.gate_proj", "model.layers.46.mlp.experts.158.gate_proj", "model.layers.46.mlp.experts.159.gate_proj", "model.layers.46.mlp.experts.0.up_proj", "model.layers.46.mlp.experts.1.up_proj", "model.layers.46.mlp.experts.2.up_proj", "model.layers.46.mlp.experts.3.up_proj", "model.layers.46.mlp.experts.4.up_proj", "model.layers.46.mlp.experts.5.up_proj", "model.layers.46.mlp.experts.6.up_proj", "model.layers.46.mlp.experts.7.up_proj", "model.layers.46.mlp.experts.8.up_proj", "model.layers.46.mlp.experts.9.up_proj", "model.layers.46.mlp.experts.10.up_proj", "model.layers.46.mlp.experts.11.up_proj", "model.layers.46.mlp.experts.12.up_proj", "model.layers.46.mlp.experts.13.up_proj", "model.layers.46.mlp.experts.14.up_proj", "model.layers.46.mlp.experts.15.up_proj", "model.layers.46.mlp.experts.16.up_proj", "model.layers.46.mlp.experts.17.up_proj", "model.layers.46.mlp.experts.18.up_proj", "model.layers.46.mlp.experts.19.up_proj", "model.layers.46.mlp.experts.20.up_proj", "model.layers.46.mlp.experts.21.up_proj", "model.layers.46.mlp.experts.22.up_proj", "model.layers.46.mlp.experts.23.up_proj", "model.layers.46.mlp.experts.24.up_proj", "model.layers.46.mlp.experts.25.up_proj", "model.layers.46.mlp.experts.26.up_proj", "model.layers.46.mlp.experts.27.up_proj", "model.layers.46.mlp.experts.28.up_proj", "model.layers.46.mlp.experts.29.up_proj", "model.layers.46.mlp.experts.30.up_proj", "model.layers.46.mlp.experts.31.up_proj", "model.layers.46.mlp.experts.32.up_proj", "model.layers.46.mlp.experts.33.up_proj", "model.layers.46.mlp.experts.34.up_proj", "model.layers.46.mlp.experts.35.up_proj", "model.layers.46.mlp.experts.36.up_proj", "model.layers.46.mlp.experts.37.up_proj", "model.layers.46.mlp.experts.38.up_proj", "model.layers.46.mlp.experts.39.up_proj", "model.layers.46.mlp.experts.40.up_proj", "model.layers.46.mlp.experts.41.up_proj", "model.layers.46.mlp.experts.42.up_proj", "model.layers.46.mlp.experts.43.up_proj", "model.layers.46.mlp.experts.44.up_proj", "model.layers.46.mlp.experts.45.up_proj", "model.layers.46.mlp.experts.46.up_proj", "model.layers.46.mlp.experts.47.up_proj", "model.layers.46.mlp.experts.48.up_proj", "model.layers.46.mlp.experts.49.up_proj", "model.layers.46.mlp.experts.50.up_proj", "model.layers.46.mlp.experts.51.up_proj", "model.layers.46.mlp.experts.52.up_proj", "model.layers.46.mlp.experts.53.up_proj", "model.layers.46.mlp.experts.54.up_proj", "model.layers.46.mlp.experts.55.up_proj", "model.layers.46.mlp.experts.56.up_proj", "model.layers.46.mlp.experts.57.up_proj", "model.layers.46.mlp.experts.58.up_proj", "model.layers.46.mlp.experts.59.up_proj", "model.layers.46.mlp.experts.60.up_proj", "model.layers.46.mlp.experts.61.up_proj", "model.layers.46.mlp.experts.62.up_proj", "model.layers.46.mlp.experts.63.up_proj", "model.layers.46.mlp.experts.64.up_proj", "model.layers.46.mlp.experts.65.up_proj", "model.layers.46.mlp.experts.66.up_proj", "model.layers.46.mlp.experts.67.up_proj", "model.layers.46.mlp.experts.68.up_proj", "model.layers.46.mlp.experts.69.up_proj", "model.layers.46.mlp.experts.70.up_proj", "model.layers.46.mlp.experts.71.up_proj", "model.layers.46.mlp.experts.72.up_proj", "model.layers.46.mlp.experts.73.up_proj", "model.layers.46.mlp.experts.74.up_proj", "model.layers.46.mlp.experts.75.up_proj", "model.layers.46.mlp.experts.76.up_proj", "model.layers.46.mlp.experts.77.up_proj", "model.layers.46.mlp.experts.78.up_proj", "model.layers.46.mlp.experts.79.up_proj", "model.layers.46.mlp.experts.80.up_proj", "model.layers.46.mlp.experts.81.up_proj", "model.layers.46.mlp.experts.82.up_proj", "model.layers.46.mlp.experts.83.up_proj", "model.layers.46.mlp.experts.84.up_proj", "model.layers.46.mlp.experts.85.up_proj", "model.layers.46.mlp.experts.86.up_proj", "model.layers.46.mlp.experts.87.up_proj", "model.layers.46.mlp.experts.88.up_proj", "model.layers.46.mlp.experts.89.up_proj", "model.layers.46.mlp.experts.90.up_proj", "model.layers.46.mlp.experts.91.up_proj", "model.layers.46.mlp.experts.92.up_proj", "model.layers.46.mlp.experts.93.up_proj", "model.layers.46.mlp.experts.94.up_proj", "model.layers.46.mlp.experts.95.up_proj", "model.layers.46.mlp.experts.96.up_proj", "model.layers.46.mlp.experts.97.up_proj", "model.layers.46.mlp.experts.98.up_proj", "model.layers.46.mlp.experts.99.up_proj", "model.layers.46.mlp.experts.100.up_proj", "model.layers.46.mlp.experts.101.up_proj", "model.layers.46.mlp.experts.102.up_proj", "model.layers.46.mlp.experts.103.up_proj", "model.layers.46.mlp.experts.104.up_proj", "model.layers.46.mlp.experts.105.up_proj", "model.layers.46.mlp.experts.106.up_proj", "model.layers.46.mlp.experts.107.up_proj", "model.layers.46.mlp.experts.108.up_proj", "model.layers.46.mlp.experts.109.up_proj", "model.layers.46.mlp.experts.110.up_proj", "model.layers.46.mlp.experts.111.up_proj", "model.layers.46.mlp.experts.112.up_proj", "model.layers.46.mlp.experts.113.up_proj", "model.layers.46.mlp.experts.114.up_proj", "model.layers.46.mlp.experts.115.up_proj", "model.layers.46.mlp.experts.116.up_proj", "model.layers.46.mlp.experts.117.up_proj", "model.layers.46.mlp.experts.118.up_proj", "model.layers.46.mlp.experts.119.up_proj", "model.layers.46.mlp.experts.120.up_proj", "model.layers.46.mlp.experts.121.up_proj", "model.layers.46.mlp.experts.122.up_proj", "model.layers.46.mlp.experts.123.up_proj", "model.layers.46.mlp.experts.124.up_proj", "model.layers.46.mlp.experts.125.up_proj", "model.layers.46.mlp.experts.126.up_proj", "model.layers.46.mlp.experts.127.up_proj", "model.layers.46.mlp.experts.128.up_proj", "model.layers.46.mlp.experts.129.up_proj", "model.layers.46.mlp.experts.130.up_proj", "model.layers.46.mlp.experts.131.up_proj", "model.layers.46.mlp.experts.132.up_proj", "model.layers.46.mlp.experts.133.up_proj", "model.layers.46.mlp.experts.134.up_proj", "model.layers.46.mlp.experts.135.up_proj", "model.layers.46.mlp.experts.136.up_proj", "model.layers.46.mlp.experts.137.up_proj", "model.layers.46.mlp.experts.138.up_proj", "model.layers.46.mlp.experts.139.up_proj", "model.layers.46.mlp.experts.140.up_proj", "model.layers.46.mlp.experts.141.up_proj", "model.layers.46.mlp.experts.142.up_proj", "model.layers.46.mlp.experts.143.up_proj", "model.layers.46.mlp.experts.144.up_proj", "model.layers.46.mlp.experts.145.up_proj", "model.layers.46.mlp.experts.146.up_proj", "model.layers.46.mlp.experts.147.up_proj", "model.layers.46.mlp.experts.148.up_proj", "model.layers.46.mlp.experts.149.up_proj", "model.layers.46.mlp.experts.150.up_proj", "model.layers.46.mlp.experts.151.up_proj", "model.layers.46.mlp.experts.152.up_proj", "model.layers.46.mlp.experts.153.up_proj", "model.layers.46.mlp.experts.154.up_proj", "model.layers.46.mlp.experts.155.up_proj", "model.layers.46.mlp.experts.156.up_proj", "model.layers.46.mlp.experts.157.up_proj", "model.layers.46.mlp.experts.158.up_proj", "model.layers.46.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -2.3964047431948576e-05, "dbits": 2516582400 }, { "dkld": -0.0001313609071075944, "dbits": 5033164800 }, { "dkld": -0.0002025883644819343, "dbits": 7549747200 }, { "dkld": -0.00018773637712002356, "dbits": 12582912000 } ] }, { "idx": 278, "layers": [ "model.layers.46.mlp.experts.0.down_proj", "model.layers.46.mlp.experts.1.down_proj", "model.layers.46.mlp.experts.2.down_proj", "model.layers.46.mlp.experts.3.down_proj", "model.layers.46.mlp.experts.4.down_proj", "model.layers.46.mlp.experts.5.down_proj", "model.layers.46.mlp.experts.6.down_proj", "model.layers.46.mlp.experts.7.down_proj", "model.layers.46.mlp.experts.8.down_proj", "model.layers.46.mlp.experts.9.down_proj", "model.layers.46.mlp.experts.10.down_proj", "model.layers.46.mlp.experts.11.down_proj", "model.layers.46.mlp.experts.12.down_proj", "model.layers.46.mlp.experts.13.down_proj", "model.layers.46.mlp.experts.14.down_proj", "model.layers.46.mlp.experts.15.down_proj", "model.layers.46.mlp.experts.16.down_proj", "model.layers.46.mlp.experts.17.down_proj", "model.layers.46.mlp.experts.18.down_proj", "model.layers.46.mlp.experts.19.down_proj", "model.layers.46.mlp.experts.20.down_proj", "model.layers.46.mlp.experts.21.down_proj", "model.layers.46.mlp.experts.22.down_proj", "model.layers.46.mlp.experts.23.down_proj", "model.layers.46.mlp.experts.24.down_proj", "model.layers.46.mlp.experts.25.down_proj", "model.layers.46.mlp.experts.26.down_proj", "model.layers.46.mlp.experts.27.down_proj", "model.layers.46.mlp.experts.28.down_proj", "model.layers.46.mlp.experts.29.down_proj", "model.layers.46.mlp.experts.30.down_proj", "model.layers.46.mlp.experts.31.down_proj", "model.layers.46.mlp.experts.32.down_proj", "model.layers.46.mlp.experts.33.down_proj", "model.layers.46.mlp.experts.34.down_proj", "model.layers.46.mlp.experts.35.down_proj", "model.layers.46.mlp.experts.36.down_proj", "model.layers.46.mlp.experts.37.down_proj", "model.layers.46.mlp.experts.38.down_proj", "model.layers.46.mlp.experts.39.down_proj", "model.layers.46.mlp.experts.40.down_proj", "model.layers.46.mlp.experts.41.down_proj", "model.layers.46.mlp.experts.42.down_proj", "model.layers.46.mlp.experts.43.down_proj", "model.layers.46.mlp.experts.44.down_proj", "model.layers.46.mlp.experts.45.down_proj", "model.layers.46.mlp.experts.46.down_proj", "model.layers.46.mlp.experts.47.down_proj", "model.layers.46.mlp.experts.48.down_proj", "model.layers.46.mlp.experts.49.down_proj", "model.layers.46.mlp.experts.50.down_proj", "model.layers.46.mlp.experts.51.down_proj", "model.layers.46.mlp.experts.52.down_proj", "model.layers.46.mlp.experts.53.down_proj", "model.layers.46.mlp.experts.54.down_proj", "model.layers.46.mlp.experts.55.down_proj", "model.layers.46.mlp.experts.56.down_proj", "model.layers.46.mlp.experts.57.down_proj", "model.layers.46.mlp.experts.58.down_proj", "model.layers.46.mlp.experts.59.down_proj", "model.layers.46.mlp.experts.60.down_proj", "model.layers.46.mlp.experts.61.down_proj", "model.layers.46.mlp.experts.62.down_proj", "model.layers.46.mlp.experts.63.down_proj", "model.layers.46.mlp.experts.64.down_proj", "model.layers.46.mlp.experts.65.down_proj", "model.layers.46.mlp.experts.66.down_proj", "model.layers.46.mlp.experts.67.down_proj", "model.layers.46.mlp.experts.68.down_proj", "model.layers.46.mlp.experts.69.down_proj", "model.layers.46.mlp.experts.70.down_proj", "model.layers.46.mlp.experts.71.down_proj", "model.layers.46.mlp.experts.72.down_proj", "model.layers.46.mlp.experts.73.down_proj", "model.layers.46.mlp.experts.74.down_proj", "model.layers.46.mlp.experts.75.down_proj", "model.layers.46.mlp.experts.76.down_proj", "model.layers.46.mlp.experts.77.down_proj", "model.layers.46.mlp.experts.78.down_proj", "model.layers.46.mlp.experts.79.down_proj", "model.layers.46.mlp.experts.80.down_proj", "model.layers.46.mlp.experts.81.down_proj", "model.layers.46.mlp.experts.82.down_proj", "model.layers.46.mlp.experts.83.down_proj", "model.layers.46.mlp.experts.84.down_proj", "model.layers.46.mlp.experts.85.down_proj", "model.layers.46.mlp.experts.86.down_proj", "model.layers.46.mlp.experts.87.down_proj", "model.layers.46.mlp.experts.88.down_proj", "model.layers.46.mlp.experts.89.down_proj", "model.layers.46.mlp.experts.90.down_proj", "model.layers.46.mlp.experts.91.down_proj", "model.layers.46.mlp.experts.92.down_proj", "model.layers.46.mlp.experts.93.down_proj", "model.layers.46.mlp.experts.94.down_proj", "model.layers.46.mlp.experts.95.down_proj", "model.layers.46.mlp.experts.96.down_proj", "model.layers.46.mlp.experts.97.down_proj", "model.layers.46.mlp.experts.98.down_proj", "model.layers.46.mlp.experts.99.down_proj", "model.layers.46.mlp.experts.100.down_proj", "model.layers.46.mlp.experts.101.down_proj", "model.layers.46.mlp.experts.102.down_proj", "model.layers.46.mlp.experts.103.down_proj", "model.layers.46.mlp.experts.104.down_proj", "model.layers.46.mlp.experts.105.down_proj", "model.layers.46.mlp.experts.106.down_proj", "model.layers.46.mlp.experts.107.down_proj", "model.layers.46.mlp.experts.108.down_proj", "model.layers.46.mlp.experts.109.down_proj", "model.layers.46.mlp.experts.110.down_proj", "model.layers.46.mlp.experts.111.down_proj", "model.layers.46.mlp.experts.112.down_proj", "model.layers.46.mlp.experts.113.down_proj", "model.layers.46.mlp.experts.114.down_proj", "model.layers.46.mlp.experts.115.down_proj", "model.layers.46.mlp.experts.116.down_proj", "model.layers.46.mlp.experts.117.down_proj", "model.layers.46.mlp.experts.118.down_proj", "model.layers.46.mlp.experts.119.down_proj", "model.layers.46.mlp.experts.120.down_proj", "model.layers.46.mlp.experts.121.down_proj", "model.layers.46.mlp.experts.122.down_proj", "model.layers.46.mlp.experts.123.down_proj", "model.layers.46.mlp.experts.124.down_proj", "model.layers.46.mlp.experts.125.down_proj", "model.layers.46.mlp.experts.126.down_proj", "model.layers.46.mlp.experts.127.down_proj", "model.layers.46.mlp.experts.128.down_proj", "model.layers.46.mlp.experts.129.down_proj", "model.layers.46.mlp.experts.130.down_proj", "model.layers.46.mlp.experts.131.down_proj", "model.layers.46.mlp.experts.132.down_proj", "model.layers.46.mlp.experts.133.down_proj", "model.layers.46.mlp.experts.134.down_proj", "model.layers.46.mlp.experts.135.down_proj", "model.layers.46.mlp.experts.136.down_proj", "model.layers.46.mlp.experts.137.down_proj", "model.layers.46.mlp.experts.138.down_proj", "model.layers.46.mlp.experts.139.down_proj", "model.layers.46.mlp.experts.140.down_proj", "model.layers.46.mlp.experts.141.down_proj", "model.layers.46.mlp.experts.142.down_proj", "model.layers.46.mlp.experts.143.down_proj", "model.layers.46.mlp.experts.144.down_proj", "model.layers.46.mlp.experts.145.down_proj", "model.layers.46.mlp.experts.146.down_proj", "model.layers.46.mlp.experts.147.down_proj", "model.layers.46.mlp.experts.148.down_proj", "model.layers.46.mlp.experts.149.down_proj", "model.layers.46.mlp.experts.150.down_proj", "model.layers.46.mlp.experts.151.down_proj", "model.layers.46.mlp.experts.152.down_proj", "model.layers.46.mlp.experts.153.down_proj", "model.layers.46.mlp.experts.154.down_proj", "model.layers.46.mlp.experts.155.down_proj", "model.layers.46.mlp.experts.156.down_proj", "model.layers.46.mlp.experts.157.down_proj", "model.layers.46.mlp.experts.158.down_proj", "model.layers.46.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002993875183165129, "dbits": 1258291200 }, { "dkld": -0.00032531656324863434, "dbits": 2516582400 }, { "dkld": -0.00036356812343002476, "dbits": 3774873600 }, { "dkld": -0.0003458688035607449, "dbits": 6291456000 } ] }, { "idx": 279, "layers": [ "model.layers.47.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00029968153685332177, "dbits": 62914560 }, { "dkld": -0.00023342343047261238, "dbits": 125829120 }, { "dkld": -0.00020310450345278896, "dbits": 188743680 }, { "dkld": -0.00018371492624283392, "dbits": 314572800 } ] }, { "idx": 280, "layers": [ "model.layers.47.self_attn.k_proj", "model.layers.47.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007816144265234554, "dbits": 10485760 }, { "dkld": -0.0007025949656963432, "dbits": 20971520 }, { "dkld": -0.0006927394308149926, "dbits": 31457280 }, { "dkld": -0.0006931796669960022, "dbits": 52428800 } ] }, { "idx": 281, "layers": [ "model.layers.47.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00022202087566257356, "dbits": 62914560 }, { "dkld": -0.00030448790639639456, "dbits": 125829120 }, { "dkld": -0.0002779069356620423, "dbits": 188743680 }, { "dkld": -0.00023386636748910783, "dbits": 314572800 } ] }, { "idx": 282, "layers": [ "model.layers.47.mlp.shared_experts.gate_proj", "model.layers.47.mlp.shared_experts.up_proj", "model.layers.47.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00021375892683862807, "dbits": 23592960 }, { "dkld": 0.00014905128628014963, "dbits": 47185920 }, { "dkld": 0.00024298615753649833, "dbits": 70778880 }, { "dkld": 0.0002005530521273502, "dbits": 117964800 } ] }, { "idx": 283, "layers": [ "model.layers.47.mlp.experts.0.gate_proj", "model.layers.47.mlp.experts.1.gate_proj", "model.layers.47.mlp.experts.2.gate_proj", "model.layers.47.mlp.experts.3.gate_proj", "model.layers.47.mlp.experts.4.gate_proj", "model.layers.47.mlp.experts.5.gate_proj", "model.layers.47.mlp.experts.6.gate_proj", "model.layers.47.mlp.experts.7.gate_proj", "model.layers.47.mlp.experts.8.gate_proj", "model.layers.47.mlp.experts.9.gate_proj", "model.layers.47.mlp.experts.10.gate_proj", "model.layers.47.mlp.experts.11.gate_proj", "model.layers.47.mlp.experts.12.gate_proj", "model.layers.47.mlp.experts.13.gate_proj", "model.layers.47.mlp.experts.14.gate_proj", "model.layers.47.mlp.experts.15.gate_proj", "model.layers.47.mlp.experts.16.gate_proj", "model.layers.47.mlp.experts.17.gate_proj", "model.layers.47.mlp.experts.18.gate_proj", "model.layers.47.mlp.experts.19.gate_proj", "model.layers.47.mlp.experts.20.gate_proj", "model.layers.47.mlp.experts.21.gate_proj", "model.layers.47.mlp.experts.22.gate_proj", "model.layers.47.mlp.experts.23.gate_proj", "model.layers.47.mlp.experts.24.gate_proj", "model.layers.47.mlp.experts.25.gate_proj", "model.layers.47.mlp.experts.26.gate_proj", "model.layers.47.mlp.experts.27.gate_proj", "model.layers.47.mlp.experts.28.gate_proj", "model.layers.47.mlp.experts.29.gate_proj", "model.layers.47.mlp.experts.30.gate_proj", "model.layers.47.mlp.experts.31.gate_proj", "model.layers.47.mlp.experts.32.gate_proj", "model.layers.47.mlp.experts.33.gate_proj", "model.layers.47.mlp.experts.34.gate_proj", "model.layers.47.mlp.experts.35.gate_proj", "model.layers.47.mlp.experts.36.gate_proj", "model.layers.47.mlp.experts.37.gate_proj", "model.layers.47.mlp.experts.38.gate_proj", "model.layers.47.mlp.experts.39.gate_proj", "model.layers.47.mlp.experts.40.gate_proj", "model.layers.47.mlp.experts.41.gate_proj", "model.layers.47.mlp.experts.42.gate_proj", "model.layers.47.mlp.experts.43.gate_proj", "model.layers.47.mlp.experts.44.gate_proj", "model.layers.47.mlp.experts.45.gate_proj", "model.layers.47.mlp.experts.46.gate_proj", "model.layers.47.mlp.experts.47.gate_proj", "model.layers.47.mlp.experts.48.gate_proj", "model.layers.47.mlp.experts.49.gate_proj", "model.layers.47.mlp.experts.50.gate_proj", "model.layers.47.mlp.experts.51.gate_proj", "model.layers.47.mlp.experts.52.gate_proj", "model.layers.47.mlp.experts.53.gate_proj", "model.layers.47.mlp.experts.54.gate_proj", "model.layers.47.mlp.experts.55.gate_proj", "model.layers.47.mlp.experts.56.gate_proj", "model.layers.47.mlp.experts.57.gate_proj", "model.layers.47.mlp.experts.58.gate_proj", "model.layers.47.mlp.experts.59.gate_proj", "model.layers.47.mlp.experts.60.gate_proj", "model.layers.47.mlp.experts.61.gate_proj", "model.layers.47.mlp.experts.62.gate_proj", "model.layers.47.mlp.experts.63.gate_proj", "model.layers.47.mlp.experts.64.gate_proj", "model.layers.47.mlp.experts.65.gate_proj", "model.layers.47.mlp.experts.66.gate_proj", "model.layers.47.mlp.experts.67.gate_proj", "model.layers.47.mlp.experts.68.gate_proj", "model.layers.47.mlp.experts.69.gate_proj", "model.layers.47.mlp.experts.70.gate_proj", "model.layers.47.mlp.experts.71.gate_proj", "model.layers.47.mlp.experts.72.gate_proj", "model.layers.47.mlp.experts.73.gate_proj", "model.layers.47.mlp.experts.74.gate_proj", "model.layers.47.mlp.experts.75.gate_proj", "model.layers.47.mlp.experts.76.gate_proj", "model.layers.47.mlp.experts.77.gate_proj", "model.layers.47.mlp.experts.78.gate_proj", "model.layers.47.mlp.experts.79.gate_proj", "model.layers.47.mlp.experts.80.gate_proj", "model.layers.47.mlp.experts.81.gate_proj", "model.layers.47.mlp.experts.82.gate_proj", "model.layers.47.mlp.experts.83.gate_proj", "model.layers.47.mlp.experts.84.gate_proj", "model.layers.47.mlp.experts.85.gate_proj", "model.layers.47.mlp.experts.86.gate_proj", "model.layers.47.mlp.experts.87.gate_proj", "model.layers.47.mlp.experts.88.gate_proj", "model.layers.47.mlp.experts.89.gate_proj", "model.layers.47.mlp.experts.90.gate_proj", "model.layers.47.mlp.experts.91.gate_proj", "model.layers.47.mlp.experts.92.gate_proj", "model.layers.47.mlp.experts.93.gate_proj", "model.layers.47.mlp.experts.94.gate_proj", "model.layers.47.mlp.experts.95.gate_proj", "model.layers.47.mlp.experts.96.gate_proj", "model.layers.47.mlp.experts.97.gate_proj", "model.layers.47.mlp.experts.98.gate_proj", "model.layers.47.mlp.experts.99.gate_proj", "model.layers.47.mlp.experts.100.gate_proj", "model.layers.47.mlp.experts.101.gate_proj", "model.layers.47.mlp.experts.102.gate_proj", "model.layers.47.mlp.experts.103.gate_proj", "model.layers.47.mlp.experts.104.gate_proj", "model.layers.47.mlp.experts.105.gate_proj", "model.layers.47.mlp.experts.106.gate_proj", "model.layers.47.mlp.experts.107.gate_proj", "model.layers.47.mlp.experts.108.gate_proj", "model.layers.47.mlp.experts.109.gate_proj", "model.layers.47.mlp.experts.110.gate_proj", "model.layers.47.mlp.experts.111.gate_proj", "model.layers.47.mlp.experts.112.gate_proj", "model.layers.47.mlp.experts.113.gate_proj", "model.layers.47.mlp.experts.114.gate_proj", "model.layers.47.mlp.experts.115.gate_proj", "model.layers.47.mlp.experts.116.gate_proj", "model.layers.47.mlp.experts.117.gate_proj", "model.layers.47.mlp.experts.118.gate_proj", "model.layers.47.mlp.experts.119.gate_proj", "model.layers.47.mlp.experts.120.gate_proj", "model.layers.47.mlp.experts.121.gate_proj", "model.layers.47.mlp.experts.122.gate_proj", "model.layers.47.mlp.experts.123.gate_proj", "model.layers.47.mlp.experts.124.gate_proj", "model.layers.47.mlp.experts.125.gate_proj", "model.layers.47.mlp.experts.126.gate_proj", "model.layers.47.mlp.experts.127.gate_proj", "model.layers.47.mlp.experts.128.gate_proj", "model.layers.47.mlp.experts.129.gate_proj", "model.layers.47.mlp.experts.130.gate_proj", "model.layers.47.mlp.experts.131.gate_proj", "model.layers.47.mlp.experts.132.gate_proj", "model.layers.47.mlp.experts.133.gate_proj", "model.layers.47.mlp.experts.134.gate_proj", "model.layers.47.mlp.experts.135.gate_proj", "model.layers.47.mlp.experts.136.gate_proj", "model.layers.47.mlp.experts.137.gate_proj", "model.layers.47.mlp.experts.138.gate_proj", "model.layers.47.mlp.experts.139.gate_proj", "model.layers.47.mlp.experts.140.gate_proj", "model.layers.47.mlp.experts.141.gate_proj", "model.layers.47.mlp.experts.142.gate_proj", "model.layers.47.mlp.experts.143.gate_proj", "model.layers.47.mlp.experts.144.gate_proj", "model.layers.47.mlp.experts.145.gate_proj", "model.layers.47.mlp.experts.146.gate_proj", "model.layers.47.mlp.experts.147.gate_proj", "model.layers.47.mlp.experts.148.gate_proj", "model.layers.47.mlp.experts.149.gate_proj", "model.layers.47.mlp.experts.150.gate_proj", "model.layers.47.mlp.experts.151.gate_proj", "model.layers.47.mlp.experts.152.gate_proj", "model.layers.47.mlp.experts.153.gate_proj", "model.layers.47.mlp.experts.154.gate_proj", "model.layers.47.mlp.experts.155.gate_proj", "model.layers.47.mlp.experts.156.gate_proj", "model.layers.47.mlp.experts.157.gate_proj", "model.layers.47.mlp.experts.158.gate_proj", "model.layers.47.mlp.experts.159.gate_proj", "model.layers.47.mlp.experts.0.up_proj", "model.layers.47.mlp.experts.1.up_proj", "model.layers.47.mlp.experts.2.up_proj", "model.layers.47.mlp.experts.3.up_proj", "model.layers.47.mlp.experts.4.up_proj", "model.layers.47.mlp.experts.5.up_proj", "model.layers.47.mlp.experts.6.up_proj", "model.layers.47.mlp.experts.7.up_proj", "model.layers.47.mlp.experts.8.up_proj", "model.layers.47.mlp.experts.9.up_proj", "model.layers.47.mlp.experts.10.up_proj", "model.layers.47.mlp.experts.11.up_proj", "model.layers.47.mlp.experts.12.up_proj", "model.layers.47.mlp.experts.13.up_proj", "model.layers.47.mlp.experts.14.up_proj", "model.layers.47.mlp.experts.15.up_proj", "model.layers.47.mlp.experts.16.up_proj", "model.layers.47.mlp.experts.17.up_proj", "model.layers.47.mlp.experts.18.up_proj", "model.layers.47.mlp.experts.19.up_proj", "model.layers.47.mlp.experts.20.up_proj", "model.layers.47.mlp.experts.21.up_proj", "model.layers.47.mlp.experts.22.up_proj", "model.layers.47.mlp.experts.23.up_proj", "model.layers.47.mlp.experts.24.up_proj", "model.layers.47.mlp.experts.25.up_proj", "model.layers.47.mlp.experts.26.up_proj", "model.layers.47.mlp.experts.27.up_proj", "model.layers.47.mlp.experts.28.up_proj", "model.layers.47.mlp.experts.29.up_proj", "model.layers.47.mlp.experts.30.up_proj", "model.layers.47.mlp.experts.31.up_proj", "model.layers.47.mlp.experts.32.up_proj", "model.layers.47.mlp.experts.33.up_proj", "model.layers.47.mlp.experts.34.up_proj", "model.layers.47.mlp.experts.35.up_proj", "model.layers.47.mlp.experts.36.up_proj", "model.layers.47.mlp.experts.37.up_proj", "model.layers.47.mlp.experts.38.up_proj", "model.layers.47.mlp.experts.39.up_proj", "model.layers.47.mlp.experts.40.up_proj", "model.layers.47.mlp.experts.41.up_proj", "model.layers.47.mlp.experts.42.up_proj", "model.layers.47.mlp.experts.43.up_proj", "model.layers.47.mlp.experts.44.up_proj", "model.layers.47.mlp.experts.45.up_proj", "model.layers.47.mlp.experts.46.up_proj", "model.layers.47.mlp.experts.47.up_proj", "model.layers.47.mlp.experts.48.up_proj", "model.layers.47.mlp.experts.49.up_proj", "model.layers.47.mlp.experts.50.up_proj", "model.layers.47.mlp.experts.51.up_proj", "model.layers.47.mlp.experts.52.up_proj", "model.layers.47.mlp.experts.53.up_proj", "model.layers.47.mlp.experts.54.up_proj", "model.layers.47.mlp.experts.55.up_proj", "model.layers.47.mlp.experts.56.up_proj", "model.layers.47.mlp.experts.57.up_proj", "model.layers.47.mlp.experts.58.up_proj", "model.layers.47.mlp.experts.59.up_proj", "model.layers.47.mlp.experts.60.up_proj", "model.layers.47.mlp.experts.61.up_proj", "model.layers.47.mlp.experts.62.up_proj", "model.layers.47.mlp.experts.63.up_proj", "model.layers.47.mlp.experts.64.up_proj", "model.layers.47.mlp.experts.65.up_proj", "model.layers.47.mlp.experts.66.up_proj", "model.layers.47.mlp.experts.67.up_proj", "model.layers.47.mlp.experts.68.up_proj", "model.layers.47.mlp.experts.69.up_proj", "model.layers.47.mlp.experts.70.up_proj", "model.layers.47.mlp.experts.71.up_proj", "model.layers.47.mlp.experts.72.up_proj", "model.layers.47.mlp.experts.73.up_proj", "model.layers.47.mlp.experts.74.up_proj", "model.layers.47.mlp.experts.75.up_proj", "model.layers.47.mlp.experts.76.up_proj", "model.layers.47.mlp.experts.77.up_proj", "model.layers.47.mlp.experts.78.up_proj", "model.layers.47.mlp.experts.79.up_proj", "model.layers.47.mlp.experts.80.up_proj", "model.layers.47.mlp.experts.81.up_proj", "model.layers.47.mlp.experts.82.up_proj", "model.layers.47.mlp.experts.83.up_proj", "model.layers.47.mlp.experts.84.up_proj", "model.layers.47.mlp.experts.85.up_proj", "model.layers.47.mlp.experts.86.up_proj", "model.layers.47.mlp.experts.87.up_proj", "model.layers.47.mlp.experts.88.up_proj", "model.layers.47.mlp.experts.89.up_proj", "model.layers.47.mlp.experts.90.up_proj", "model.layers.47.mlp.experts.91.up_proj", "model.layers.47.mlp.experts.92.up_proj", "model.layers.47.mlp.experts.93.up_proj", "model.layers.47.mlp.experts.94.up_proj", "model.layers.47.mlp.experts.95.up_proj", "model.layers.47.mlp.experts.96.up_proj", "model.layers.47.mlp.experts.97.up_proj", "model.layers.47.mlp.experts.98.up_proj", "model.layers.47.mlp.experts.99.up_proj", "model.layers.47.mlp.experts.100.up_proj", "model.layers.47.mlp.experts.101.up_proj", "model.layers.47.mlp.experts.102.up_proj", "model.layers.47.mlp.experts.103.up_proj", "model.layers.47.mlp.experts.104.up_proj", "model.layers.47.mlp.experts.105.up_proj", "model.layers.47.mlp.experts.106.up_proj", "model.layers.47.mlp.experts.107.up_proj", "model.layers.47.mlp.experts.108.up_proj", "model.layers.47.mlp.experts.109.up_proj", "model.layers.47.mlp.experts.110.up_proj", "model.layers.47.mlp.experts.111.up_proj", "model.layers.47.mlp.experts.112.up_proj", "model.layers.47.mlp.experts.113.up_proj", "model.layers.47.mlp.experts.114.up_proj", "model.layers.47.mlp.experts.115.up_proj", "model.layers.47.mlp.experts.116.up_proj", "model.layers.47.mlp.experts.117.up_proj", "model.layers.47.mlp.experts.118.up_proj", "model.layers.47.mlp.experts.119.up_proj", "model.layers.47.mlp.experts.120.up_proj", "model.layers.47.mlp.experts.121.up_proj", "model.layers.47.mlp.experts.122.up_proj", "model.layers.47.mlp.experts.123.up_proj", "model.layers.47.mlp.experts.124.up_proj", "model.layers.47.mlp.experts.125.up_proj", "model.layers.47.mlp.experts.126.up_proj", "model.layers.47.mlp.experts.127.up_proj", "model.layers.47.mlp.experts.128.up_proj", "model.layers.47.mlp.experts.129.up_proj", "model.layers.47.mlp.experts.130.up_proj", "model.layers.47.mlp.experts.131.up_proj", "model.layers.47.mlp.experts.132.up_proj", "model.layers.47.mlp.experts.133.up_proj", "model.layers.47.mlp.experts.134.up_proj", "model.layers.47.mlp.experts.135.up_proj", "model.layers.47.mlp.experts.136.up_proj", "model.layers.47.mlp.experts.137.up_proj", "model.layers.47.mlp.experts.138.up_proj", "model.layers.47.mlp.experts.139.up_proj", "model.layers.47.mlp.experts.140.up_proj", "model.layers.47.mlp.experts.141.up_proj", "model.layers.47.mlp.experts.142.up_proj", "model.layers.47.mlp.experts.143.up_proj", "model.layers.47.mlp.experts.144.up_proj", "model.layers.47.mlp.experts.145.up_proj", "model.layers.47.mlp.experts.146.up_proj", "model.layers.47.mlp.experts.147.up_proj", "model.layers.47.mlp.experts.148.up_proj", "model.layers.47.mlp.experts.149.up_proj", "model.layers.47.mlp.experts.150.up_proj", "model.layers.47.mlp.experts.151.up_proj", "model.layers.47.mlp.experts.152.up_proj", "model.layers.47.mlp.experts.153.up_proj", "model.layers.47.mlp.experts.154.up_proj", "model.layers.47.mlp.experts.155.up_proj", "model.layers.47.mlp.experts.156.up_proj", "model.layers.47.mlp.experts.157.up_proj", "model.layers.47.mlp.experts.158.up_proj", "model.layers.47.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0002377570606768159, "dbits": 2516582400 }, { "dkld": -0.00023700352758170007, "dbits": 5033164800 }, { "dkld": -0.00023041237145662863, "dbits": 7549747200 }, { "dkld": -0.00024204980581998825, "dbits": 12582912000 } ] }, { "idx": 284, "layers": [ "model.layers.47.mlp.experts.0.down_proj", "model.layers.47.mlp.experts.1.down_proj", "model.layers.47.mlp.experts.2.down_proj", "model.layers.47.mlp.experts.3.down_proj", "model.layers.47.mlp.experts.4.down_proj", "model.layers.47.mlp.experts.5.down_proj", "model.layers.47.mlp.experts.6.down_proj", "model.layers.47.mlp.experts.7.down_proj", "model.layers.47.mlp.experts.8.down_proj", "model.layers.47.mlp.experts.9.down_proj", "model.layers.47.mlp.experts.10.down_proj", "model.layers.47.mlp.experts.11.down_proj", "model.layers.47.mlp.experts.12.down_proj", "model.layers.47.mlp.experts.13.down_proj", "model.layers.47.mlp.experts.14.down_proj", "model.layers.47.mlp.experts.15.down_proj", "model.layers.47.mlp.experts.16.down_proj", "model.layers.47.mlp.experts.17.down_proj", "model.layers.47.mlp.experts.18.down_proj", "model.layers.47.mlp.experts.19.down_proj", "model.layers.47.mlp.experts.20.down_proj", "model.layers.47.mlp.experts.21.down_proj", "model.layers.47.mlp.experts.22.down_proj", "model.layers.47.mlp.experts.23.down_proj", "model.layers.47.mlp.experts.24.down_proj", "model.layers.47.mlp.experts.25.down_proj", "model.layers.47.mlp.experts.26.down_proj", "model.layers.47.mlp.experts.27.down_proj", "model.layers.47.mlp.experts.28.down_proj", "model.layers.47.mlp.experts.29.down_proj", "model.layers.47.mlp.experts.30.down_proj", "model.layers.47.mlp.experts.31.down_proj", "model.layers.47.mlp.experts.32.down_proj", "model.layers.47.mlp.experts.33.down_proj", "model.layers.47.mlp.experts.34.down_proj", "model.layers.47.mlp.experts.35.down_proj", "model.layers.47.mlp.experts.36.down_proj", "model.layers.47.mlp.experts.37.down_proj", "model.layers.47.mlp.experts.38.down_proj", "model.layers.47.mlp.experts.39.down_proj", "model.layers.47.mlp.experts.40.down_proj", "model.layers.47.mlp.experts.41.down_proj", "model.layers.47.mlp.experts.42.down_proj", "model.layers.47.mlp.experts.43.down_proj", "model.layers.47.mlp.experts.44.down_proj", "model.layers.47.mlp.experts.45.down_proj", "model.layers.47.mlp.experts.46.down_proj", "model.layers.47.mlp.experts.47.down_proj", "model.layers.47.mlp.experts.48.down_proj", "model.layers.47.mlp.experts.49.down_proj", "model.layers.47.mlp.experts.50.down_proj", "model.layers.47.mlp.experts.51.down_proj", "model.layers.47.mlp.experts.52.down_proj", "model.layers.47.mlp.experts.53.down_proj", "model.layers.47.mlp.experts.54.down_proj", "model.layers.47.mlp.experts.55.down_proj", "model.layers.47.mlp.experts.56.down_proj", "model.layers.47.mlp.experts.57.down_proj", "model.layers.47.mlp.experts.58.down_proj", "model.layers.47.mlp.experts.59.down_proj", "model.layers.47.mlp.experts.60.down_proj", "model.layers.47.mlp.experts.61.down_proj", "model.layers.47.mlp.experts.62.down_proj", "model.layers.47.mlp.experts.63.down_proj", "model.layers.47.mlp.experts.64.down_proj", "model.layers.47.mlp.experts.65.down_proj", "model.layers.47.mlp.experts.66.down_proj", "model.layers.47.mlp.experts.67.down_proj", "model.layers.47.mlp.experts.68.down_proj", "model.layers.47.mlp.experts.69.down_proj", "model.layers.47.mlp.experts.70.down_proj", "model.layers.47.mlp.experts.71.down_proj", "model.layers.47.mlp.experts.72.down_proj", "model.layers.47.mlp.experts.73.down_proj", "model.layers.47.mlp.experts.74.down_proj", "model.layers.47.mlp.experts.75.down_proj", "model.layers.47.mlp.experts.76.down_proj", "model.layers.47.mlp.experts.77.down_proj", "model.layers.47.mlp.experts.78.down_proj", "model.layers.47.mlp.experts.79.down_proj", "model.layers.47.mlp.experts.80.down_proj", "model.layers.47.mlp.experts.81.down_proj", "model.layers.47.mlp.experts.82.down_proj", "model.layers.47.mlp.experts.83.down_proj", "model.layers.47.mlp.experts.84.down_proj", "model.layers.47.mlp.experts.85.down_proj", "model.layers.47.mlp.experts.86.down_proj", "model.layers.47.mlp.experts.87.down_proj", "model.layers.47.mlp.experts.88.down_proj", "model.layers.47.mlp.experts.89.down_proj", "model.layers.47.mlp.experts.90.down_proj", "model.layers.47.mlp.experts.91.down_proj", "model.layers.47.mlp.experts.92.down_proj", "model.layers.47.mlp.experts.93.down_proj", "model.layers.47.mlp.experts.94.down_proj", "model.layers.47.mlp.experts.95.down_proj", "model.layers.47.mlp.experts.96.down_proj", "model.layers.47.mlp.experts.97.down_proj", "model.layers.47.mlp.experts.98.down_proj", "model.layers.47.mlp.experts.99.down_proj", "model.layers.47.mlp.experts.100.down_proj", "model.layers.47.mlp.experts.101.down_proj", "model.layers.47.mlp.experts.102.down_proj", "model.layers.47.mlp.experts.103.down_proj", "model.layers.47.mlp.experts.104.down_proj", "model.layers.47.mlp.experts.105.down_proj", "model.layers.47.mlp.experts.106.down_proj", "model.layers.47.mlp.experts.107.down_proj", "model.layers.47.mlp.experts.108.down_proj", "model.layers.47.mlp.experts.109.down_proj", "model.layers.47.mlp.experts.110.down_proj", "model.layers.47.mlp.experts.111.down_proj", "model.layers.47.mlp.experts.112.down_proj", "model.layers.47.mlp.experts.113.down_proj", "model.layers.47.mlp.experts.114.down_proj", "model.layers.47.mlp.experts.115.down_proj", "model.layers.47.mlp.experts.116.down_proj", "model.layers.47.mlp.experts.117.down_proj", "model.layers.47.mlp.experts.118.down_proj", "model.layers.47.mlp.experts.119.down_proj", "model.layers.47.mlp.experts.120.down_proj", "model.layers.47.mlp.experts.121.down_proj", "model.layers.47.mlp.experts.122.down_proj", "model.layers.47.mlp.experts.123.down_proj", "model.layers.47.mlp.experts.124.down_proj", "model.layers.47.mlp.experts.125.down_proj", "model.layers.47.mlp.experts.126.down_proj", "model.layers.47.mlp.experts.127.down_proj", "model.layers.47.mlp.experts.128.down_proj", "model.layers.47.mlp.experts.129.down_proj", "model.layers.47.mlp.experts.130.down_proj", "model.layers.47.mlp.experts.131.down_proj", "model.layers.47.mlp.experts.132.down_proj", "model.layers.47.mlp.experts.133.down_proj", "model.layers.47.mlp.experts.134.down_proj", "model.layers.47.mlp.experts.135.down_proj", "model.layers.47.mlp.experts.136.down_proj", "model.layers.47.mlp.experts.137.down_proj", "model.layers.47.mlp.experts.138.down_proj", "model.layers.47.mlp.experts.139.down_proj", "model.layers.47.mlp.experts.140.down_proj", "model.layers.47.mlp.experts.141.down_proj", "model.layers.47.mlp.experts.142.down_proj", "model.layers.47.mlp.experts.143.down_proj", "model.layers.47.mlp.experts.144.down_proj", "model.layers.47.mlp.experts.145.down_proj", "model.layers.47.mlp.experts.146.down_proj", "model.layers.47.mlp.experts.147.down_proj", "model.layers.47.mlp.experts.148.down_proj", "model.layers.47.mlp.experts.149.down_proj", "model.layers.47.mlp.experts.150.down_proj", "model.layers.47.mlp.experts.151.down_proj", "model.layers.47.mlp.experts.152.down_proj", "model.layers.47.mlp.experts.153.down_proj", "model.layers.47.mlp.experts.154.down_proj", "model.layers.47.mlp.experts.155.down_proj", "model.layers.47.mlp.experts.156.down_proj", "model.layers.47.mlp.experts.157.down_proj", "model.layers.47.mlp.experts.158.down_proj", "model.layers.47.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00015937630087137222, "dbits": 1258291200 }, { "dkld": -0.00019172988831997473, "dbits": 2516582400 }, { "dkld": -0.0002679459750652424, "dbits": 3774873600 }, { "dkld": -0.00024993866682053445, "dbits": 6291456000 } ] }, { "idx": 285, "layers": [ "model.layers.48.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0003480928950011647, "dbits": 62914560 }, { "dkld": 8.105374872684201e-05, "dbits": 125829120 }, { "dkld": 0.00019134934991597852, "dbits": 188743680 }, { "dkld": 0.00017693974077701013, "dbits": 314572800 } ] }, { "idx": 286, "layers": [ "model.layers.48.self_attn.k_proj", "model.layers.48.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0008548403158784024, "dbits": 10485760 }, { "dkld": -0.0012789228931069374, "dbits": 20971520 }, { "dkld": -0.0011828462593257427, "dbits": 31457280 }, { "dkld": -0.0011655594222247656, "dbits": 52428800 } ] }, { "idx": 287, "layers": [ "model.layers.48.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0005704309791326606, "dbits": 62914560 }, { "dkld": -0.0005753517150878934, "dbits": 125829120 }, { "dkld": -0.0005072675645351465, "dbits": 188743680 }, { "dkld": -0.0004758642986416872, "dbits": 314572800 } ] }, { "idx": 288, "layers": [ "model.layers.48.mlp.shared_experts.gate_proj", "model.layers.48.mlp.shared_experts.up_proj", "model.layers.48.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005481851287186257, "dbits": 23592960 }, { "dkld": -0.0003412337973713875, "dbits": 47185920 }, { "dkld": -0.0003336050547659425, "dbits": 70778880 }, { "dkld": -0.00039302082732320387, "dbits": 117964800 } ] }, { "idx": 289, "layers": [ "model.layers.48.mlp.experts.0.gate_proj", "model.layers.48.mlp.experts.1.gate_proj", "model.layers.48.mlp.experts.2.gate_proj", "model.layers.48.mlp.experts.3.gate_proj", "model.layers.48.mlp.experts.4.gate_proj", "model.layers.48.mlp.experts.5.gate_proj", "model.layers.48.mlp.experts.6.gate_proj", "model.layers.48.mlp.experts.7.gate_proj", "model.layers.48.mlp.experts.8.gate_proj", "model.layers.48.mlp.experts.9.gate_proj", "model.layers.48.mlp.experts.10.gate_proj", "model.layers.48.mlp.experts.11.gate_proj", "model.layers.48.mlp.experts.12.gate_proj", "model.layers.48.mlp.experts.13.gate_proj", "model.layers.48.mlp.experts.14.gate_proj", "model.layers.48.mlp.experts.15.gate_proj", "model.layers.48.mlp.experts.16.gate_proj", "model.layers.48.mlp.experts.17.gate_proj", "model.layers.48.mlp.experts.18.gate_proj", "model.layers.48.mlp.experts.19.gate_proj", "model.layers.48.mlp.experts.20.gate_proj", "model.layers.48.mlp.experts.21.gate_proj", "model.layers.48.mlp.experts.22.gate_proj", "model.layers.48.mlp.experts.23.gate_proj", "model.layers.48.mlp.experts.24.gate_proj", "model.layers.48.mlp.experts.25.gate_proj", "model.layers.48.mlp.experts.26.gate_proj", "model.layers.48.mlp.experts.27.gate_proj", "model.layers.48.mlp.experts.28.gate_proj", "model.layers.48.mlp.experts.29.gate_proj", "model.layers.48.mlp.experts.30.gate_proj", "model.layers.48.mlp.experts.31.gate_proj", "model.layers.48.mlp.experts.32.gate_proj", "model.layers.48.mlp.experts.33.gate_proj", "model.layers.48.mlp.experts.34.gate_proj", "model.layers.48.mlp.experts.35.gate_proj", "model.layers.48.mlp.experts.36.gate_proj", "model.layers.48.mlp.experts.37.gate_proj", "model.layers.48.mlp.experts.38.gate_proj", "model.layers.48.mlp.experts.39.gate_proj", "model.layers.48.mlp.experts.40.gate_proj", "model.layers.48.mlp.experts.41.gate_proj", "model.layers.48.mlp.experts.42.gate_proj", "model.layers.48.mlp.experts.43.gate_proj", "model.layers.48.mlp.experts.44.gate_proj", "model.layers.48.mlp.experts.45.gate_proj", "model.layers.48.mlp.experts.46.gate_proj", "model.layers.48.mlp.experts.47.gate_proj", "model.layers.48.mlp.experts.48.gate_proj", "model.layers.48.mlp.experts.49.gate_proj", "model.layers.48.mlp.experts.50.gate_proj", "model.layers.48.mlp.experts.51.gate_proj", "model.layers.48.mlp.experts.52.gate_proj", "model.layers.48.mlp.experts.53.gate_proj", "model.layers.48.mlp.experts.54.gate_proj", "model.layers.48.mlp.experts.55.gate_proj", "model.layers.48.mlp.experts.56.gate_proj", "model.layers.48.mlp.experts.57.gate_proj", "model.layers.48.mlp.experts.58.gate_proj", "model.layers.48.mlp.experts.59.gate_proj", "model.layers.48.mlp.experts.60.gate_proj", "model.layers.48.mlp.experts.61.gate_proj", "model.layers.48.mlp.experts.62.gate_proj", "model.layers.48.mlp.experts.63.gate_proj", "model.layers.48.mlp.experts.64.gate_proj", "model.layers.48.mlp.experts.65.gate_proj", "model.layers.48.mlp.experts.66.gate_proj", "model.layers.48.mlp.experts.67.gate_proj", "model.layers.48.mlp.experts.68.gate_proj", "model.layers.48.mlp.experts.69.gate_proj", "model.layers.48.mlp.experts.70.gate_proj", "model.layers.48.mlp.experts.71.gate_proj", "model.layers.48.mlp.experts.72.gate_proj", "model.layers.48.mlp.experts.73.gate_proj", "model.layers.48.mlp.experts.74.gate_proj", "model.layers.48.mlp.experts.75.gate_proj", "model.layers.48.mlp.experts.76.gate_proj", "model.layers.48.mlp.experts.77.gate_proj", "model.layers.48.mlp.experts.78.gate_proj", "model.layers.48.mlp.experts.79.gate_proj", "model.layers.48.mlp.experts.80.gate_proj", "model.layers.48.mlp.experts.81.gate_proj", "model.layers.48.mlp.experts.82.gate_proj", "model.layers.48.mlp.experts.83.gate_proj", "model.layers.48.mlp.experts.84.gate_proj", "model.layers.48.mlp.experts.85.gate_proj", "model.layers.48.mlp.experts.86.gate_proj", "model.layers.48.mlp.experts.87.gate_proj", "model.layers.48.mlp.experts.88.gate_proj", "model.layers.48.mlp.experts.89.gate_proj", "model.layers.48.mlp.experts.90.gate_proj", "model.layers.48.mlp.experts.91.gate_proj", "model.layers.48.mlp.experts.92.gate_proj", "model.layers.48.mlp.experts.93.gate_proj", "model.layers.48.mlp.experts.94.gate_proj", "model.layers.48.mlp.experts.95.gate_proj", "model.layers.48.mlp.experts.96.gate_proj", "model.layers.48.mlp.experts.97.gate_proj", "model.layers.48.mlp.experts.98.gate_proj", "model.layers.48.mlp.experts.99.gate_proj", "model.layers.48.mlp.experts.100.gate_proj", "model.layers.48.mlp.experts.101.gate_proj", "model.layers.48.mlp.experts.102.gate_proj", "model.layers.48.mlp.experts.103.gate_proj", "model.layers.48.mlp.experts.104.gate_proj", "model.layers.48.mlp.experts.105.gate_proj", "model.layers.48.mlp.experts.106.gate_proj", "model.layers.48.mlp.experts.107.gate_proj", "model.layers.48.mlp.experts.108.gate_proj", "model.layers.48.mlp.experts.109.gate_proj", "model.layers.48.mlp.experts.110.gate_proj", "model.layers.48.mlp.experts.111.gate_proj", "model.layers.48.mlp.experts.112.gate_proj", "model.layers.48.mlp.experts.113.gate_proj", "model.layers.48.mlp.experts.114.gate_proj", "model.layers.48.mlp.experts.115.gate_proj", "model.layers.48.mlp.experts.116.gate_proj", "model.layers.48.mlp.experts.117.gate_proj", "model.layers.48.mlp.experts.118.gate_proj", "model.layers.48.mlp.experts.119.gate_proj", "model.layers.48.mlp.experts.120.gate_proj", "model.layers.48.mlp.experts.121.gate_proj", "model.layers.48.mlp.experts.122.gate_proj", "model.layers.48.mlp.experts.123.gate_proj", "model.layers.48.mlp.experts.124.gate_proj", "model.layers.48.mlp.experts.125.gate_proj", "model.layers.48.mlp.experts.126.gate_proj", "model.layers.48.mlp.experts.127.gate_proj", "model.layers.48.mlp.experts.128.gate_proj", "model.layers.48.mlp.experts.129.gate_proj", "model.layers.48.mlp.experts.130.gate_proj", "model.layers.48.mlp.experts.131.gate_proj", "model.layers.48.mlp.experts.132.gate_proj", "model.layers.48.mlp.experts.133.gate_proj", "model.layers.48.mlp.experts.134.gate_proj", "model.layers.48.mlp.experts.135.gate_proj", "model.layers.48.mlp.experts.136.gate_proj", "model.layers.48.mlp.experts.137.gate_proj", "model.layers.48.mlp.experts.138.gate_proj", "model.layers.48.mlp.experts.139.gate_proj", "model.layers.48.mlp.experts.140.gate_proj", "model.layers.48.mlp.experts.141.gate_proj", "model.layers.48.mlp.experts.142.gate_proj", "model.layers.48.mlp.experts.143.gate_proj", "model.layers.48.mlp.experts.144.gate_proj", "model.layers.48.mlp.experts.145.gate_proj", "model.layers.48.mlp.experts.146.gate_proj", "model.layers.48.mlp.experts.147.gate_proj", "model.layers.48.mlp.experts.148.gate_proj", "model.layers.48.mlp.experts.149.gate_proj", "model.layers.48.mlp.experts.150.gate_proj", "model.layers.48.mlp.experts.151.gate_proj", "model.layers.48.mlp.experts.152.gate_proj", "model.layers.48.mlp.experts.153.gate_proj", "model.layers.48.mlp.experts.154.gate_proj", "model.layers.48.mlp.experts.155.gate_proj", "model.layers.48.mlp.experts.156.gate_proj", "model.layers.48.mlp.experts.157.gate_proj", "model.layers.48.mlp.experts.158.gate_proj", "model.layers.48.mlp.experts.159.gate_proj", "model.layers.48.mlp.experts.0.up_proj", "model.layers.48.mlp.experts.1.up_proj", "model.layers.48.mlp.experts.2.up_proj", "model.layers.48.mlp.experts.3.up_proj", "model.layers.48.mlp.experts.4.up_proj", "model.layers.48.mlp.experts.5.up_proj", "model.layers.48.mlp.experts.6.up_proj", "model.layers.48.mlp.experts.7.up_proj", "model.layers.48.mlp.experts.8.up_proj", "model.layers.48.mlp.experts.9.up_proj", "model.layers.48.mlp.experts.10.up_proj", "model.layers.48.mlp.experts.11.up_proj", "model.layers.48.mlp.experts.12.up_proj", "model.layers.48.mlp.experts.13.up_proj", "model.layers.48.mlp.experts.14.up_proj", "model.layers.48.mlp.experts.15.up_proj", "model.layers.48.mlp.experts.16.up_proj", "model.layers.48.mlp.experts.17.up_proj", "model.layers.48.mlp.experts.18.up_proj", "model.layers.48.mlp.experts.19.up_proj", "model.layers.48.mlp.experts.20.up_proj", "model.layers.48.mlp.experts.21.up_proj", "model.layers.48.mlp.experts.22.up_proj", "model.layers.48.mlp.experts.23.up_proj", "model.layers.48.mlp.experts.24.up_proj", "model.layers.48.mlp.experts.25.up_proj", "model.layers.48.mlp.experts.26.up_proj", "model.layers.48.mlp.experts.27.up_proj", "model.layers.48.mlp.experts.28.up_proj", "model.layers.48.mlp.experts.29.up_proj", "model.layers.48.mlp.experts.30.up_proj", "model.layers.48.mlp.experts.31.up_proj", "model.layers.48.mlp.experts.32.up_proj", "model.layers.48.mlp.experts.33.up_proj", "model.layers.48.mlp.experts.34.up_proj", "model.layers.48.mlp.experts.35.up_proj", "model.layers.48.mlp.experts.36.up_proj", "model.layers.48.mlp.experts.37.up_proj", "model.layers.48.mlp.experts.38.up_proj", "model.layers.48.mlp.experts.39.up_proj", "model.layers.48.mlp.experts.40.up_proj", "model.layers.48.mlp.experts.41.up_proj", "model.layers.48.mlp.experts.42.up_proj", "model.layers.48.mlp.experts.43.up_proj", "model.layers.48.mlp.experts.44.up_proj", "model.layers.48.mlp.experts.45.up_proj", "model.layers.48.mlp.experts.46.up_proj", "model.layers.48.mlp.experts.47.up_proj", "model.layers.48.mlp.experts.48.up_proj", "model.layers.48.mlp.experts.49.up_proj", "model.layers.48.mlp.experts.50.up_proj", "model.layers.48.mlp.experts.51.up_proj", "model.layers.48.mlp.experts.52.up_proj", "model.layers.48.mlp.experts.53.up_proj", "model.layers.48.mlp.experts.54.up_proj", "model.layers.48.mlp.experts.55.up_proj", "model.layers.48.mlp.experts.56.up_proj", "model.layers.48.mlp.experts.57.up_proj", "model.layers.48.mlp.experts.58.up_proj", "model.layers.48.mlp.experts.59.up_proj", "model.layers.48.mlp.experts.60.up_proj", "model.layers.48.mlp.experts.61.up_proj", "model.layers.48.mlp.experts.62.up_proj", "model.layers.48.mlp.experts.63.up_proj", "model.layers.48.mlp.experts.64.up_proj", "model.layers.48.mlp.experts.65.up_proj", "model.layers.48.mlp.experts.66.up_proj", "model.layers.48.mlp.experts.67.up_proj", "model.layers.48.mlp.experts.68.up_proj", "model.layers.48.mlp.experts.69.up_proj", "model.layers.48.mlp.experts.70.up_proj", "model.layers.48.mlp.experts.71.up_proj", "model.layers.48.mlp.experts.72.up_proj", "model.layers.48.mlp.experts.73.up_proj", "model.layers.48.mlp.experts.74.up_proj", "model.layers.48.mlp.experts.75.up_proj", "model.layers.48.mlp.experts.76.up_proj", "model.layers.48.mlp.experts.77.up_proj", "model.layers.48.mlp.experts.78.up_proj", "model.layers.48.mlp.experts.79.up_proj", "model.layers.48.mlp.experts.80.up_proj", "model.layers.48.mlp.experts.81.up_proj", "model.layers.48.mlp.experts.82.up_proj", "model.layers.48.mlp.experts.83.up_proj", "model.layers.48.mlp.experts.84.up_proj", "model.layers.48.mlp.experts.85.up_proj", "model.layers.48.mlp.experts.86.up_proj", "model.layers.48.mlp.experts.87.up_proj", "model.layers.48.mlp.experts.88.up_proj", "model.layers.48.mlp.experts.89.up_proj", "model.layers.48.mlp.experts.90.up_proj", "model.layers.48.mlp.experts.91.up_proj", "model.layers.48.mlp.experts.92.up_proj", "model.layers.48.mlp.experts.93.up_proj", "model.layers.48.mlp.experts.94.up_proj", "model.layers.48.mlp.experts.95.up_proj", "model.layers.48.mlp.experts.96.up_proj", "model.layers.48.mlp.experts.97.up_proj", "model.layers.48.mlp.experts.98.up_proj", "model.layers.48.mlp.experts.99.up_proj", "model.layers.48.mlp.experts.100.up_proj", "model.layers.48.mlp.experts.101.up_proj", "model.layers.48.mlp.experts.102.up_proj", "model.layers.48.mlp.experts.103.up_proj", "model.layers.48.mlp.experts.104.up_proj", "model.layers.48.mlp.experts.105.up_proj", "model.layers.48.mlp.experts.106.up_proj", "model.layers.48.mlp.experts.107.up_proj", "model.layers.48.mlp.experts.108.up_proj", "model.layers.48.mlp.experts.109.up_proj", "model.layers.48.mlp.experts.110.up_proj", "model.layers.48.mlp.experts.111.up_proj", "model.layers.48.mlp.experts.112.up_proj", "model.layers.48.mlp.experts.113.up_proj", "model.layers.48.mlp.experts.114.up_proj", "model.layers.48.mlp.experts.115.up_proj", "model.layers.48.mlp.experts.116.up_proj", "model.layers.48.mlp.experts.117.up_proj", "model.layers.48.mlp.experts.118.up_proj", "model.layers.48.mlp.experts.119.up_proj", "model.layers.48.mlp.experts.120.up_proj", "model.layers.48.mlp.experts.121.up_proj", "model.layers.48.mlp.experts.122.up_proj", "model.layers.48.mlp.experts.123.up_proj", "model.layers.48.mlp.experts.124.up_proj", "model.layers.48.mlp.experts.125.up_proj", "model.layers.48.mlp.experts.126.up_proj", "model.layers.48.mlp.experts.127.up_proj", "model.layers.48.mlp.experts.128.up_proj", "model.layers.48.mlp.experts.129.up_proj", "model.layers.48.mlp.experts.130.up_proj", "model.layers.48.mlp.experts.131.up_proj", "model.layers.48.mlp.experts.132.up_proj", "model.layers.48.mlp.experts.133.up_proj", "model.layers.48.mlp.experts.134.up_proj", "model.layers.48.mlp.experts.135.up_proj", "model.layers.48.mlp.experts.136.up_proj", "model.layers.48.mlp.experts.137.up_proj", "model.layers.48.mlp.experts.138.up_proj", "model.layers.48.mlp.experts.139.up_proj", "model.layers.48.mlp.experts.140.up_proj", "model.layers.48.mlp.experts.141.up_proj", "model.layers.48.mlp.experts.142.up_proj", "model.layers.48.mlp.experts.143.up_proj", "model.layers.48.mlp.experts.144.up_proj", "model.layers.48.mlp.experts.145.up_proj", "model.layers.48.mlp.experts.146.up_proj", "model.layers.48.mlp.experts.147.up_proj", "model.layers.48.mlp.experts.148.up_proj", "model.layers.48.mlp.experts.149.up_proj", "model.layers.48.mlp.experts.150.up_proj", "model.layers.48.mlp.experts.151.up_proj", "model.layers.48.mlp.experts.152.up_proj", "model.layers.48.mlp.experts.153.up_proj", "model.layers.48.mlp.experts.154.up_proj", "model.layers.48.mlp.experts.155.up_proj", "model.layers.48.mlp.experts.156.up_proj", "model.layers.48.mlp.experts.157.up_proj", "model.layers.48.mlp.experts.158.up_proj", "model.layers.48.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0003868428058922402, "dbits": 2516582400 }, { "dkld": -0.00038525937125087617, "dbits": 5033164800 }, { "dkld": -0.0004037499427795438, "dbits": 7549747200 }, { "dkld": -0.0004001275636255769, "dbits": 12582912000 } ] }, { "idx": 290, "layers": [ "model.layers.48.mlp.experts.0.down_proj", "model.layers.48.mlp.experts.1.down_proj", "model.layers.48.mlp.experts.2.down_proj", "model.layers.48.mlp.experts.3.down_proj", "model.layers.48.mlp.experts.4.down_proj", "model.layers.48.mlp.experts.5.down_proj", "model.layers.48.mlp.experts.6.down_proj", "model.layers.48.mlp.experts.7.down_proj", "model.layers.48.mlp.experts.8.down_proj", "model.layers.48.mlp.experts.9.down_proj", "model.layers.48.mlp.experts.10.down_proj", "model.layers.48.mlp.experts.11.down_proj", "model.layers.48.mlp.experts.12.down_proj", "model.layers.48.mlp.experts.13.down_proj", "model.layers.48.mlp.experts.14.down_proj", "model.layers.48.mlp.experts.15.down_proj", "model.layers.48.mlp.experts.16.down_proj", "model.layers.48.mlp.experts.17.down_proj", "model.layers.48.mlp.experts.18.down_proj", "model.layers.48.mlp.experts.19.down_proj", "model.layers.48.mlp.experts.20.down_proj", "model.layers.48.mlp.experts.21.down_proj", "model.layers.48.mlp.experts.22.down_proj", "model.layers.48.mlp.experts.23.down_proj", "model.layers.48.mlp.experts.24.down_proj", "model.layers.48.mlp.experts.25.down_proj", "model.layers.48.mlp.experts.26.down_proj", "model.layers.48.mlp.experts.27.down_proj", "model.layers.48.mlp.experts.28.down_proj", "model.layers.48.mlp.experts.29.down_proj", "model.layers.48.mlp.experts.30.down_proj", "model.layers.48.mlp.experts.31.down_proj", "model.layers.48.mlp.experts.32.down_proj", "model.layers.48.mlp.experts.33.down_proj", "model.layers.48.mlp.experts.34.down_proj", "model.layers.48.mlp.experts.35.down_proj", "model.layers.48.mlp.experts.36.down_proj", "model.layers.48.mlp.experts.37.down_proj", "model.layers.48.mlp.experts.38.down_proj", "model.layers.48.mlp.experts.39.down_proj", "model.layers.48.mlp.experts.40.down_proj", "model.layers.48.mlp.experts.41.down_proj", "model.layers.48.mlp.experts.42.down_proj", "model.layers.48.mlp.experts.43.down_proj", "model.layers.48.mlp.experts.44.down_proj", "model.layers.48.mlp.experts.45.down_proj", "model.layers.48.mlp.experts.46.down_proj", "model.layers.48.mlp.experts.47.down_proj", "model.layers.48.mlp.experts.48.down_proj", "model.layers.48.mlp.experts.49.down_proj", "model.layers.48.mlp.experts.50.down_proj", "model.layers.48.mlp.experts.51.down_proj", "model.layers.48.mlp.experts.52.down_proj", "model.layers.48.mlp.experts.53.down_proj", "model.layers.48.mlp.experts.54.down_proj", "model.layers.48.mlp.experts.55.down_proj", "model.layers.48.mlp.experts.56.down_proj", "model.layers.48.mlp.experts.57.down_proj", "model.layers.48.mlp.experts.58.down_proj", "model.layers.48.mlp.experts.59.down_proj", "model.layers.48.mlp.experts.60.down_proj", "model.layers.48.mlp.experts.61.down_proj", "model.layers.48.mlp.experts.62.down_proj", "model.layers.48.mlp.experts.63.down_proj", "model.layers.48.mlp.experts.64.down_proj", "model.layers.48.mlp.experts.65.down_proj", "model.layers.48.mlp.experts.66.down_proj", "model.layers.48.mlp.experts.67.down_proj", "model.layers.48.mlp.experts.68.down_proj", "model.layers.48.mlp.experts.69.down_proj", "model.layers.48.mlp.experts.70.down_proj", "model.layers.48.mlp.experts.71.down_proj", "model.layers.48.mlp.experts.72.down_proj", "model.layers.48.mlp.experts.73.down_proj", "model.layers.48.mlp.experts.74.down_proj", "model.layers.48.mlp.experts.75.down_proj", "model.layers.48.mlp.experts.76.down_proj", "model.layers.48.mlp.experts.77.down_proj", "model.layers.48.mlp.experts.78.down_proj", "model.layers.48.mlp.experts.79.down_proj", "model.layers.48.mlp.experts.80.down_proj", "model.layers.48.mlp.experts.81.down_proj", "model.layers.48.mlp.experts.82.down_proj", "model.layers.48.mlp.experts.83.down_proj", "model.layers.48.mlp.experts.84.down_proj", "model.layers.48.mlp.experts.85.down_proj", "model.layers.48.mlp.experts.86.down_proj", "model.layers.48.mlp.experts.87.down_proj", "model.layers.48.mlp.experts.88.down_proj", "model.layers.48.mlp.experts.89.down_proj", "model.layers.48.mlp.experts.90.down_proj", "model.layers.48.mlp.experts.91.down_proj", "model.layers.48.mlp.experts.92.down_proj", "model.layers.48.mlp.experts.93.down_proj", "model.layers.48.mlp.experts.94.down_proj", "model.layers.48.mlp.experts.95.down_proj", "model.layers.48.mlp.experts.96.down_proj", "model.layers.48.mlp.experts.97.down_proj", "model.layers.48.mlp.experts.98.down_proj", "model.layers.48.mlp.experts.99.down_proj", "model.layers.48.mlp.experts.100.down_proj", "model.layers.48.mlp.experts.101.down_proj", "model.layers.48.mlp.experts.102.down_proj", "model.layers.48.mlp.experts.103.down_proj", "model.layers.48.mlp.experts.104.down_proj", "model.layers.48.mlp.experts.105.down_proj", "model.layers.48.mlp.experts.106.down_proj", "model.layers.48.mlp.experts.107.down_proj", "model.layers.48.mlp.experts.108.down_proj", "model.layers.48.mlp.experts.109.down_proj", "model.layers.48.mlp.experts.110.down_proj", "model.layers.48.mlp.experts.111.down_proj", "model.layers.48.mlp.experts.112.down_proj", "model.layers.48.mlp.experts.113.down_proj", "model.layers.48.mlp.experts.114.down_proj", "model.layers.48.mlp.experts.115.down_proj", "model.layers.48.mlp.experts.116.down_proj", "model.layers.48.mlp.experts.117.down_proj", "model.layers.48.mlp.experts.118.down_proj", "model.layers.48.mlp.experts.119.down_proj", "model.layers.48.mlp.experts.120.down_proj", "model.layers.48.mlp.experts.121.down_proj", "model.layers.48.mlp.experts.122.down_proj", "model.layers.48.mlp.experts.123.down_proj", "model.layers.48.mlp.experts.124.down_proj", "model.layers.48.mlp.experts.125.down_proj", "model.layers.48.mlp.experts.126.down_proj", "model.layers.48.mlp.experts.127.down_proj", "model.layers.48.mlp.experts.128.down_proj", "model.layers.48.mlp.experts.129.down_proj", "model.layers.48.mlp.experts.130.down_proj", "model.layers.48.mlp.experts.131.down_proj", "model.layers.48.mlp.experts.132.down_proj", "model.layers.48.mlp.experts.133.down_proj", "model.layers.48.mlp.experts.134.down_proj", "model.layers.48.mlp.experts.135.down_proj", "model.layers.48.mlp.experts.136.down_proj", "model.layers.48.mlp.experts.137.down_proj", "model.layers.48.mlp.experts.138.down_proj", "model.layers.48.mlp.experts.139.down_proj", "model.layers.48.mlp.experts.140.down_proj", "model.layers.48.mlp.experts.141.down_proj", "model.layers.48.mlp.experts.142.down_proj", "model.layers.48.mlp.experts.143.down_proj", "model.layers.48.mlp.experts.144.down_proj", "model.layers.48.mlp.experts.145.down_proj", "model.layers.48.mlp.experts.146.down_proj", "model.layers.48.mlp.experts.147.down_proj", "model.layers.48.mlp.experts.148.down_proj", "model.layers.48.mlp.experts.149.down_proj", "model.layers.48.mlp.experts.150.down_proj", "model.layers.48.mlp.experts.151.down_proj", "model.layers.48.mlp.experts.152.down_proj", "model.layers.48.mlp.experts.153.down_proj", "model.layers.48.mlp.experts.154.down_proj", "model.layers.48.mlp.experts.155.down_proj", "model.layers.48.mlp.experts.156.down_proj", "model.layers.48.mlp.experts.157.down_proj", "model.layers.48.mlp.experts.158.down_proj", "model.layers.48.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00025311093777419247, "dbits": 1258291200 }, { "dkld": -0.00026842784136534015, "dbits": 2516582400 }, { "dkld": -0.0002357296645641438, "dbits": 3774873600 }, { "dkld": -0.00024682758376003144, "dbits": 6291456000 } ] }, { "idx": 291, "layers": [ "model.layers.49.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00038392208516598025, "dbits": 62914560 }, { "dkld": -0.0003332739695906667, "dbits": 125829120 }, { "dkld": -0.00021435935050249932, "dbits": 188743680 }, { "dkld": -0.0003170181065797889, "dbits": 314572800 } ] }, { "idx": 292, "layers": [ "model.layers.49.self_attn.k_proj", "model.layers.49.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0002741411328315707, "dbits": 10485760 }, { "dkld": 0.0001932259649038287, "dbits": 20971520 }, { "dkld": 0.0001714436337351688, "dbits": 31457280 }, { "dkld": 0.0002001512795686694, "dbits": 52428800 } ] }, { "idx": 293, "layers": [ "model.layers.49.self_attn.o_proj" ], "candidates": [ { "dkld": -5.6166667491197586e-05, "dbits": 62914560 }, { "dkld": 4.934836179017188e-05, "dbits": 125829120 }, { "dkld": 4.3130759149787035e-05, "dbits": 188743680 }, { "dkld": 1.0940246284008026e-05, "dbits": 314572800 } ] }, { "idx": 294, "layers": [ "model.layers.49.mlp.shared_experts.gate_proj", "model.layers.49.mlp.shared_experts.up_proj", "model.layers.49.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 2.729250118135851e-05, "dbits": 23592960 }, { "dkld": -0.00022060181945562918, "dbits": 47185920 }, { "dkld": -2.9728282243013382e-05, "dbits": 70778880 }, { "dkld": -0.00011159293353557587, "dbits": 117964800 } ] }, { "idx": 295, "layers": [ "model.layers.49.mlp.experts.0.gate_proj", "model.layers.49.mlp.experts.1.gate_proj", "model.layers.49.mlp.experts.2.gate_proj", "model.layers.49.mlp.experts.3.gate_proj", "model.layers.49.mlp.experts.4.gate_proj", "model.layers.49.mlp.experts.5.gate_proj", "model.layers.49.mlp.experts.6.gate_proj", "model.layers.49.mlp.experts.7.gate_proj", "model.layers.49.mlp.experts.8.gate_proj", "model.layers.49.mlp.experts.9.gate_proj", "model.layers.49.mlp.experts.10.gate_proj", "model.layers.49.mlp.experts.11.gate_proj", "model.layers.49.mlp.experts.12.gate_proj", "model.layers.49.mlp.experts.13.gate_proj", "model.layers.49.mlp.experts.14.gate_proj", "model.layers.49.mlp.experts.15.gate_proj", "model.layers.49.mlp.experts.16.gate_proj", "model.layers.49.mlp.experts.17.gate_proj", "model.layers.49.mlp.experts.18.gate_proj", "model.layers.49.mlp.experts.19.gate_proj", "model.layers.49.mlp.experts.20.gate_proj", "model.layers.49.mlp.experts.21.gate_proj", "model.layers.49.mlp.experts.22.gate_proj", "model.layers.49.mlp.experts.23.gate_proj", "model.layers.49.mlp.experts.24.gate_proj", "model.layers.49.mlp.experts.25.gate_proj", "model.layers.49.mlp.experts.26.gate_proj", "model.layers.49.mlp.experts.27.gate_proj", "model.layers.49.mlp.experts.28.gate_proj", "model.layers.49.mlp.experts.29.gate_proj", "model.layers.49.mlp.experts.30.gate_proj", "model.layers.49.mlp.experts.31.gate_proj", "model.layers.49.mlp.experts.32.gate_proj", "model.layers.49.mlp.experts.33.gate_proj", "model.layers.49.mlp.experts.34.gate_proj", "model.layers.49.mlp.experts.35.gate_proj", "model.layers.49.mlp.experts.36.gate_proj", "model.layers.49.mlp.experts.37.gate_proj", "model.layers.49.mlp.experts.38.gate_proj", "model.layers.49.mlp.experts.39.gate_proj", "model.layers.49.mlp.experts.40.gate_proj", "model.layers.49.mlp.experts.41.gate_proj", "model.layers.49.mlp.experts.42.gate_proj", "model.layers.49.mlp.experts.43.gate_proj", "model.layers.49.mlp.experts.44.gate_proj", "model.layers.49.mlp.experts.45.gate_proj", "model.layers.49.mlp.experts.46.gate_proj", "model.layers.49.mlp.experts.47.gate_proj", "model.layers.49.mlp.experts.48.gate_proj", "model.layers.49.mlp.experts.49.gate_proj", "model.layers.49.mlp.experts.50.gate_proj", "model.layers.49.mlp.experts.51.gate_proj", "model.layers.49.mlp.experts.52.gate_proj", "model.layers.49.mlp.experts.53.gate_proj", "model.layers.49.mlp.experts.54.gate_proj", "model.layers.49.mlp.experts.55.gate_proj", "model.layers.49.mlp.experts.56.gate_proj", "model.layers.49.mlp.experts.57.gate_proj", "model.layers.49.mlp.experts.58.gate_proj", "model.layers.49.mlp.experts.59.gate_proj", "model.layers.49.mlp.experts.60.gate_proj", "model.layers.49.mlp.experts.61.gate_proj", "model.layers.49.mlp.experts.62.gate_proj", "model.layers.49.mlp.experts.63.gate_proj", "model.layers.49.mlp.experts.64.gate_proj", "model.layers.49.mlp.experts.65.gate_proj", "model.layers.49.mlp.experts.66.gate_proj", "model.layers.49.mlp.experts.67.gate_proj", "model.layers.49.mlp.experts.68.gate_proj", "model.layers.49.mlp.experts.69.gate_proj", "model.layers.49.mlp.experts.70.gate_proj", "model.layers.49.mlp.experts.71.gate_proj", "model.layers.49.mlp.experts.72.gate_proj", "model.layers.49.mlp.experts.73.gate_proj", "model.layers.49.mlp.experts.74.gate_proj", "model.layers.49.mlp.experts.75.gate_proj", "model.layers.49.mlp.experts.76.gate_proj", "model.layers.49.mlp.experts.77.gate_proj", "model.layers.49.mlp.experts.78.gate_proj", "model.layers.49.mlp.experts.79.gate_proj", "model.layers.49.mlp.experts.80.gate_proj", "model.layers.49.mlp.experts.81.gate_proj", "model.layers.49.mlp.experts.82.gate_proj", "model.layers.49.mlp.experts.83.gate_proj", "model.layers.49.mlp.experts.84.gate_proj", "model.layers.49.mlp.experts.85.gate_proj", "model.layers.49.mlp.experts.86.gate_proj", "model.layers.49.mlp.experts.87.gate_proj", "model.layers.49.mlp.experts.88.gate_proj", "model.layers.49.mlp.experts.89.gate_proj", "model.layers.49.mlp.experts.90.gate_proj", "model.layers.49.mlp.experts.91.gate_proj", "model.layers.49.mlp.experts.92.gate_proj", "model.layers.49.mlp.experts.93.gate_proj", "model.layers.49.mlp.experts.94.gate_proj", "model.layers.49.mlp.experts.95.gate_proj", "model.layers.49.mlp.experts.96.gate_proj", "model.layers.49.mlp.experts.97.gate_proj", "model.layers.49.mlp.experts.98.gate_proj", "model.layers.49.mlp.experts.99.gate_proj", "model.layers.49.mlp.experts.100.gate_proj", "model.layers.49.mlp.experts.101.gate_proj", "model.layers.49.mlp.experts.102.gate_proj", "model.layers.49.mlp.experts.103.gate_proj", "model.layers.49.mlp.experts.104.gate_proj", "model.layers.49.mlp.experts.105.gate_proj", "model.layers.49.mlp.experts.106.gate_proj", "model.layers.49.mlp.experts.107.gate_proj", "model.layers.49.mlp.experts.108.gate_proj", "model.layers.49.mlp.experts.109.gate_proj", "model.layers.49.mlp.experts.110.gate_proj", "model.layers.49.mlp.experts.111.gate_proj", "model.layers.49.mlp.experts.112.gate_proj", "model.layers.49.mlp.experts.113.gate_proj", "model.layers.49.mlp.experts.114.gate_proj", "model.layers.49.mlp.experts.115.gate_proj", "model.layers.49.mlp.experts.116.gate_proj", "model.layers.49.mlp.experts.117.gate_proj", "model.layers.49.mlp.experts.118.gate_proj", "model.layers.49.mlp.experts.119.gate_proj", "model.layers.49.mlp.experts.120.gate_proj", "model.layers.49.mlp.experts.121.gate_proj", "model.layers.49.mlp.experts.122.gate_proj", "model.layers.49.mlp.experts.123.gate_proj", "model.layers.49.mlp.experts.124.gate_proj", "model.layers.49.mlp.experts.125.gate_proj", "model.layers.49.mlp.experts.126.gate_proj", "model.layers.49.mlp.experts.127.gate_proj", "model.layers.49.mlp.experts.128.gate_proj", "model.layers.49.mlp.experts.129.gate_proj", "model.layers.49.mlp.experts.130.gate_proj", "model.layers.49.mlp.experts.131.gate_proj", "model.layers.49.mlp.experts.132.gate_proj", "model.layers.49.mlp.experts.133.gate_proj", "model.layers.49.mlp.experts.134.gate_proj", "model.layers.49.mlp.experts.135.gate_proj", "model.layers.49.mlp.experts.136.gate_proj", "model.layers.49.mlp.experts.137.gate_proj", "model.layers.49.mlp.experts.138.gate_proj", "model.layers.49.mlp.experts.139.gate_proj", "model.layers.49.mlp.experts.140.gate_proj", "model.layers.49.mlp.experts.141.gate_proj", "model.layers.49.mlp.experts.142.gate_proj", "model.layers.49.mlp.experts.143.gate_proj", "model.layers.49.mlp.experts.144.gate_proj", "model.layers.49.mlp.experts.145.gate_proj", "model.layers.49.mlp.experts.146.gate_proj", "model.layers.49.mlp.experts.147.gate_proj", "model.layers.49.mlp.experts.148.gate_proj", "model.layers.49.mlp.experts.149.gate_proj", "model.layers.49.mlp.experts.150.gate_proj", "model.layers.49.mlp.experts.151.gate_proj", "model.layers.49.mlp.experts.152.gate_proj", "model.layers.49.mlp.experts.153.gate_proj", "model.layers.49.mlp.experts.154.gate_proj", "model.layers.49.mlp.experts.155.gate_proj", "model.layers.49.mlp.experts.156.gate_proj", "model.layers.49.mlp.experts.157.gate_proj", "model.layers.49.mlp.experts.158.gate_proj", "model.layers.49.mlp.experts.159.gate_proj", "model.layers.49.mlp.experts.0.up_proj", "model.layers.49.mlp.experts.1.up_proj", "model.layers.49.mlp.experts.2.up_proj", "model.layers.49.mlp.experts.3.up_proj", "model.layers.49.mlp.experts.4.up_proj", "model.layers.49.mlp.experts.5.up_proj", "model.layers.49.mlp.experts.6.up_proj", "model.layers.49.mlp.experts.7.up_proj", "model.layers.49.mlp.experts.8.up_proj", "model.layers.49.mlp.experts.9.up_proj", "model.layers.49.mlp.experts.10.up_proj", "model.layers.49.mlp.experts.11.up_proj", "model.layers.49.mlp.experts.12.up_proj", "model.layers.49.mlp.experts.13.up_proj", "model.layers.49.mlp.experts.14.up_proj", "model.layers.49.mlp.experts.15.up_proj", "model.layers.49.mlp.experts.16.up_proj", "model.layers.49.mlp.experts.17.up_proj", "model.layers.49.mlp.experts.18.up_proj", "model.layers.49.mlp.experts.19.up_proj", "model.layers.49.mlp.experts.20.up_proj", "model.layers.49.mlp.experts.21.up_proj", "model.layers.49.mlp.experts.22.up_proj", "model.layers.49.mlp.experts.23.up_proj", "model.layers.49.mlp.experts.24.up_proj", "model.layers.49.mlp.experts.25.up_proj", "model.layers.49.mlp.experts.26.up_proj", "model.layers.49.mlp.experts.27.up_proj", "model.layers.49.mlp.experts.28.up_proj", "model.layers.49.mlp.experts.29.up_proj", "model.layers.49.mlp.experts.30.up_proj", "model.layers.49.mlp.experts.31.up_proj", "model.layers.49.mlp.experts.32.up_proj", "model.layers.49.mlp.experts.33.up_proj", "model.layers.49.mlp.experts.34.up_proj", "model.layers.49.mlp.experts.35.up_proj", "model.layers.49.mlp.experts.36.up_proj", "model.layers.49.mlp.experts.37.up_proj", "model.layers.49.mlp.experts.38.up_proj", "model.layers.49.mlp.experts.39.up_proj", "model.layers.49.mlp.experts.40.up_proj", "model.layers.49.mlp.experts.41.up_proj", "model.layers.49.mlp.experts.42.up_proj", "model.layers.49.mlp.experts.43.up_proj", "model.layers.49.mlp.experts.44.up_proj", "model.layers.49.mlp.experts.45.up_proj", "model.layers.49.mlp.experts.46.up_proj", "model.layers.49.mlp.experts.47.up_proj", "model.layers.49.mlp.experts.48.up_proj", "model.layers.49.mlp.experts.49.up_proj", "model.layers.49.mlp.experts.50.up_proj", "model.layers.49.mlp.experts.51.up_proj", "model.layers.49.mlp.experts.52.up_proj", "model.layers.49.mlp.experts.53.up_proj", "model.layers.49.mlp.experts.54.up_proj", "model.layers.49.mlp.experts.55.up_proj", "model.layers.49.mlp.experts.56.up_proj", "model.layers.49.mlp.experts.57.up_proj", "model.layers.49.mlp.experts.58.up_proj", "model.layers.49.mlp.experts.59.up_proj", "model.layers.49.mlp.experts.60.up_proj", "model.layers.49.mlp.experts.61.up_proj", "model.layers.49.mlp.experts.62.up_proj", "model.layers.49.mlp.experts.63.up_proj", "model.layers.49.mlp.experts.64.up_proj", "model.layers.49.mlp.experts.65.up_proj", "model.layers.49.mlp.experts.66.up_proj", "model.layers.49.mlp.experts.67.up_proj", "model.layers.49.mlp.experts.68.up_proj", "model.layers.49.mlp.experts.69.up_proj", "model.layers.49.mlp.experts.70.up_proj", "model.layers.49.mlp.experts.71.up_proj", "model.layers.49.mlp.experts.72.up_proj", "model.layers.49.mlp.experts.73.up_proj", "model.layers.49.mlp.experts.74.up_proj", "model.layers.49.mlp.experts.75.up_proj", "model.layers.49.mlp.experts.76.up_proj", "model.layers.49.mlp.experts.77.up_proj", "model.layers.49.mlp.experts.78.up_proj", "model.layers.49.mlp.experts.79.up_proj", "model.layers.49.mlp.experts.80.up_proj", "model.layers.49.mlp.experts.81.up_proj", "model.layers.49.mlp.experts.82.up_proj", "model.layers.49.mlp.experts.83.up_proj", "model.layers.49.mlp.experts.84.up_proj", "model.layers.49.mlp.experts.85.up_proj", "model.layers.49.mlp.experts.86.up_proj", "model.layers.49.mlp.experts.87.up_proj", "model.layers.49.mlp.experts.88.up_proj", "model.layers.49.mlp.experts.89.up_proj", "model.layers.49.mlp.experts.90.up_proj", "model.layers.49.mlp.experts.91.up_proj", "model.layers.49.mlp.experts.92.up_proj", "model.layers.49.mlp.experts.93.up_proj", "model.layers.49.mlp.experts.94.up_proj", "model.layers.49.mlp.experts.95.up_proj", "model.layers.49.mlp.experts.96.up_proj", "model.layers.49.mlp.experts.97.up_proj", "model.layers.49.mlp.experts.98.up_proj", "model.layers.49.mlp.experts.99.up_proj", "model.layers.49.mlp.experts.100.up_proj", "model.layers.49.mlp.experts.101.up_proj", "model.layers.49.mlp.experts.102.up_proj", "model.layers.49.mlp.experts.103.up_proj", "model.layers.49.mlp.experts.104.up_proj", "model.layers.49.mlp.experts.105.up_proj", "model.layers.49.mlp.experts.106.up_proj", "model.layers.49.mlp.experts.107.up_proj", "model.layers.49.mlp.experts.108.up_proj", "model.layers.49.mlp.experts.109.up_proj", "model.layers.49.mlp.experts.110.up_proj", "model.layers.49.mlp.experts.111.up_proj", "model.layers.49.mlp.experts.112.up_proj", "model.layers.49.mlp.experts.113.up_proj", "model.layers.49.mlp.experts.114.up_proj", "model.layers.49.mlp.experts.115.up_proj", "model.layers.49.mlp.experts.116.up_proj", "model.layers.49.mlp.experts.117.up_proj", "model.layers.49.mlp.experts.118.up_proj", "model.layers.49.mlp.experts.119.up_proj", "model.layers.49.mlp.experts.120.up_proj", "model.layers.49.mlp.experts.121.up_proj", "model.layers.49.mlp.experts.122.up_proj", "model.layers.49.mlp.experts.123.up_proj", "model.layers.49.mlp.experts.124.up_proj", "model.layers.49.mlp.experts.125.up_proj", "model.layers.49.mlp.experts.126.up_proj", "model.layers.49.mlp.experts.127.up_proj", "model.layers.49.mlp.experts.128.up_proj", "model.layers.49.mlp.experts.129.up_proj", "model.layers.49.mlp.experts.130.up_proj", "model.layers.49.mlp.experts.131.up_proj", "model.layers.49.mlp.experts.132.up_proj", "model.layers.49.mlp.experts.133.up_proj", "model.layers.49.mlp.experts.134.up_proj", "model.layers.49.mlp.experts.135.up_proj", "model.layers.49.mlp.experts.136.up_proj", "model.layers.49.mlp.experts.137.up_proj", "model.layers.49.mlp.experts.138.up_proj", "model.layers.49.mlp.experts.139.up_proj", "model.layers.49.mlp.experts.140.up_proj", "model.layers.49.mlp.experts.141.up_proj", "model.layers.49.mlp.experts.142.up_proj", "model.layers.49.mlp.experts.143.up_proj", "model.layers.49.mlp.experts.144.up_proj", "model.layers.49.mlp.experts.145.up_proj", "model.layers.49.mlp.experts.146.up_proj", "model.layers.49.mlp.experts.147.up_proj", "model.layers.49.mlp.experts.148.up_proj", "model.layers.49.mlp.experts.149.up_proj", "model.layers.49.mlp.experts.150.up_proj", "model.layers.49.mlp.experts.151.up_proj", "model.layers.49.mlp.experts.152.up_proj", "model.layers.49.mlp.experts.153.up_proj", "model.layers.49.mlp.experts.154.up_proj", "model.layers.49.mlp.experts.155.up_proj", "model.layers.49.mlp.experts.156.up_proj", "model.layers.49.mlp.experts.157.up_proj", "model.layers.49.mlp.experts.158.up_proj", "model.layers.49.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000311811640858653, "dbits": 2516582400 }, { "dkld": -0.0003333836793899536, "dbits": 5033164800 }, { "dkld": -0.00029933704063296596, "dbits": 7549747200 }, { "dkld": -0.0003596052527427701, "dbits": 12582912000 } ] }, { "idx": 296, "layers": [ "model.layers.49.mlp.experts.0.down_proj", "model.layers.49.mlp.experts.1.down_proj", "model.layers.49.mlp.experts.2.down_proj", "model.layers.49.mlp.experts.3.down_proj", "model.layers.49.mlp.experts.4.down_proj", "model.layers.49.mlp.experts.5.down_proj", "model.layers.49.mlp.experts.6.down_proj", "model.layers.49.mlp.experts.7.down_proj", "model.layers.49.mlp.experts.8.down_proj", "model.layers.49.mlp.experts.9.down_proj", "model.layers.49.mlp.experts.10.down_proj", "model.layers.49.mlp.experts.11.down_proj", "model.layers.49.mlp.experts.12.down_proj", "model.layers.49.mlp.experts.13.down_proj", "model.layers.49.mlp.experts.14.down_proj", "model.layers.49.mlp.experts.15.down_proj", "model.layers.49.mlp.experts.16.down_proj", "model.layers.49.mlp.experts.17.down_proj", "model.layers.49.mlp.experts.18.down_proj", "model.layers.49.mlp.experts.19.down_proj", "model.layers.49.mlp.experts.20.down_proj", "model.layers.49.mlp.experts.21.down_proj", "model.layers.49.mlp.experts.22.down_proj", "model.layers.49.mlp.experts.23.down_proj", "model.layers.49.mlp.experts.24.down_proj", "model.layers.49.mlp.experts.25.down_proj", "model.layers.49.mlp.experts.26.down_proj", "model.layers.49.mlp.experts.27.down_proj", "model.layers.49.mlp.experts.28.down_proj", "model.layers.49.mlp.experts.29.down_proj", "model.layers.49.mlp.experts.30.down_proj", "model.layers.49.mlp.experts.31.down_proj", "model.layers.49.mlp.experts.32.down_proj", "model.layers.49.mlp.experts.33.down_proj", "model.layers.49.mlp.experts.34.down_proj", "model.layers.49.mlp.experts.35.down_proj", "model.layers.49.mlp.experts.36.down_proj", "model.layers.49.mlp.experts.37.down_proj", "model.layers.49.mlp.experts.38.down_proj", "model.layers.49.mlp.experts.39.down_proj", "model.layers.49.mlp.experts.40.down_proj", "model.layers.49.mlp.experts.41.down_proj", "model.layers.49.mlp.experts.42.down_proj", "model.layers.49.mlp.experts.43.down_proj", "model.layers.49.mlp.experts.44.down_proj", "model.layers.49.mlp.experts.45.down_proj", "model.layers.49.mlp.experts.46.down_proj", "model.layers.49.mlp.experts.47.down_proj", "model.layers.49.mlp.experts.48.down_proj", "model.layers.49.mlp.experts.49.down_proj", "model.layers.49.mlp.experts.50.down_proj", "model.layers.49.mlp.experts.51.down_proj", "model.layers.49.mlp.experts.52.down_proj", "model.layers.49.mlp.experts.53.down_proj", "model.layers.49.mlp.experts.54.down_proj", "model.layers.49.mlp.experts.55.down_proj", "model.layers.49.mlp.experts.56.down_proj", "model.layers.49.mlp.experts.57.down_proj", "model.layers.49.mlp.experts.58.down_proj", "model.layers.49.mlp.experts.59.down_proj", "model.layers.49.mlp.experts.60.down_proj", "model.layers.49.mlp.experts.61.down_proj", "model.layers.49.mlp.experts.62.down_proj", "model.layers.49.mlp.experts.63.down_proj", "model.layers.49.mlp.experts.64.down_proj", "model.layers.49.mlp.experts.65.down_proj", "model.layers.49.mlp.experts.66.down_proj", "model.layers.49.mlp.experts.67.down_proj", "model.layers.49.mlp.experts.68.down_proj", "model.layers.49.mlp.experts.69.down_proj", "model.layers.49.mlp.experts.70.down_proj", "model.layers.49.mlp.experts.71.down_proj", "model.layers.49.mlp.experts.72.down_proj", "model.layers.49.mlp.experts.73.down_proj", "model.layers.49.mlp.experts.74.down_proj", "model.layers.49.mlp.experts.75.down_proj", "model.layers.49.mlp.experts.76.down_proj", "model.layers.49.mlp.experts.77.down_proj", "model.layers.49.mlp.experts.78.down_proj", "model.layers.49.mlp.experts.79.down_proj", "model.layers.49.mlp.experts.80.down_proj", "model.layers.49.mlp.experts.81.down_proj", "model.layers.49.mlp.experts.82.down_proj", "model.layers.49.mlp.experts.83.down_proj", "model.layers.49.mlp.experts.84.down_proj", "model.layers.49.mlp.experts.85.down_proj", "model.layers.49.mlp.experts.86.down_proj", "model.layers.49.mlp.experts.87.down_proj", "model.layers.49.mlp.experts.88.down_proj", "model.layers.49.mlp.experts.89.down_proj", "model.layers.49.mlp.experts.90.down_proj", "model.layers.49.mlp.experts.91.down_proj", "model.layers.49.mlp.experts.92.down_proj", "model.layers.49.mlp.experts.93.down_proj", "model.layers.49.mlp.experts.94.down_proj", "model.layers.49.mlp.experts.95.down_proj", "model.layers.49.mlp.experts.96.down_proj", "model.layers.49.mlp.experts.97.down_proj", "model.layers.49.mlp.experts.98.down_proj", "model.layers.49.mlp.experts.99.down_proj", "model.layers.49.mlp.experts.100.down_proj", "model.layers.49.mlp.experts.101.down_proj", "model.layers.49.mlp.experts.102.down_proj", "model.layers.49.mlp.experts.103.down_proj", "model.layers.49.mlp.experts.104.down_proj", "model.layers.49.mlp.experts.105.down_proj", "model.layers.49.mlp.experts.106.down_proj", "model.layers.49.mlp.experts.107.down_proj", "model.layers.49.mlp.experts.108.down_proj", "model.layers.49.mlp.experts.109.down_proj", "model.layers.49.mlp.experts.110.down_proj", "model.layers.49.mlp.experts.111.down_proj", "model.layers.49.mlp.experts.112.down_proj", "model.layers.49.mlp.experts.113.down_proj", "model.layers.49.mlp.experts.114.down_proj", "model.layers.49.mlp.experts.115.down_proj", "model.layers.49.mlp.experts.116.down_proj", "model.layers.49.mlp.experts.117.down_proj", "model.layers.49.mlp.experts.118.down_proj", "model.layers.49.mlp.experts.119.down_proj", "model.layers.49.mlp.experts.120.down_proj", "model.layers.49.mlp.experts.121.down_proj", "model.layers.49.mlp.experts.122.down_proj", "model.layers.49.mlp.experts.123.down_proj", "model.layers.49.mlp.experts.124.down_proj", "model.layers.49.mlp.experts.125.down_proj", "model.layers.49.mlp.experts.126.down_proj", "model.layers.49.mlp.experts.127.down_proj", "model.layers.49.mlp.experts.128.down_proj", "model.layers.49.mlp.experts.129.down_proj", "model.layers.49.mlp.experts.130.down_proj", "model.layers.49.mlp.experts.131.down_proj", "model.layers.49.mlp.experts.132.down_proj", "model.layers.49.mlp.experts.133.down_proj", "model.layers.49.mlp.experts.134.down_proj", "model.layers.49.mlp.experts.135.down_proj", "model.layers.49.mlp.experts.136.down_proj", "model.layers.49.mlp.experts.137.down_proj", "model.layers.49.mlp.experts.138.down_proj", "model.layers.49.mlp.experts.139.down_proj", "model.layers.49.mlp.experts.140.down_proj", "model.layers.49.mlp.experts.141.down_proj", "model.layers.49.mlp.experts.142.down_proj", "model.layers.49.mlp.experts.143.down_proj", "model.layers.49.mlp.experts.144.down_proj", "model.layers.49.mlp.experts.145.down_proj", "model.layers.49.mlp.experts.146.down_proj", "model.layers.49.mlp.experts.147.down_proj", "model.layers.49.mlp.experts.148.down_proj", "model.layers.49.mlp.experts.149.down_proj", "model.layers.49.mlp.experts.150.down_proj", "model.layers.49.mlp.experts.151.down_proj", "model.layers.49.mlp.experts.152.down_proj", "model.layers.49.mlp.experts.153.down_proj", "model.layers.49.mlp.experts.154.down_proj", "model.layers.49.mlp.experts.155.down_proj", "model.layers.49.mlp.experts.156.down_proj", "model.layers.49.mlp.experts.157.down_proj", "model.layers.49.mlp.experts.158.down_proj", "model.layers.49.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00019941609352827627, "dbits": 1258291200 }, { "dkld": -0.00020891372114420492, "dbits": 2516582400 }, { "dkld": -0.00021634092554451145, "dbits": 3774873600 }, { "dkld": -0.00019441833719612, "dbits": 6291456000 } ] }, { "idx": 297, "layers": [ "model.layers.50.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00012101419270038605, "dbits": 62914560 }, { "dkld": -4.562214016914923e-05, "dbits": 125829120 }, { "dkld": 1.1675618588918857e-05, "dbits": 188743680 }, { "dkld": -3.204382956029095e-05, "dbits": 314572800 } ] }, { "idx": 298, "layers": [ "model.layers.50.self_attn.k_proj", "model.layers.50.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00017510578036307456, "dbits": 10485760 }, { "dkld": -9.677540510893978e-05, "dbits": 20971520 }, { "dkld": -0.00011203726753593324, "dbits": 31457280 }, { "dkld": -0.00015326980501413623, "dbits": 52428800 } ] }, { "idx": 299, "layers": [ "model.layers.50.self_attn.o_proj" ], "candidates": [ { "dkld": -6.335871294141371e-05, "dbits": 62914560 }, { "dkld": -0.00020257281139493544, "dbits": 125829120 }, { "dkld": -0.0001929306425154237, "dbits": 188743680 }, { "dkld": -0.00020050266757608293, "dbits": 314572800 } ] }, { "idx": 300, "layers": [ "model.layers.50.mlp.shared_experts.gate_proj", "model.layers.50.mlp.shared_experts.up_proj", "model.layers.50.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -2.7759000658988953e-05, "dbits": 23592960 }, { "dkld": -0.00011147251352668086, "dbits": 47185920 }, { "dkld": -3.2794661819934845e-05, "dbits": 70778880 }, { "dkld": -2.497080713510791e-05, "dbits": 117964800 } ] }, { "idx": 301, "layers": [ "model.layers.50.mlp.experts.0.gate_proj", "model.layers.50.mlp.experts.1.gate_proj", "model.layers.50.mlp.experts.2.gate_proj", "model.layers.50.mlp.experts.3.gate_proj", "model.layers.50.mlp.experts.4.gate_proj", "model.layers.50.mlp.experts.5.gate_proj", "model.layers.50.mlp.experts.6.gate_proj", "model.layers.50.mlp.experts.7.gate_proj", "model.layers.50.mlp.experts.8.gate_proj", "model.layers.50.mlp.experts.9.gate_proj", "model.layers.50.mlp.experts.10.gate_proj", "model.layers.50.mlp.experts.11.gate_proj", "model.layers.50.mlp.experts.12.gate_proj", "model.layers.50.mlp.experts.13.gate_proj", "model.layers.50.mlp.experts.14.gate_proj", "model.layers.50.mlp.experts.15.gate_proj", "model.layers.50.mlp.experts.16.gate_proj", "model.layers.50.mlp.experts.17.gate_proj", "model.layers.50.mlp.experts.18.gate_proj", "model.layers.50.mlp.experts.19.gate_proj", "model.layers.50.mlp.experts.20.gate_proj", "model.layers.50.mlp.experts.21.gate_proj", "model.layers.50.mlp.experts.22.gate_proj", "model.layers.50.mlp.experts.23.gate_proj", "model.layers.50.mlp.experts.24.gate_proj", "model.layers.50.mlp.experts.25.gate_proj", "model.layers.50.mlp.experts.26.gate_proj", "model.layers.50.mlp.experts.27.gate_proj", "model.layers.50.mlp.experts.28.gate_proj", "model.layers.50.mlp.experts.29.gate_proj", "model.layers.50.mlp.experts.30.gate_proj", "model.layers.50.mlp.experts.31.gate_proj", "model.layers.50.mlp.experts.32.gate_proj", "model.layers.50.mlp.experts.33.gate_proj", "model.layers.50.mlp.experts.34.gate_proj", "model.layers.50.mlp.experts.35.gate_proj", "model.layers.50.mlp.experts.36.gate_proj", "model.layers.50.mlp.experts.37.gate_proj", "model.layers.50.mlp.experts.38.gate_proj", "model.layers.50.mlp.experts.39.gate_proj", "model.layers.50.mlp.experts.40.gate_proj", "model.layers.50.mlp.experts.41.gate_proj", "model.layers.50.mlp.experts.42.gate_proj", "model.layers.50.mlp.experts.43.gate_proj", "model.layers.50.mlp.experts.44.gate_proj", "model.layers.50.mlp.experts.45.gate_proj", "model.layers.50.mlp.experts.46.gate_proj", "model.layers.50.mlp.experts.47.gate_proj", "model.layers.50.mlp.experts.48.gate_proj", "model.layers.50.mlp.experts.49.gate_proj", "model.layers.50.mlp.experts.50.gate_proj", "model.layers.50.mlp.experts.51.gate_proj", "model.layers.50.mlp.experts.52.gate_proj", "model.layers.50.mlp.experts.53.gate_proj", "model.layers.50.mlp.experts.54.gate_proj", "model.layers.50.mlp.experts.55.gate_proj", "model.layers.50.mlp.experts.56.gate_proj", "model.layers.50.mlp.experts.57.gate_proj", "model.layers.50.mlp.experts.58.gate_proj", "model.layers.50.mlp.experts.59.gate_proj", "model.layers.50.mlp.experts.60.gate_proj", "model.layers.50.mlp.experts.61.gate_proj", "model.layers.50.mlp.experts.62.gate_proj", "model.layers.50.mlp.experts.63.gate_proj", "model.layers.50.mlp.experts.64.gate_proj", "model.layers.50.mlp.experts.65.gate_proj", "model.layers.50.mlp.experts.66.gate_proj", "model.layers.50.mlp.experts.67.gate_proj", "model.layers.50.mlp.experts.68.gate_proj", "model.layers.50.mlp.experts.69.gate_proj", "model.layers.50.mlp.experts.70.gate_proj", "model.layers.50.mlp.experts.71.gate_proj", "model.layers.50.mlp.experts.72.gate_proj", "model.layers.50.mlp.experts.73.gate_proj", "model.layers.50.mlp.experts.74.gate_proj", "model.layers.50.mlp.experts.75.gate_proj", "model.layers.50.mlp.experts.76.gate_proj", "model.layers.50.mlp.experts.77.gate_proj", "model.layers.50.mlp.experts.78.gate_proj", "model.layers.50.mlp.experts.79.gate_proj", "model.layers.50.mlp.experts.80.gate_proj", "model.layers.50.mlp.experts.81.gate_proj", "model.layers.50.mlp.experts.82.gate_proj", "model.layers.50.mlp.experts.83.gate_proj", "model.layers.50.mlp.experts.84.gate_proj", "model.layers.50.mlp.experts.85.gate_proj", "model.layers.50.mlp.experts.86.gate_proj", "model.layers.50.mlp.experts.87.gate_proj", "model.layers.50.mlp.experts.88.gate_proj", "model.layers.50.mlp.experts.89.gate_proj", "model.layers.50.mlp.experts.90.gate_proj", "model.layers.50.mlp.experts.91.gate_proj", "model.layers.50.mlp.experts.92.gate_proj", "model.layers.50.mlp.experts.93.gate_proj", "model.layers.50.mlp.experts.94.gate_proj", "model.layers.50.mlp.experts.95.gate_proj", "model.layers.50.mlp.experts.96.gate_proj", "model.layers.50.mlp.experts.97.gate_proj", "model.layers.50.mlp.experts.98.gate_proj", "model.layers.50.mlp.experts.99.gate_proj", "model.layers.50.mlp.experts.100.gate_proj", "model.layers.50.mlp.experts.101.gate_proj", "model.layers.50.mlp.experts.102.gate_proj", "model.layers.50.mlp.experts.103.gate_proj", "model.layers.50.mlp.experts.104.gate_proj", "model.layers.50.mlp.experts.105.gate_proj", "model.layers.50.mlp.experts.106.gate_proj", "model.layers.50.mlp.experts.107.gate_proj", "model.layers.50.mlp.experts.108.gate_proj", "model.layers.50.mlp.experts.109.gate_proj", "model.layers.50.mlp.experts.110.gate_proj", "model.layers.50.mlp.experts.111.gate_proj", "model.layers.50.mlp.experts.112.gate_proj", "model.layers.50.mlp.experts.113.gate_proj", "model.layers.50.mlp.experts.114.gate_proj", "model.layers.50.mlp.experts.115.gate_proj", "model.layers.50.mlp.experts.116.gate_proj", "model.layers.50.mlp.experts.117.gate_proj", "model.layers.50.mlp.experts.118.gate_proj", "model.layers.50.mlp.experts.119.gate_proj", "model.layers.50.mlp.experts.120.gate_proj", "model.layers.50.mlp.experts.121.gate_proj", "model.layers.50.mlp.experts.122.gate_proj", "model.layers.50.mlp.experts.123.gate_proj", "model.layers.50.mlp.experts.124.gate_proj", "model.layers.50.mlp.experts.125.gate_proj", "model.layers.50.mlp.experts.126.gate_proj", "model.layers.50.mlp.experts.127.gate_proj", "model.layers.50.mlp.experts.128.gate_proj", "model.layers.50.mlp.experts.129.gate_proj", "model.layers.50.mlp.experts.130.gate_proj", "model.layers.50.mlp.experts.131.gate_proj", "model.layers.50.mlp.experts.132.gate_proj", "model.layers.50.mlp.experts.133.gate_proj", "model.layers.50.mlp.experts.134.gate_proj", "model.layers.50.mlp.experts.135.gate_proj", "model.layers.50.mlp.experts.136.gate_proj", "model.layers.50.mlp.experts.137.gate_proj", "model.layers.50.mlp.experts.138.gate_proj", "model.layers.50.mlp.experts.139.gate_proj", "model.layers.50.mlp.experts.140.gate_proj", "model.layers.50.mlp.experts.141.gate_proj", "model.layers.50.mlp.experts.142.gate_proj", "model.layers.50.mlp.experts.143.gate_proj", "model.layers.50.mlp.experts.144.gate_proj", "model.layers.50.mlp.experts.145.gate_proj", "model.layers.50.mlp.experts.146.gate_proj", "model.layers.50.mlp.experts.147.gate_proj", "model.layers.50.mlp.experts.148.gate_proj", "model.layers.50.mlp.experts.149.gate_proj", "model.layers.50.mlp.experts.150.gate_proj", "model.layers.50.mlp.experts.151.gate_proj", "model.layers.50.mlp.experts.152.gate_proj", "model.layers.50.mlp.experts.153.gate_proj", "model.layers.50.mlp.experts.154.gate_proj", "model.layers.50.mlp.experts.155.gate_proj", "model.layers.50.mlp.experts.156.gate_proj", "model.layers.50.mlp.experts.157.gate_proj", "model.layers.50.mlp.experts.158.gate_proj", "model.layers.50.mlp.experts.159.gate_proj", "model.layers.50.mlp.experts.0.up_proj", "model.layers.50.mlp.experts.1.up_proj", "model.layers.50.mlp.experts.2.up_proj", "model.layers.50.mlp.experts.3.up_proj", "model.layers.50.mlp.experts.4.up_proj", "model.layers.50.mlp.experts.5.up_proj", "model.layers.50.mlp.experts.6.up_proj", "model.layers.50.mlp.experts.7.up_proj", "model.layers.50.mlp.experts.8.up_proj", "model.layers.50.mlp.experts.9.up_proj", "model.layers.50.mlp.experts.10.up_proj", "model.layers.50.mlp.experts.11.up_proj", "model.layers.50.mlp.experts.12.up_proj", "model.layers.50.mlp.experts.13.up_proj", "model.layers.50.mlp.experts.14.up_proj", "model.layers.50.mlp.experts.15.up_proj", "model.layers.50.mlp.experts.16.up_proj", "model.layers.50.mlp.experts.17.up_proj", "model.layers.50.mlp.experts.18.up_proj", "model.layers.50.mlp.experts.19.up_proj", "model.layers.50.mlp.experts.20.up_proj", "model.layers.50.mlp.experts.21.up_proj", "model.layers.50.mlp.experts.22.up_proj", "model.layers.50.mlp.experts.23.up_proj", "model.layers.50.mlp.experts.24.up_proj", "model.layers.50.mlp.experts.25.up_proj", "model.layers.50.mlp.experts.26.up_proj", "model.layers.50.mlp.experts.27.up_proj", "model.layers.50.mlp.experts.28.up_proj", "model.layers.50.mlp.experts.29.up_proj", "model.layers.50.mlp.experts.30.up_proj", "model.layers.50.mlp.experts.31.up_proj", "model.layers.50.mlp.experts.32.up_proj", "model.layers.50.mlp.experts.33.up_proj", "model.layers.50.mlp.experts.34.up_proj", "model.layers.50.mlp.experts.35.up_proj", "model.layers.50.mlp.experts.36.up_proj", "model.layers.50.mlp.experts.37.up_proj", "model.layers.50.mlp.experts.38.up_proj", "model.layers.50.mlp.experts.39.up_proj", "model.layers.50.mlp.experts.40.up_proj", "model.layers.50.mlp.experts.41.up_proj", "model.layers.50.mlp.experts.42.up_proj", "model.layers.50.mlp.experts.43.up_proj", "model.layers.50.mlp.experts.44.up_proj", "model.layers.50.mlp.experts.45.up_proj", "model.layers.50.mlp.experts.46.up_proj", "model.layers.50.mlp.experts.47.up_proj", "model.layers.50.mlp.experts.48.up_proj", "model.layers.50.mlp.experts.49.up_proj", "model.layers.50.mlp.experts.50.up_proj", "model.layers.50.mlp.experts.51.up_proj", "model.layers.50.mlp.experts.52.up_proj", "model.layers.50.mlp.experts.53.up_proj", "model.layers.50.mlp.experts.54.up_proj", "model.layers.50.mlp.experts.55.up_proj", "model.layers.50.mlp.experts.56.up_proj", "model.layers.50.mlp.experts.57.up_proj", "model.layers.50.mlp.experts.58.up_proj", "model.layers.50.mlp.experts.59.up_proj", "model.layers.50.mlp.experts.60.up_proj", "model.layers.50.mlp.experts.61.up_proj", "model.layers.50.mlp.experts.62.up_proj", "model.layers.50.mlp.experts.63.up_proj", "model.layers.50.mlp.experts.64.up_proj", "model.layers.50.mlp.experts.65.up_proj", "model.layers.50.mlp.experts.66.up_proj", "model.layers.50.mlp.experts.67.up_proj", "model.layers.50.mlp.experts.68.up_proj", "model.layers.50.mlp.experts.69.up_proj", "model.layers.50.mlp.experts.70.up_proj", "model.layers.50.mlp.experts.71.up_proj", "model.layers.50.mlp.experts.72.up_proj", "model.layers.50.mlp.experts.73.up_proj", "model.layers.50.mlp.experts.74.up_proj", "model.layers.50.mlp.experts.75.up_proj", "model.layers.50.mlp.experts.76.up_proj", "model.layers.50.mlp.experts.77.up_proj", "model.layers.50.mlp.experts.78.up_proj", "model.layers.50.mlp.experts.79.up_proj", "model.layers.50.mlp.experts.80.up_proj", "model.layers.50.mlp.experts.81.up_proj", "model.layers.50.mlp.experts.82.up_proj", "model.layers.50.mlp.experts.83.up_proj", "model.layers.50.mlp.experts.84.up_proj", "model.layers.50.mlp.experts.85.up_proj", "model.layers.50.mlp.experts.86.up_proj", "model.layers.50.mlp.experts.87.up_proj", "model.layers.50.mlp.experts.88.up_proj", "model.layers.50.mlp.experts.89.up_proj", "model.layers.50.mlp.experts.90.up_proj", "model.layers.50.mlp.experts.91.up_proj", "model.layers.50.mlp.experts.92.up_proj", "model.layers.50.mlp.experts.93.up_proj", "model.layers.50.mlp.experts.94.up_proj", "model.layers.50.mlp.experts.95.up_proj", "model.layers.50.mlp.experts.96.up_proj", "model.layers.50.mlp.experts.97.up_proj", "model.layers.50.mlp.experts.98.up_proj", "model.layers.50.mlp.experts.99.up_proj", "model.layers.50.mlp.experts.100.up_proj", "model.layers.50.mlp.experts.101.up_proj", "model.layers.50.mlp.experts.102.up_proj", "model.layers.50.mlp.experts.103.up_proj", "model.layers.50.mlp.experts.104.up_proj", "model.layers.50.mlp.experts.105.up_proj", "model.layers.50.mlp.experts.106.up_proj", "model.layers.50.mlp.experts.107.up_proj", "model.layers.50.mlp.experts.108.up_proj", "model.layers.50.mlp.experts.109.up_proj", "model.layers.50.mlp.experts.110.up_proj", "model.layers.50.mlp.experts.111.up_proj", "model.layers.50.mlp.experts.112.up_proj", "model.layers.50.mlp.experts.113.up_proj", "model.layers.50.mlp.experts.114.up_proj", "model.layers.50.mlp.experts.115.up_proj", "model.layers.50.mlp.experts.116.up_proj", "model.layers.50.mlp.experts.117.up_proj", "model.layers.50.mlp.experts.118.up_proj", "model.layers.50.mlp.experts.119.up_proj", "model.layers.50.mlp.experts.120.up_proj", "model.layers.50.mlp.experts.121.up_proj", "model.layers.50.mlp.experts.122.up_proj", "model.layers.50.mlp.experts.123.up_proj", "model.layers.50.mlp.experts.124.up_proj", "model.layers.50.mlp.experts.125.up_proj", "model.layers.50.mlp.experts.126.up_proj", "model.layers.50.mlp.experts.127.up_proj", "model.layers.50.mlp.experts.128.up_proj", "model.layers.50.mlp.experts.129.up_proj", "model.layers.50.mlp.experts.130.up_proj", "model.layers.50.mlp.experts.131.up_proj", "model.layers.50.mlp.experts.132.up_proj", "model.layers.50.mlp.experts.133.up_proj", "model.layers.50.mlp.experts.134.up_proj", "model.layers.50.mlp.experts.135.up_proj", "model.layers.50.mlp.experts.136.up_proj", "model.layers.50.mlp.experts.137.up_proj", "model.layers.50.mlp.experts.138.up_proj", "model.layers.50.mlp.experts.139.up_proj", "model.layers.50.mlp.experts.140.up_proj", "model.layers.50.mlp.experts.141.up_proj", "model.layers.50.mlp.experts.142.up_proj", "model.layers.50.mlp.experts.143.up_proj", "model.layers.50.mlp.experts.144.up_proj", "model.layers.50.mlp.experts.145.up_proj", "model.layers.50.mlp.experts.146.up_proj", "model.layers.50.mlp.experts.147.up_proj", "model.layers.50.mlp.experts.148.up_proj", "model.layers.50.mlp.experts.149.up_proj", "model.layers.50.mlp.experts.150.up_proj", "model.layers.50.mlp.experts.151.up_proj", "model.layers.50.mlp.experts.152.up_proj", "model.layers.50.mlp.experts.153.up_proj", "model.layers.50.mlp.experts.154.up_proj", "model.layers.50.mlp.experts.155.up_proj", "model.layers.50.mlp.experts.156.up_proj", "model.layers.50.mlp.experts.157.up_proj", "model.layers.50.mlp.experts.158.up_proj", "model.layers.50.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00025224182754755853, "dbits": 2516582400 }, { "dkld": -0.0002411151304841097, "dbits": 5033164800 }, { "dkld": -0.00029180459678174175, "dbits": 7549747200 }, { "dkld": -0.0003105135634541567, "dbits": 12582912000 } ] }, { "idx": 302, "layers": [ "model.layers.50.mlp.experts.0.down_proj", "model.layers.50.mlp.experts.1.down_proj", "model.layers.50.mlp.experts.2.down_proj", "model.layers.50.mlp.experts.3.down_proj", "model.layers.50.mlp.experts.4.down_proj", "model.layers.50.mlp.experts.5.down_proj", "model.layers.50.mlp.experts.6.down_proj", "model.layers.50.mlp.experts.7.down_proj", "model.layers.50.mlp.experts.8.down_proj", "model.layers.50.mlp.experts.9.down_proj", "model.layers.50.mlp.experts.10.down_proj", "model.layers.50.mlp.experts.11.down_proj", "model.layers.50.mlp.experts.12.down_proj", "model.layers.50.mlp.experts.13.down_proj", "model.layers.50.mlp.experts.14.down_proj", "model.layers.50.mlp.experts.15.down_proj", "model.layers.50.mlp.experts.16.down_proj", "model.layers.50.mlp.experts.17.down_proj", "model.layers.50.mlp.experts.18.down_proj", "model.layers.50.mlp.experts.19.down_proj", "model.layers.50.mlp.experts.20.down_proj", "model.layers.50.mlp.experts.21.down_proj", "model.layers.50.mlp.experts.22.down_proj", "model.layers.50.mlp.experts.23.down_proj", "model.layers.50.mlp.experts.24.down_proj", "model.layers.50.mlp.experts.25.down_proj", "model.layers.50.mlp.experts.26.down_proj", "model.layers.50.mlp.experts.27.down_proj", "model.layers.50.mlp.experts.28.down_proj", "model.layers.50.mlp.experts.29.down_proj", "model.layers.50.mlp.experts.30.down_proj", "model.layers.50.mlp.experts.31.down_proj", "model.layers.50.mlp.experts.32.down_proj", "model.layers.50.mlp.experts.33.down_proj", "model.layers.50.mlp.experts.34.down_proj", "model.layers.50.mlp.experts.35.down_proj", "model.layers.50.mlp.experts.36.down_proj", "model.layers.50.mlp.experts.37.down_proj", "model.layers.50.mlp.experts.38.down_proj", "model.layers.50.mlp.experts.39.down_proj", "model.layers.50.mlp.experts.40.down_proj", "model.layers.50.mlp.experts.41.down_proj", "model.layers.50.mlp.experts.42.down_proj", "model.layers.50.mlp.experts.43.down_proj", "model.layers.50.mlp.experts.44.down_proj", "model.layers.50.mlp.experts.45.down_proj", "model.layers.50.mlp.experts.46.down_proj", "model.layers.50.mlp.experts.47.down_proj", "model.layers.50.mlp.experts.48.down_proj", "model.layers.50.mlp.experts.49.down_proj", "model.layers.50.mlp.experts.50.down_proj", "model.layers.50.mlp.experts.51.down_proj", "model.layers.50.mlp.experts.52.down_proj", "model.layers.50.mlp.experts.53.down_proj", "model.layers.50.mlp.experts.54.down_proj", "model.layers.50.mlp.experts.55.down_proj", "model.layers.50.mlp.experts.56.down_proj", "model.layers.50.mlp.experts.57.down_proj", "model.layers.50.mlp.experts.58.down_proj", "model.layers.50.mlp.experts.59.down_proj", "model.layers.50.mlp.experts.60.down_proj", "model.layers.50.mlp.experts.61.down_proj", "model.layers.50.mlp.experts.62.down_proj", "model.layers.50.mlp.experts.63.down_proj", "model.layers.50.mlp.experts.64.down_proj", "model.layers.50.mlp.experts.65.down_proj", "model.layers.50.mlp.experts.66.down_proj", "model.layers.50.mlp.experts.67.down_proj", "model.layers.50.mlp.experts.68.down_proj", "model.layers.50.mlp.experts.69.down_proj", "model.layers.50.mlp.experts.70.down_proj", "model.layers.50.mlp.experts.71.down_proj", "model.layers.50.mlp.experts.72.down_proj", "model.layers.50.mlp.experts.73.down_proj", "model.layers.50.mlp.experts.74.down_proj", "model.layers.50.mlp.experts.75.down_proj", "model.layers.50.mlp.experts.76.down_proj", "model.layers.50.mlp.experts.77.down_proj", "model.layers.50.mlp.experts.78.down_proj", "model.layers.50.mlp.experts.79.down_proj", "model.layers.50.mlp.experts.80.down_proj", "model.layers.50.mlp.experts.81.down_proj", "model.layers.50.mlp.experts.82.down_proj", "model.layers.50.mlp.experts.83.down_proj", "model.layers.50.mlp.experts.84.down_proj", "model.layers.50.mlp.experts.85.down_proj", "model.layers.50.mlp.experts.86.down_proj", "model.layers.50.mlp.experts.87.down_proj", "model.layers.50.mlp.experts.88.down_proj", "model.layers.50.mlp.experts.89.down_proj", "model.layers.50.mlp.experts.90.down_proj", "model.layers.50.mlp.experts.91.down_proj", "model.layers.50.mlp.experts.92.down_proj", "model.layers.50.mlp.experts.93.down_proj", "model.layers.50.mlp.experts.94.down_proj", "model.layers.50.mlp.experts.95.down_proj", "model.layers.50.mlp.experts.96.down_proj", "model.layers.50.mlp.experts.97.down_proj", "model.layers.50.mlp.experts.98.down_proj", "model.layers.50.mlp.experts.99.down_proj", "model.layers.50.mlp.experts.100.down_proj", "model.layers.50.mlp.experts.101.down_proj", "model.layers.50.mlp.experts.102.down_proj", "model.layers.50.mlp.experts.103.down_proj", "model.layers.50.mlp.experts.104.down_proj", "model.layers.50.mlp.experts.105.down_proj", "model.layers.50.mlp.experts.106.down_proj", "model.layers.50.mlp.experts.107.down_proj", "model.layers.50.mlp.experts.108.down_proj", "model.layers.50.mlp.experts.109.down_proj", "model.layers.50.mlp.experts.110.down_proj", "model.layers.50.mlp.experts.111.down_proj", "model.layers.50.mlp.experts.112.down_proj", "model.layers.50.mlp.experts.113.down_proj", "model.layers.50.mlp.experts.114.down_proj", "model.layers.50.mlp.experts.115.down_proj", "model.layers.50.mlp.experts.116.down_proj", "model.layers.50.mlp.experts.117.down_proj", "model.layers.50.mlp.experts.118.down_proj", "model.layers.50.mlp.experts.119.down_proj", "model.layers.50.mlp.experts.120.down_proj", "model.layers.50.mlp.experts.121.down_proj", "model.layers.50.mlp.experts.122.down_proj", "model.layers.50.mlp.experts.123.down_proj", "model.layers.50.mlp.experts.124.down_proj", "model.layers.50.mlp.experts.125.down_proj", "model.layers.50.mlp.experts.126.down_proj", "model.layers.50.mlp.experts.127.down_proj", "model.layers.50.mlp.experts.128.down_proj", "model.layers.50.mlp.experts.129.down_proj", "model.layers.50.mlp.experts.130.down_proj", "model.layers.50.mlp.experts.131.down_proj", "model.layers.50.mlp.experts.132.down_proj", "model.layers.50.mlp.experts.133.down_proj", "model.layers.50.mlp.experts.134.down_proj", "model.layers.50.mlp.experts.135.down_proj", "model.layers.50.mlp.experts.136.down_proj", "model.layers.50.mlp.experts.137.down_proj", "model.layers.50.mlp.experts.138.down_proj", "model.layers.50.mlp.experts.139.down_proj", "model.layers.50.mlp.experts.140.down_proj", "model.layers.50.mlp.experts.141.down_proj", "model.layers.50.mlp.experts.142.down_proj", "model.layers.50.mlp.experts.143.down_proj", "model.layers.50.mlp.experts.144.down_proj", "model.layers.50.mlp.experts.145.down_proj", "model.layers.50.mlp.experts.146.down_proj", "model.layers.50.mlp.experts.147.down_proj", "model.layers.50.mlp.experts.148.down_proj", "model.layers.50.mlp.experts.149.down_proj", "model.layers.50.mlp.experts.150.down_proj", "model.layers.50.mlp.experts.151.down_proj", "model.layers.50.mlp.experts.152.down_proj", "model.layers.50.mlp.experts.153.down_proj", "model.layers.50.mlp.experts.154.down_proj", "model.layers.50.mlp.experts.155.down_proj", "model.layers.50.mlp.experts.156.down_proj", "model.layers.50.mlp.experts.157.down_proj", "model.layers.50.mlp.experts.158.down_proj", "model.layers.50.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0001597605645656669, "dbits": 1258291200 }, { "dkld": -0.0001602055504918154, "dbits": 2516582400 }, { "dkld": -0.00016897022724152166, "dbits": 3774873600 }, { "dkld": -0.00019561378285289088, "dbits": 6291456000 } ] }, { "idx": 303, "layers": [ "model.layers.51.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00013685096055270074, "dbits": 62914560 }, { "dkld": -0.00011848825961352505, "dbits": 125829120 }, { "dkld": -0.00013270089402794838, "dbits": 188743680 }, { "dkld": -0.00010460522025823593, "dbits": 314572800 } ] }, { "idx": 304, "layers": [ "model.layers.51.self_attn.k_proj", "model.layers.51.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0003290327265858678, "dbits": 10485760 }, { "dkld": -0.00033227391541004736, "dbits": 20971520 }, { "dkld": -0.00047439318150283016, "dbits": 31457280 }, { "dkld": -0.00033868588507175723, "dbits": 52428800 } ] }, { "idx": 305, "layers": [ "model.layers.51.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00015056468546391366, "dbits": 62914560 }, { "dkld": 2.534128725473206e-07, "dbits": 125829120 }, { "dkld": 3.8327276706690006e-05, "dbits": 188743680 }, { "dkld": -1.2653972953566295e-05, "dbits": 314572800 } ] }, { "idx": 306, "layers": [ "model.layers.51.mlp.shared_experts.gate_proj", "model.layers.51.mlp.shared_experts.up_proj", "model.layers.51.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00015510860830546258, "dbits": 23592960 }, { "dkld": -0.00031036036089063246, "dbits": 47185920 }, { "dkld": -0.0002384061925113229, "dbits": 70778880 }, { "dkld": -0.0002456204034388121, "dbits": 117964800 } ] }, { "idx": 307, "layers": [ "model.layers.51.mlp.experts.0.gate_proj", "model.layers.51.mlp.experts.1.gate_proj", "model.layers.51.mlp.experts.2.gate_proj", "model.layers.51.mlp.experts.3.gate_proj", "model.layers.51.mlp.experts.4.gate_proj", "model.layers.51.mlp.experts.5.gate_proj", "model.layers.51.mlp.experts.6.gate_proj", "model.layers.51.mlp.experts.7.gate_proj", "model.layers.51.mlp.experts.8.gate_proj", "model.layers.51.mlp.experts.9.gate_proj", "model.layers.51.mlp.experts.10.gate_proj", "model.layers.51.mlp.experts.11.gate_proj", "model.layers.51.mlp.experts.12.gate_proj", "model.layers.51.mlp.experts.13.gate_proj", "model.layers.51.mlp.experts.14.gate_proj", "model.layers.51.mlp.experts.15.gate_proj", "model.layers.51.mlp.experts.16.gate_proj", "model.layers.51.mlp.experts.17.gate_proj", "model.layers.51.mlp.experts.18.gate_proj", "model.layers.51.mlp.experts.19.gate_proj", "model.layers.51.mlp.experts.20.gate_proj", "model.layers.51.mlp.experts.21.gate_proj", "model.layers.51.mlp.experts.22.gate_proj", "model.layers.51.mlp.experts.23.gate_proj", "model.layers.51.mlp.experts.24.gate_proj", "model.layers.51.mlp.experts.25.gate_proj", "model.layers.51.mlp.experts.26.gate_proj", "model.layers.51.mlp.experts.27.gate_proj", "model.layers.51.mlp.experts.28.gate_proj", "model.layers.51.mlp.experts.29.gate_proj", "model.layers.51.mlp.experts.30.gate_proj", "model.layers.51.mlp.experts.31.gate_proj", "model.layers.51.mlp.experts.32.gate_proj", "model.layers.51.mlp.experts.33.gate_proj", "model.layers.51.mlp.experts.34.gate_proj", "model.layers.51.mlp.experts.35.gate_proj", "model.layers.51.mlp.experts.36.gate_proj", "model.layers.51.mlp.experts.37.gate_proj", "model.layers.51.mlp.experts.38.gate_proj", "model.layers.51.mlp.experts.39.gate_proj", "model.layers.51.mlp.experts.40.gate_proj", "model.layers.51.mlp.experts.41.gate_proj", "model.layers.51.mlp.experts.42.gate_proj", "model.layers.51.mlp.experts.43.gate_proj", "model.layers.51.mlp.experts.44.gate_proj", "model.layers.51.mlp.experts.45.gate_proj", "model.layers.51.mlp.experts.46.gate_proj", "model.layers.51.mlp.experts.47.gate_proj", "model.layers.51.mlp.experts.48.gate_proj", "model.layers.51.mlp.experts.49.gate_proj", "model.layers.51.mlp.experts.50.gate_proj", "model.layers.51.mlp.experts.51.gate_proj", "model.layers.51.mlp.experts.52.gate_proj", "model.layers.51.mlp.experts.53.gate_proj", "model.layers.51.mlp.experts.54.gate_proj", "model.layers.51.mlp.experts.55.gate_proj", "model.layers.51.mlp.experts.56.gate_proj", "model.layers.51.mlp.experts.57.gate_proj", "model.layers.51.mlp.experts.58.gate_proj", "model.layers.51.mlp.experts.59.gate_proj", "model.layers.51.mlp.experts.60.gate_proj", "model.layers.51.mlp.experts.61.gate_proj", "model.layers.51.mlp.experts.62.gate_proj", "model.layers.51.mlp.experts.63.gate_proj", "model.layers.51.mlp.experts.64.gate_proj", "model.layers.51.mlp.experts.65.gate_proj", "model.layers.51.mlp.experts.66.gate_proj", "model.layers.51.mlp.experts.67.gate_proj", "model.layers.51.mlp.experts.68.gate_proj", "model.layers.51.mlp.experts.69.gate_proj", "model.layers.51.mlp.experts.70.gate_proj", "model.layers.51.mlp.experts.71.gate_proj", "model.layers.51.mlp.experts.72.gate_proj", "model.layers.51.mlp.experts.73.gate_proj", "model.layers.51.mlp.experts.74.gate_proj", "model.layers.51.mlp.experts.75.gate_proj", "model.layers.51.mlp.experts.76.gate_proj", "model.layers.51.mlp.experts.77.gate_proj", "model.layers.51.mlp.experts.78.gate_proj", "model.layers.51.mlp.experts.79.gate_proj", "model.layers.51.mlp.experts.80.gate_proj", "model.layers.51.mlp.experts.81.gate_proj", "model.layers.51.mlp.experts.82.gate_proj", "model.layers.51.mlp.experts.83.gate_proj", "model.layers.51.mlp.experts.84.gate_proj", "model.layers.51.mlp.experts.85.gate_proj", "model.layers.51.mlp.experts.86.gate_proj", "model.layers.51.mlp.experts.87.gate_proj", "model.layers.51.mlp.experts.88.gate_proj", "model.layers.51.mlp.experts.89.gate_proj", "model.layers.51.mlp.experts.90.gate_proj", "model.layers.51.mlp.experts.91.gate_proj", "model.layers.51.mlp.experts.92.gate_proj", "model.layers.51.mlp.experts.93.gate_proj", "model.layers.51.mlp.experts.94.gate_proj", "model.layers.51.mlp.experts.95.gate_proj", "model.layers.51.mlp.experts.96.gate_proj", "model.layers.51.mlp.experts.97.gate_proj", "model.layers.51.mlp.experts.98.gate_proj", "model.layers.51.mlp.experts.99.gate_proj", "model.layers.51.mlp.experts.100.gate_proj", "model.layers.51.mlp.experts.101.gate_proj", "model.layers.51.mlp.experts.102.gate_proj", "model.layers.51.mlp.experts.103.gate_proj", "model.layers.51.mlp.experts.104.gate_proj", "model.layers.51.mlp.experts.105.gate_proj", "model.layers.51.mlp.experts.106.gate_proj", "model.layers.51.mlp.experts.107.gate_proj", "model.layers.51.mlp.experts.108.gate_proj", "model.layers.51.mlp.experts.109.gate_proj", "model.layers.51.mlp.experts.110.gate_proj", "model.layers.51.mlp.experts.111.gate_proj", "model.layers.51.mlp.experts.112.gate_proj", "model.layers.51.mlp.experts.113.gate_proj", "model.layers.51.mlp.experts.114.gate_proj", "model.layers.51.mlp.experts.115.gate_proj", "model.layers.51.mlp.experts.116.gate_proj", "model.layers.51.mlp.experts.117.gate_proj", "model.layers.51.mlp.experts.118.gate_proj", "model.layers.51.mlp.experts.119.gate_proj", "model.layers.51.mlp.experts.120.gate_proj", "model.layers.51.mlp.experts.121.gate_proj", "model.layers.51.mlp.experts.122.gate_proj", "model.layers.51.mlp.experts.123.gate_proj", "model.layers.51.mlp.experts.124.gate_proj", "model.layers.51.mlp.experts.125.gate_proj", "model.layers.51.mlp.experts.126.gate_proj", "model.layers.51.mlp.experts.127.gate_proj", "model.layers.51.mlp.experts.128.gate_proj", "model.layers.51.mlp.experts.129.gate_proj", "model.layers.51.mlp.experts.130.gate_proj", "model.layers.51.mlp.experts.131.gate_proj", "model.layers.51.mlp.experts.132.gate_proj", "model.layers.51.mlp.experts.133.gate_proj", "model.layers.51.mlp.experts.134.gate_proj", "model.layers.51.mlp.experts.135.gate_proj", "model.layers.51.mlp.experts.136.gate_proj", "model.layers.51.mlp.experts.137.gate_proj", "model.layers.51.mlp.experts.138.gate_proj", "model.layers.51.mlp.experts.139.gate_proj", "model.layers.51.mlp.experts.140.gate_proj", "model.layers.51.mlp.experts.141.gate_proj", "model.layers.51.mlp.experts.142.gate_proj", "model.layers.51.mlp.experts.143.gate_proj", "model.layers.51.mlp.experts.144.gate_proj", "model.layers.51.mlp.experts.145.gate_proj", "model.layers.51.mlp.experts.146.gate_proj", "model.layers.51.mlp.experts.147.gate_proj", "model.layers.51.mlp.experts.148.gate_proj", "model.layers.51.mlp.experts.149.gate_proj", "model.layers.51.mlp.experts.150.gate_proj", "model.layers.51.mlp.experts.151.gate_proj", "model.layers.51.mlp.experts.152.gate_proj", "model.layers.51.mlp.experts.153.gate_proj", "model.layers.51.mlp.experts.154.gate_proj", "model.layers.51.mlp.experts.155.gate_proj", "model.layers.51.mlp.experts.156.gate_proj", "model.layers.51.mlp.experts.157.gate_proj", "model.layers.51.mlp.experts.158.gate_proj", "model.layers.51.mlp.experts.159.gate_proj", "model.layers.51.mlp.experts.0.up_proj", "model.layers.51.mlp.experts.1.up_proj", "model.layers.51.mlp.experts.2.up_proj", "model.layers.51.mlp.experts.3.up_proj", "model.layers.51.mlp.experts.4.up_proj", "model.layers.51.mlp.experts.5.up_proj", "model.layers.51.mlp.experts.6.up_proj", "model.layers.51.mlp.experts.7.up_proj", "model.layers.51.mlp.experts.8.up_proj", "model.layers.51.mlp.experts.9.up_proj", "model.layers.51.mlp.experts.10.up_proj", "model.layers.51.mlp.experts.11.up_proj", "model.layers.51.mlp.experts.12.up_proj", "model.layers.51.mlp.experts.13.up_proj", "model.layers.51.mlp.experts.14.up_proj", "model.layers.51.mlp.experts.15.up_proj", "model.layers.51.mlp.experts.16.up_proj", "model.layers.51.mlp.experts.17.up_proj", "model.layers.51.mlp.experts.18.up_proj", "model.layers.51.mlp.experts.19.up_proj", "model.layers.51.mlp.experts.20.up_proj", "model.layers.51.mlp.experts.21.up_proj", "model.layers.51.mlp.experts.22.up_proj", "model.layers.51.mlp.experts.23.up_proj", "model.layers.51.mlp.experts.24.up_proj", "model.layers.51.mlp.experts.25.up_proj", "model.layers.51.mlp.experts.26.up_proj", "model.layers.51.mlp.experts.27.up_proj", "model.layers.51.mlp.experts.28.up_proj", "model.layers.51.mlp.experts.29.up_proj", "model.layers.51.mlp.experts.30.up_proj", "model.layers.51.mlp.experts.31.up_proj", "model.layers.51.mlp.experts.32.up_proj", "model.layers.51.mlp.experts.33.up_proj", "model.layers.51.mlp.experts.34.up_proj", "model.layers.51.mlp.experts.35.up_proj", "model.layers.51.mlp.experts.36.up_proj", "model.layers.51.mlp.experts.37.up_proj", "model.layers.51.mlp.experts.38.up_proj", "model.layers.51.mlp.experts.39.up_proj", "model.layers.51.mlp.experts.40.up_proj", "model.layers.51.mlp.experts.41.up_proj", "model.layers.51.mlp.experts.42.up_proj", "model.layers.51.mlp.experts.43.up_proj", "model.layers.51.mlp.experts.44.up_proj", "model.layers.51.mlp.experts.45.up_proj", "model.layers.51.mlp.experts.46.up_proj", "model.layers.51.mlp.experts.47.up_proj", "model.layers.51.mlp.experts.48.up_proj", "model.layers.51.mlp.experts.49.up_proj", "model.layers.51.mlp.experts.50.up_proj", "model.layers.51.mlp.experts.51.up_proj", "model.layers.51.mlp.experts.52.up_proj", "model.layers.51.mlp.experts.53.up_proj", "model.layers.51.mlp.experts.54.up_proj", "model.layers.51.mlp.experts.55.up_proj", "model.layers.51.mlp.experts.56.up_proj", "model.layers.51.mlp.experts.57.up_proj", "model.layers.51.mlp.experts.58.up_proj", "model.layers.51.mlp.experts.59.up_proj", "model.layers.51.mlp.experts.60.up_proj", "model.layers.51.mlp.experts.61.up_proj", "model.layers.51.mlp.experts.62.up_proj", "model.layers.51.mlp.experts.63.up_proj", "model.layers.51.mlp.experts.64.up_proj", "model.layers.51.mlp.experts.65.up_proj", "model.layers.51.mlp.experts.66.up_proj", "model.layers.51.mlp.experts.67.up_proj", "model.layers.51.mlp.experts.68.up_proj", "model.layers.51.mlp.experts.69.up_proj", "model.layers.51.mlp.experts.70.up_proj", "model.layers.51.mlp.experts.71.up_proj", "model.layers.51.mlp.experts.72.up_proj", "model.layers.51.mlp.experts.73.up_proj", "model.layers.51.mlp.experts.74.up_proj", "model.layers.51.mlp.experts.75.up_proj", "model.layers.51.mlp.experts.76.up_proj", "model.layers.51.mlp.experts.77.up_proj", "model.layers.51.mlp.experts.78.up_proj", "model.layers.51.mlp.experts.79.up_proj", "model.layers.51.mlp.experts.80.up_proj", "model.layers.51.mlp.experts.81.up_proj", "model.layers.51.mlp.experts.82.up_proj", "model.layers.51.mlp.experts.83.up_proj", "model.layers.51.mlp.experts.84.up_proj", "model.layers.51.mlp.experts.85.up_proj", "model.layers.51.mlp.experts.86.up_proj", "model.layers.51.mlp.experts.87.up_proj", "model.layers.51.mlp.experts.88.up_proj", "model.layers.51.mlp.experts.89.up_proj", "model.layers.51.mlp.experts.90.up_proj", "model.layers.51.mlp.experts.91.up_proj", "model.layers.51.mlp.experts.92.up_proj", "model.layers.51.mlp.experts.93.up_proj", "model.layers.51.mlp.experts.94.up_proj", "model.layers.51.mlp.experts.95.up_proj", "model.layers.51.mlp.experts.96.up_proj", "model.layers.51.mlp.experts.97.up_proj", "model.layers.51.mlp.experts.98.up_proj", "model.layers.51.mlp.experts.99.up_proj", "model.layers.51.mlp.experts.100.up_proj", "model.layers.51.mlp.experts.101.up_proj", "model.layers.51.mlp.experts.102.up_proj", "model.layers.51.mlp.experts.103.up_proj", "model.layers.51.mlp.experts.104.up_proj", "model.layers.51.mlp.experts.105.up_proj", "model.layers.51.mlp.experts.106.up_proj", "model.layers.51.mlp.experts.107.up_proj", "model.layers.51.mlp.experts.108.up_proj", "model.layers.51.mlp.experts.109.up_proj", "model.layers.51.mlp.experts.110.up_proj", "model.layers.51.mlp.experts.111.up_proj", "model.layers.51.mlp.experts.112.up_proj", "model.layers.51.mlp.experts.113.up_proj", "model.layers.51.mlp.experts.114.up_proj", "model.layers.51.mlp.experts.115.up_proj", "model.layers.51.mlp.experts.116.up_proj", "model.layers.51.mlp.experts.117.up_proj", "model.layers.51.mlp.experts.118.up_proj", "model.layers.51.mlp.experts.119.up_proj", "model.layers.51.mlp.experts.120.up_proj", "model.layers.51.mlp.experts.121.up_proj", "model.layers.51.mlp.experts.122.up_proj", "model.layers.51.mlp.experts.123.up_proj", "model.layers.51.mlp.experts.124.up_proj", "model.layers.51.mlp.experts.125.up_proj", "model.layers.51.mlp.experts.126.up_proj", "model.layers.51.mlp.experts.127.up_proj", "model.layers.51.mlp.experts.128.up_proj", "model.layers.51.mlp.experts.129.up_proj", "model.layers.51.mlp.experts.130.up_proj", "model.layers.51.mlp.experts.131.up_proj", "model.layers.51.mlp.experts.132.up_proj", "model.layers.51.mlp.experts.133.up_proj", "model.layers.51.mlp.experts.134.up_proj", "model.layers.51.mlp.experts.135.up_proj", "model.layers.51.mlp.experts.136.up_proj", "model.layers.51.mlp.experts.137.up_proj", "model.layers.51.mlp.experts.138.up_proj", "model.layers.51.mlp.experts.139.up_proj", "model.layers.51.mlp.experts.140.up_proj", "model.layers.51.mlp.experts.141.up_proj", "model.layers.51.mlp.experts.142.up_proj", "model.layers.51.mlp.experts.143.up_proj", "model.layers.51.mlp.experts.144.up_proj", "model.layers.51.mlp.experts.145.up_proj", "model.layers.51.mlp.experts.146.up_proj", "model.layers.51.mlp.experts.147.up_proj", "model.layers.51.mlp.experts.148.up_proj", "model.layers.51.mlp.experts.149.up_proj", "model.layers.51.mlp.experts.150.up_proj", "model.layers.51.mlp.experts.151.up_proj", "model.layers.51.mlp.experts.152.up_proj", "model.layers.51.mlp.experts.153.up_proj", "model.layers.51.mlp.experts.154.up_proj", "model.layers.51.mlp.experts.155.up_proj", "model.layers.51.mlp.experts.156.up_proj", "model.layers.51.mlp.experts.157.up_proj", "model.layers.51.mlp.experts.158.up_proj", "model.layers.51.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00024493867531419355, "dbits": 2516582400 }, { "dkld": -0.00026508867740632214, "dbits": 5033164800 }, { "dkld": -0.00029461514204741357, "dbits": 7549747200 }, { "dkld": -0.00029129236936570324, "dbits": 12582912000 } ] }, { "idx": 308, "layers": [ "model.layers.51.mlp.experts.0.down_proj", "model.layers.51.mlp.experts.1.down_proj", "model.layers.51.mlp.experts.2.down_proj", "model.layers.51.mlp.experts.3.down_proj", "model.layers.51.mlp.experts.4.down_proj", "model.layers.51.mlp.experts.5.down_proj", "model.layers.51.mlp.experts.6.down_proj", "model.layers.51.mlp.experts.7.down_proj", "model.layers.51.mlp.experts.8.down_proj", "model.layers.51.mlp.experts.9.down_proj", "model.layers.51.mlp.experts.10.down_proj", "model.layers.51.mlp.experts.11.down_proj", "model.layers.51.mlp.experts.12.down_proj", "model.layers.51.mlp.experts.13.down_proj", "model.layers.51.mlp.experts.14.down_proj", "model.layers.51.mlp.experts.15.down_proj", "model.layers.51.mlp.experts.16.down_proj", "model.layers.51.mlp.experts.17.down_proj", "model.layers.51.mlp.experts.18.down_proj", "model.layers.51.mlp.experts.19.down_proj", "model.layers.51.mlp.experts.20.down_proj", "model.layers.51.mlp.experts.21.down_proj", "model.layers.51.mlp.experts.22.down_proj", "model.layers.51.mlp.experts.23.down_proj", "model.layers.51.mlp.experts.24.down_proj", "model.layers.51.mlp.experts.25.down_proj", "model.layers.51.mlp.experts.26.down_proj", "model.layers.51.mlp.experts.27.down_proj", "model.layers.51.mlp.experts.28.down_proj", "model.layers.51.mlp.experts.29.down_proj", "model.layers.51.mlp.experts.30.down_proj", "model.layers.51.mlp.experts.31.down_proj", "model.layers.51.mlp.experts.32.down_proj", "model.layers.51.mlp.experts.33.down_proj", "model.layers.51.mlp.experts.34.down_proj", "model.layers.51.mlp.experts.35.down_proj", "model.layers.51.mlp.experts.36.down_proj", "model.layers.51.mlp.experts.37.down_proj", "model.layers.51.mlp.experts.38.down_proj", "model.layers.51.mlp.experts.39.down_proj", "model.layers.51.mlp.experts.40.down_proj", "model.layers.51.mlp.experts.41.down_proj", "model.layers.51.mlp.experts.42.down_proj", "model.layers.51.mlp.experts.43.down_proj", "model.layers.51.mlp.experts.44.down_proj", "model.layers.51.mlp.experts.45.down_proj", "model.layers.51.mlp.experts.46.down_proj", "model.layers.51.mlp.experts.47.down_proj", "model.layers.51.mlp.experts.48.down_proj", "model.layers.51.mlp.experts.49.down_proj", "model.layers.51.mlp.experts.50.down_proj", "model.layers.51.mlp.experts.51.down_proj", "model.layers.51.mlp.experts.52.down_proj", "model.layers.51.mlp.experts.53.down_proj", "model.layers.51.mlp.experts.54.down_proj", "model.layers.51.mlp.experts.55.down_proj", "model.layers.51.mlp.experts.56.down_proj", "model.layers.51.mlp.experts.57.down_proj", "model.layers.51.mlp.experts.58.down_proj", "model.layers.51.mlp.experts.59.down_proj", "model.layers.51.mlp.experts.60.down_proj", "model.layers.51.mlp.experts.61.down_proj", "model.layers.51.mlp.experts.62.down_proj", "model.layers.51.mlp.experts.63.down_proj", "model.layers.51.mlp.experts.64.down_proj", "model.layers.51.mlp.experts.65.down_proj", "model.layers.51.mlp.experts.66.down_proj", "model.layers.51.mlp.experts.67.down_proj", "model.layers.51.mlp.experts.68.down_proj", "model.layers.51.mlp.experts.69.down_proj", "model.layers.51.mlp.experts.70.down_proj", "model.layers.51.mlp.experts.71.down_proj", "model.layers.51.mlp.experts.72.down_proj", "model.layers.51.mlp.experts.73.down_proj", "model.layers.51.mlp.experts.74.down_proj", "model.layers.51.mlp.experts.75.down_proj", "model.layers.51.mlp.experts.76.down_proj", "model.layers.51.mlp.experts.77.down_proj", "model.layers.51.mlp.experts.78.down_proj", "model.layers.51.mlp.experts.79.down_proj", "model.layers.51.mlp.experts.80.down_proj", "model.layers.51.mlp.experts.81.down_proj", "model.layers.51.mlp.experts.82.down_proj", "model.layers.51.mlp.experts.83.down_proj", "model.layers.51.mlp.experts.84.down_proj", "model.layers.51.mlp.experts.85.down_proj", "model.layers.51.mlp.experts.86.down_proj", "model.layers.51.mlp.experts.87.down_proj", "model.layers.51.mlp.experts.88.down_proj", "model.layers.51.mlp.experts.89.down_proj", "model.layers.51.mlp.experts.90.down_proj", "model.layers.51.mlp.experts.91.down_proj", "model.layers.51.mlp.experts.92.down_proj", "model.layers.51.mlp.experts.93.down_proj", "model.layers.51.mlp.experts.94.down_proj", "model.layers.51.mlp.experts.95.down_proj", "model.layers.51.mlp.experts.96.down_proj", "model.layers.51.mlp.experts.97.down_proj", "model.layers.51.mlp.experts.98.down_proj", "model.layers.51.mlp.experts.99.down_proj", "model.layers.51.mlp.experts.100.down_proj", "model.layers.51.mlp.experts.101.down_proj", "model.layers.51.mlp.experts.102.down_proj", "model.layers.51.mlp.experts.103.down_proj", "model.layers.51.mlp.experts.104.down_proj", "model.layers.51.mlp.experts.105.down_proj", "model.layers.51.mlp.experts.106.down_proj", "model.layers.51.mlp.experts.107.down_proj", "model.layers.51.mlp.experts.108.down_proj", "model.layers.51.mlp.experts.109.down_proj", "model.layers.51.mlp.experts.110.down_proj", "model.layers.51.mlp.experts.111.down_proj", "model.layers.51.mlp.experts.112.down_proj", "model.layers.51.mlp.experts.113.down_proj", "model.layers.51.mlp.experts.114.down_proj", "model.layers.51.mlp.experts.115.down_proj", "model.layers.51.mlp.experts.116.down_proj", "model.layers.51.mlp.experts.117.down_proj", "model.layers.51.mlp.experts.118.down_proj", "model.layers.51.mlp.experts.119.down_proj", "model.layers.51.mlp.experts.120.down_proj", "model.layers.51.mlp.experts.121.down_proj", "model.layers.51.mlp.experts.122.down_proj", "model.layers.51.mlp.experts.123.down_proj", "model.layers.51.mlp.experts.124.down_proj", "model.layers.51.mlp.experts.125.down_proj", "model.layers.51.mlp.experts.126.down_proj", "model.layers.51.mlp.experts.127.down_proj", "model.layers.51.mlp.experts.128.down_proj", "model.layers.51.mlp.experts.129.down_proj", "model.layers.51.mlp.experts.130.down_proj", "model.layers.51.mlp.experts.131.down_proj", "model.layers.51.mlp.experts.132.down_proj", "model.layers.51.mlp.experts.133.down_proj", "model.layers.51.mlp.experts.134.down_proj", "model.layers.51.mlp.experts.135.down_proj", "model.layers.51.mlp.experts.136.down_proj", "model.layers.51.mlp.experts.137.down_proj", "model.layers.51.mlp.experts.138.down_proj", "model.layers.51.mlp.experts.139.down_proj", "model.layers.51.mlp.experts.140.down_proj", "model.layers.51.mlp.experts.141.down_proj", "model.layers.51.mlp.experts.142.down_proj", "model.layers.51.mlp.experts.143.down_proj", "model.layers.51.mlp.experts.144.down_proj", "model.layers.51.mlp.experts.145.down_proj", "model.layers.51.mlp.experts.146.down_proj", "model.layers.51.mlp.experts.147.down_proj", "model.layers.51.mlp.experts.148.down_proj", "model.layers.51.mlp.experts.149.down_proj", "model.layers.51.mlp.experts.150.down_proj", "model.layers.51.mlp.experts.151.down_proj", "model.layers.51.mlp.experts.152.down_proj", "model.layers.51.mlp.experts.153.down_proj", "model.layers.51.mlp.experts.154.down_proj", "model.layers.51.mlp.experts.155.down_proj", "model.layers.51.mlp.experts.156.down_proj", "model.layers.51.mlp.experts.157.down_proj", "model.layers.51.mlp.experts.158.down_proj", "model.layers.51.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002582391723990468, "dbits": 1258291200 }, { "dkld": -0.0002077866345644025, "dbits": 2516582400 }, { "dkld": -0.00021667797118425647, "dbits": 3774873600 }, { "dkld": -0.0002125414088368499, "dbits": 6291456000 } ] }, { "idx": 309, "layers": [ "model.layers.52.self_attn.q_proj" ], "candidates": [ { "dkld": 1.1325441300863437e-05, "dbits": 62914560 }, { "dkld": 0.00010482212528585555, "dbits": 125829120 }, { "dkld": 8.653784170746526e-05, "dbits": 188743680 }, { "dkld": 6.597163155674934e-05, "dbits": 314572800 } ] }, { "idx": 310, "layers": [ "model.layers.52.self_attn.k_proj", "model.layers.52.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007451565936207882, "dbits": 10485760 }, { "dkld": -0.0004891932010650663, "dbits": 20971520 }, { "dkld": -0.0006296861916780527, "dbits": 31457280 }, { "dkld": -0.000570616405457261, "dbits": 52428800 } ] }, { "idx": 311, "layers": [ "model.layers.52.self_attn.o_proj" ], "candidates": [ { "dkld": -8.949749171734134e-05, "dbits": 62914560 }, { "dkld": -8.491594344378628e-05, "dbits": 125829120 }, { "dkld": -0.0002518134191632354, "dbits": 188743680 }, { "dkld": -0.00022841878235341229, "dbits": 314572800 } ] }, { "idx": 312, "layers": [ "model.layers.52.mlp.shared_experts.gate_proj", "model.layers.52.mlp.shared_experts.up_proj", "model.layers.52.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00025985324755310735, "dbits": 23592960 }, { "dkld": 0.00021450361236929893, "dbits": 47185920 }, { "dkld": 0.00019294703379273137, "dbits": 70778880 }, { "dkld": 0.00021821781992911737, "dbits": 117964800 } ] }, { "idx": 313, "layers": [ "model.layers.52.mlp.experts.0.gate_proj", "model.layers.52.mlp.experts.1.gate_proj", "model.layers.52.mlp.experts.2.gate_proj", "model.layers.52.mlp.experts.3.gate_proj", "model.layers.52.mlp.experts.4.gate_proj", "model.layers.52.mlp.experts.5.gate_proj", "model.layers.52.mlp.experts.6.gate_proj", "model.layers.52.mlp.experts.7.gate_proj", "model.layers.52.mlp.experts.8.gate_proj", "model.layers.52.mlp.experts.9.gate_proj", "model.layers.52.mlp.experts.10.gate_proj", "model.layers.52.mlp.experts.11.gate_proj", "model.layers.52.mlp.experts.12.gate_proj", "model.layers.52.mlp.experts.13.gate_proj", "model.layers.52.mlp.experts.14.gate_proj", "model.layers.52.mlp.experts.15.gate_proj", "model.layers.52.mlp.experts.16.gate_proj", "model.layers.52.mlp.experts.17.gate_proj", "model.layers.52.mlp.experts.18.gate_proj", "model.layers.52.mlp.experts.19.gate_proj", "model.layers.52.mlp.experts.20.gate_proj", "model.layers.52.mlp.experts.21.gate_proj", "model.layers.52.mlp.experts.22.gate_proj", "model.layers.52.mlp.experts.23.gate_proj", "model.layers.52.mlp.experts.24.gate_proj", "model.layers.52.mlp.experts.25.gate_proj", "model.layers.52.mlp.experts.26.gate_proj", "model.layers.52.mlp.experts.27.gate_proj", "model.layers.52.mlp.experts.28.gate_proj", "model.layers.52.mlp.experts.29.gate_proj", "model.layers.52.mlp.experts.30.gate_proj", "model.layers.52.mlp.experts.31.gate_proj", "model.layers.52.mlp.experts.32.gate_proj", "model.layers.52.mlp.experts.33.gate_proj", "model.layers.52.mlp.experts.34.gate_proj", "model.layers.52.mlp.experts.35.gate_proj", "model.layers.52.mlp.experts.36.gate_proj", "model.layers.52.mlp.experts.37.gate_proj", "model.layers.52.mlp.experts.38.gate_proj", "model.layers.52.mlp.experts.39.gate_proj", "model.layers.52.mlp.experts.40.gate_proj", "model.layers.52.mlp.experts.41.gate_proj", "model.layers.52.mlp.experts.42.gate_proj", "model.layers.52.mlp.experts.43.gate_proj", "model.layers.52.mlp.experts.44.gate_proj", "model.layers.52.mlp.experts.45.gate_proj", "model.layers.52.mlp.experts.46.gate_proj", "model.layers.52.mlp.experts.47.gate_proj", "model.layers.52.mlp.experts.48.gate_proj", "model.layers.52.mlp.experts.49.gate_proj", "model.layers.52.mlp.experts.50.gate_proj", "model.layers.52.mlp.experts.51.gate_proj", "model.layers.52.mlp.experts.52.gate_proj", "model.layers.52.mlp.experts.53.gate_proj", "model.layers.52.mlp.experts.54.gate_proj", "model.layers.52.mlp.experts.55.gate_proj", "model.layers.52.mlp.experts.56.gate_proj", "model.layers.52.mlp.experts.57.gate_proj", "model.layers.52.mlp.experts.58.gate_proj", "model.layers.52.mlp.experts.59.gate_proj", "model.layers.52.mlp.experts.60.gate_proj", "model.layers.52.mlp.experts.61.gate_proj", "model.layers.52.mlp.experts.62.gate_proj", "model.layers.52.mlp.experts.63.gate_proj", "model.layers.52.mlp.experts.64.gate_proj", "model.layers.52.mlp.experts.65.gate_proj", "model.layers.52.mlp.experts.66.gate_proj", "model.layers.52.mlp.experts.67.gate_proj", "model.layers.52.mlp.experts.68.gate_proj", "model.layers.52.mlp.experts.69.gate_proj", "model.layers.52.mlp.experts.70.gate_proj", "model.layers.52.mlp.experts.71.gate_proj", "model.layers.52.mlp.experts.72.gate_proj", "model.layers.52.mlp.experts.73.gate_proj", "model.layers.52.mlp.experts.74.gate_proj", "model.layers.52.mlp.experts.75.gate_proj", "model.layers.52.mlp.experts.76.gate_proj", "model.layers.52.mlp.experts.77.gate_proj", "model.layers.52.mlp.experts.78.gate_proj", "model.layers.52.mlp.experts.79.gate_proj", "model.layers.52.mlp.experts.80.gate_proj", "model.layers.52.mlp.experts.81.gate_proj", "model.layers.52.mlp.experts.82.gate_proj", "model.layers.52.mlp.experts.83.gate_proj", "model.layers.52.mlp.experts.84.gate_proj", "model.layers.52.mlp.experts.85.gate_proj", "model.layers.52.mlp.experts.86.gate_proj", "model.layers.52.mlp.experts.87.gate_proj", "model.layers.52.mlp.experts.88.gate_proj", "model.layers.52.mlp.experts.89.gate_proj", "model.layers.52.mlp.experts.90.gate_proj", "model.layers.52.mlp.experts.91.gate_proj", "model.layers.52.mlp.experts.92.gate_proj", "model.layers.52.mlp.experts.93.gate_proj", "model.layers.52.mlp.experts.94.gate_proj", "model.layers.52.mlp.experts.95.gate_proj", "model.layers.52.mlp.experts.96.gate_proj", "model.layers.52.mlp.experts.97.gate_proj", "model.layers.52.mlp.experts.98.gate_proj", "model.layers.52.mlp.experts.99.gate_proj", "model.layers.52.mlp.experts.100.gate_proj", "model.layers.52.mlp.experts.101.gate_proj", "model.layers.52.mlp.experts.102.gate_proj", "model.layers.52.mlp.experts.103.gate_proj", "model.layers.52.mlp.experts.104.gate_proj", "model.layers.52.mlp.experts.105.gate_proj", "model.layers.52.mlp.experts.106.gate_proj", "model.layers.52.mlp.experts.107.gate_proj", "model.layers.52.mlp.experts.108.gate_proj", "model.layers.52.mlp.experts.109.gate_proj", "model.layers.52.mlp.experts.110.gate_proj", "model.layers.52.mlp.experts.111.gate_proj", "model.layers.52.mlp.experts.112.gate_proj", "model.layers.52.mlp.experts.113.gate_proj", "model.layers.52.mlp.experts.114.gate_proj", "model.layers.52.mlp.experts.115.gate_proj", "model.layers.52.mlp.experts.116.gate_proj", "model.layers.52.mlp.experts.117.gate_proj", "model.layers.52.mlp.experts.118.gate_proj", "model.layers.52.mlp.experts.119.gate_proj", "model.layers.52.mlp.experts.120.gate_proj", "model.layers.52.mlp.experts.121.gate_proj", "model.layers.52.mlp.experts.122.gate_proj", "model.layers.52.mlp.experts.123.gate_proj", "model.layers.52.mlp.experts.124.gate_proj", "model.layers.52.mlp.experts.125.gate_proj", "model.layers.52.mlp.experts.126.gate_proj", "model.layers.52.mlp.experts.127.gate_proj", "model.layers.52.mlp.experts.128.gate_proj", "model.layers.52.mlp.experts.129.gate_proj", "model.layers.52.mlp.experts.130.gate_proj", "model.layers.52.mlp.experts.131.gate_proj", "model.layers.52.mlp.experts.132.gate_proj", "model.layers.52.mlp.experts.133.gate_proj", "model.layers.52.mlp.experts.134.gate_proj", "model.layers.52.mlp.experts.135.gate_proj", "model.layers.52.mlp.experts.136.gate_proj", "model.layers.52.mlp.experts.137.gate_proj", "model.layers.52.mlp.experts.138.gate_proj", "model.layers.52.mlp.experts.139.gate_proj", "model.layers.52.mlp.experts.140.gate_proj", "model.layers.52.mlp.experts.141.gate_proj", "model.layers.52.mlp.experts.142.gate_proj", "model.layers.52.mlp.experts.143.gate_proj", "model.layers.52.mlp.experts.144.gate_proj", "model.layers.52.mlp.experts.145.gate_proj", "model.layers.52.mlp.experts.146.gate_proj", "model.layers.52.mlp.experts.147.gate_proj", "model.layers.52.mlp.experts.148.gate_proj", "model.layers.52.mlp.experts.149.gate_proj", "model.layers.52.mlp.experts.150.gate_proj", "model.layers.52.mlp.experts.151.gate_proj", "model.layers.52.mlp.experts.152.gate_proj", "model.layers.52.mlp.experts.153.gate_proj", "model.layers.52.mlp.experts.154.gate_proj", "model.layers.52.mlp.experts.155.gate_proj", "model.layers.52.mlp.experts.156.gate_proj", "model.layers.52.mlp.experts.157.gate_proj", "model.layers.52.mlp.experts.158.gate_proj", "model.layers.52.mlp.experts.159.gate_proj", "model.layers.52.mlp.experts.0.up_proj", "model.layers.52.mlp.experts.1.up_proj", "model.layers.52.mlp.experts.2.up_proj", "model.layers.52.mlp.experts.3.up_proj", "model.layers.52.mlp.experts.4.up_proj", "model.layers.52.mlp.experts.5.up_proj", "model.layers.52.mlp.experts.6.up_proj", "model.layers.52.mlp.experts.7.up_proj", "model.layers.52.mlp.experts.8.up_proj", "model.layers.52.mlp.experts.9.up_proj", "model.layers.52.mlp.experts.10.up_proj", "model.layers.52.mlp.experts.11.up_proj", "model.layers.52.mlp.experts.12.up_proj", "model.layers.52.mlp.experts.13.up_proj", "model.layers.52.mlp.experts.14.up_proj", "model.layers.52.mlp.experts.15.up_proj", "model.layers.52.mlp.experts.16.up_proj", "model.layers.52.mlp.experts.17.up_proj", "model.layers.52.mlp.experts.18.up_proj", "model.layers.52.mlp.experts.19.up_proj", "model.layers.52.mlp.experts.20.up_proj", "model.layers.52.mlp.experts.21.up_proj", "model.layers.52.mlp.experts.22.up_proj", "model.layers.52.mlp.experts.23.up_proj", "model.layers.52.mlp.experts.24.up_proj", "model.layers.52.mlp.experts.25.up_proj", "model.layers.52.mlp.experts.26.up_proj", "model.layers.52.mlp.experts.27.up_proj", "model.layers.52.mlp.experts.28.up_proj", "model.layers.52.mlp.experts.29.up_proj", "model.layers.52.mlp.experts.30.up_proj", "model.layers.52.mlp.experts.31.up_proj", "model.layers.52.mlp.experts.32.up_proj", "model.layers.52.mlp.experts.33.up_proj", "model.layers.52.mlp.experts.34.up_proj", "model.layers.52.mlp.experts.35.up_proj", "model.layers.52.mlp.experts.36.up_proj", "model.layers.52.mlp.experts.37.up_proj", "model.layers.52.mlp.experts.38.up_proj", "model.layers.52.mlp.experts.39.up_proj", "model.layers.52.mlp.experts.40.up_proj", "model.layers.52.mlp.experts.41.up_proj", "model.layers.52.mlp.experts.42.up_proj", "model.layers.52.mlp.experts.43.up_proj", "model.layers.52.mlp.experts.44.up_proj", "model.layers.52.mlp.experts.45.up_proj", "model.layers.52.mlp.experts.46.up_proj", "model.layers.52.mlp.experts.47.up_proj", "model.layers.52.mlp.experts.48.up_proj", "model.layers.52.mlp.experts.49.up_proj", "model.layers.52.mlp.experts.50.up_proj", "model.layers.52.mlp.experts.51.up_proj", "model.layers.52.mlp.experts.52.up_proj", "model.layers.52.mlp.experts.53.up_proj", "model.layers.52.mlp.experts.54.up_proj", "model.layers.52.mlp.experts.55.up_proj", "model.layers.52.mlp.experts.56.up_proj", "model.layers.52.mlp.experts.57.up_proj", "model.layers.52.mlp.experts.58.up_proj", "model.layers.52.mlp.experts.59.up_proj", "model.layers.52.mlp.experts.60.up_proj", "model.layers.52.mlp.experts.61.up_proj", "model.layers.52.mlp.experts.62.up_proj", "model.layers.52.mlp.experts.63.up_proj", "model.layers.52.mlp.experts.64.up_proj", "model.layers.52.mlp.experts.65.up_proj", "model.layers.52.mlp.experts.66.up_proj", "model.layers.52.mlp.experts.67.up_proj", "model.layers.52.mlp.experts.68.up_proj", "model.layers.52.mlp.experts.69.up_proj", "model.layers.52.mlp.experts.70.up_proj", "model.layers.52.mlp.experts.71.up_proj", "model.layers.52.mlp.experts.72.up_proj", "model.layers.52.mlp.experts.73.up_proj", "model.layers.52.mlp.experts.74.up_proj", "model.layers.52.mlp.experts.75.up_proj", "model.layers.52.mlp.experts.76.up_proj", "model.layers.52.mlp.experts.77.up_proj", "model.layers.52.mlp.experts.78.up_proj", "model.layers.52.mlp.experts.79.up_proj", "model.layers.52.mlp.experts.80.up_proj", "model.layers.52.mlp.experts.81.up_proj", "model.layers.52.mlp.experts.82.up_proj", "model.layers.52.mlp.experts.83.up_proj", "model.layers.52.mlp.experts.84.up_proj", "model.layers.52.mlp.experts.85.up_proj", "model.layers.52.mlp.experts.86.up_proj", "model.layers.52.mlp.experts.87.up_proj", "model.layers.52.mlp.experts.88.up_proj", "model.layers.52.mlp.experts.89.up_proj", "model.layers.52.mlp.experts.90.up_proj", "model.layers.52.mlp.experts.91.up_proj", "model.layers.52.mlp.experts.92.up_proj", "model.layers.52.mlp.experts.93.up_proj", "model.layers.52.mlp.experts.94.up_proj", "model.layers.52.mlp.experts.95.up_proj", "model.layers.52.mlp.experts.96.up_proj", "model.layers.52.mlp.experts.97.up_proj", "model.layers.52.mlp.experts.98.up_proj", "model.layers.52.mlp.experts.99.up_proj", "model.layers.52.mlp.experts.100.up_proj", "model.layers.52.mlp.experts.101.up_proj", "model.layers.52.mlp.experts.102.up_proj", "model.layers.52.mlp.experts.103.up_proj", "model.layers.52.mlp.experts.104.up_proj", "model.layers.52.mlp.experts.105.up_proj", "model.layers.52.mlp.experts.106.up_proj", "model.layers.52.mlp.experts.107.up_proj", "model.layers.52.mlp.experts.108.up_proj", "model.layers.52.mlp.experts.109.up_proj", "model.layers.52.mlp.experts.110.up_proj", "model.layers.52.mlp.experts.111.up_proj", "model.layers.52.mlp.experts.112.up_proj", "model.layers.52.mlp.experts.113.up_proj", "model.layers.52.mlp.experts.114.up_proj", "model.layers.52.mlp.experts.115.up_proj", "model.layers.52.mlp.experts.116.up_proj", "model.layers.52.mlp.experts.117.up_proj", "model.layers.52.mlp.experts.118.up_proj", "model.layers.52.mlp.experts.119.up_proj", "model.layers.52.mlp.experts.120.up_proj", "model.layers.52.mlp.experts.121.up_proj", "model.layers.52.mlp.experts.122.up_proj", "model.layers.52.mlp.experts.123.up_proj", "model.layers.52.mlp.experts.124.up_proj", "model.layers.52.mlp.experts.125.up_proj", "model.layers.52.mlp.experts.126.up_proj", "model.layers.52.mlp.experts.127.up_proj", "model.layers.52.mlp.experts.128.up_proj", "model.layers.52.mlp.experts.129.up_proj", "model.layers.52.mlp.experts.130.up_proj", "model.layers.52.mlp.experts.131.up_proj", "model.layers.52.mlp.experts.132.up_proj", "model.layers.52.mlp.experts.133.up_proj", "model.layers.52.mlp.experts.134.up_proj", "model.layers.52.mlp.experts.135.up_proj", "model.layers.52.mlp.experts.136.up_proj", "model.layers.52.mlp.experts.137.up_proj", "model.layers.52.mlp.experts.138.up_proj", "model.layers.52.mlp.experts.139.up_proj", "model.layers.52.mlp.experts.140.up_proj", "model.layers.52.mlp.experts.141.up_proj", "model.layers.52.mlp.experts.142.up_proj", "model.layers.52.mlp.experts.143.up_proj", "model.layers.52.mlp.experts.144.up_proj", "model.layers.52.mlp.experts.145.up_proj", "model.layers.52.mlp.experts.146.up_proj", "model.layers.52.mlp.experts.147.up_proj", "model.layers.52.mlp.experts.148.up_proj", "model.layers.52.mlp.experts.149.up_proj", "model.layers.52.mlp.experts.150.up_proj", "model.layers.52.mlp.experts.151.up_proj", "model.layers.52.mlp.experts.152.up_proj", "model.layers.52.mlp.experts.153.up_proj", "model.layers.52.mlp.experts.154.up_proj", "model.layers.52.mlp.experts.155.up_proj", "model.layers.52.mlp.experts.156.up_proj", "model.layers.52.mlp.experts.157.up_proj", "model.layers.52.mlp.experts.158.up_proj", "model.layers.52.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0001880816183984363, "dbits": 2516582400 }, { "dkld": -0.0002517359331250246, "dbits": 5033164800 }, { "dkld": -0.00019384771585465033, "dbits": 7549747200 }, { "dkld": -0.0002311928197741564, "dbits": 12582912000 } ] }, { "idx": 314, "layers": [ "model.layers.52.mlp.experts.0.down_proj", "model.layers.52.mlp.experts.1.down_proj", "model.layers.52.mlp.experts.2.down_proj", "model.layers.52.mlp.experts.3.down_proj", "model.layers.52.mlp.experts.4.down_proj", "model.layers.52.mlp.experts.5.down_proj", "model.layers.52.mlp.experts.6.down_proj", "model.layers.52.mlp.experts.7.down_proj", "model.layers.52.mlp.experts.8.down_proj", "model.layers.52.mlp.experts.9.down_proj", "model.layers.52.mlp.experts.10.down_proj", "model.layers.52.mlp.experts.11.down_proj", "model.layers.52.mlp.experts.12.down_proj", "model.layers.52.mlp.experts.13.down_proj", "model.layers.52.mlp.experts.14.down_proj", "model.layers.52.mlp.experts.15.down_proj", "model.layers.52.mlp.experts.16.down_proj", "model.layers.52.mlp.experts.17.down_proj", "model.layers.52.mlp.experts.18.down_proj", "model.layers.52.mlp.experts.19.down_proj", "model.layers.52.mlp.experts.20.down_proj", "model.layers.52.mlp.experts.21.down_proj", "model.layers.52.mlp.experts.22.down_proj", "model.layers.52.mlp.experts.23.down_proj", "model.layers.52.mlp.experts.24.down_proj", "model.layers.52.mlp.experts.25.down_proj", "model.layers.52.mlp.experts.26.down_proj", "model.layers.52.mlp.experts.27.down_proj", "model.layers.52.mlp.experts.28.down_proj", "model.layers.52.mlp.experts.29.down_proj", "model.layers.52.mlp.experts.30.down_proj", "model.layers.52.mlp.experts.31.down_proj", "model.layers.52.mlp.experts.32.down_proj", "model.layers.52.mlp.experts.33.down_proj", "model.layers.52.mlp.experts.34.down_proj", "model.layers.52.mlp.experts.35.down_proj", "model.layers.52.mlp.experts.36.down_proj", "model.layers.52.mlp.experts.37.down_proj", "model.layers.52.mlp.experts.38.down_proj", "model.layers.52.mlp.experts.39.down_proj", "model.layers.52.mlp.experts.40.down_proj", "model.layers.52.mlp.experts.41.down_proj", "model.layers.52.mlp.experts.42.down_proj", "model.layers.52.mlp.experts.43.down_proj", "model.layers.52.mlp.experts.44.down_proj", "model.layers.52.mlp.experts.45.down_proj", "model.layers.52.mlp.experts.46.down_proj", "model.layers.52.mlp.experts.47.down_proj", "model.layers.52.mlp.experts.48.down_proj", "model.layers.52.mlp.experts.49.down_proj", "model.layers.52.mlp.experts.50.down_proj", "model.layers.52.mlp.experts.51.down_proj", "model.layers.52.mlp.experts.52.down_proj", "model.layers.52.mlp.experts.53.down_proj", "model.layers.52.mlp.experts.54.down_proj", "model.layers.52.mlp.experts.55.down_proj", "model.layers.52.mlp.experts.56.down_proj", "model.layers.52.mlp.experts.57.down_proj", "model.layers.52.mlp.experts.58.down_proj", "model.layers.52.mlp.experts.59.down_proj", "model.layers.52.mlp.experts.60.down_proj", "model.layers.52.mlp.experts.61.down_proj", "model.layers.52.mlp.experts.62.down_proj", "model.layers.52.mlp.experts.63.down_proj", "model.layers.52.mlp.experts.64.down_proj", "model.layers.52.mlp.experts.65.down_proj", "model.layers.52.mlp.experts.66.down_proj", "model.layers.52.mlp.experts.67.down_proj", "model.layers.52.mlp.experts.68.down_proj", "model.layers.52.mlp.experts.69.down_proj", "model.layers.52.mlp.experts.70.down_proj", "model.layers.52.mlp.experts.71.down_proj", "model.layers.52.mlp.experts.72.down_proj", "model.layers.52.mlp.experts.73.down_proj", "model.layers.52.mlp.experts.74.down_proj", "model.layers.52.mlp.experts.75.down_proj", "model.layers.52.mlp.experts.76.down_proj", "model.layers.52.mlp.experts.77.down_proj", "model.layers.52.mlp.experts.78.down_proj", "model.layers.52.mlp.experts.79.down_proj", "model.layers.52.mlp.experts.80.down_proj", "model.layers.52.mlp.experts.81.down_proj", "model.layers.52.mlp.experts.82.down_proj", "model.layers.52.mlp.experts.83.down_proj", "model.layers.52.mlp.experts.84.down_proj", "model.layers.52.mlp.experts.85.down_proj", "model.layers.52.mlp.experts.86.down_proj", "model.layers.52.mlp.experts.87.down_proj", "model.layers.52.mlp.experts.88.down_proj", "model.layers.52.mlp.experts.89.down_proj", "model.layers.52.mlp.experts.90.down_proj", "model.layers.52.mlp.experts.91.down_proj", "model.layers.52.mlp.experts.92.down_proj", "model.layers.52.mlp.experts.93.down_proj", "model.layers.52.mlp.experts.94.down_proj", "model.layers.52.mlp.experts.95.down_proj", "model.layers.52.mlp.experts.96.down_proj", "model.layers.52.mlp.experts.97.down_proj", "model.layers.52.mlp.experts.98.down_proj", "model.layers.52.mlp.experts.99.down_proj", "model.layers.52.mlp.experts.100.down_proj", "model.layers.52.mlp.experts.101.down_proj", "model.layers.52.mlp.experts.102.down_proj", "model.layers.52.mlp.experts.103.down_proj", "model.layers.52.mlp.experts.104.down_proj", "model.layers.52.mlp.experts.105.down_proj", "model.layers.52.mlp.experts.106.down_proj", "model.layers.52.mlp.experts.107.down_proj", "model.layers.52.mlp.experts.108.down_proj", "model.layers.52.mlp.experts.109.down_proj", "model.layers.52.mlp.experts.110.down_proj", "model.layers.52.mlp.experts.111.down_proj", "model.layers.52.mlp.experts.112.down_proj", "model.layers.52.mlp.experts.113.down_proj", "model.layers.52.mlp.experts.114.down_proj", "model.layers.52.mlp.experts.115.down_proj", "model.layers.52.mlp.experts.116.down_proj", "model.layers.52.mlp.experts.117.down_proj", "model.layers.52.mlp.experts.118.down_proj", "model.layers.52.mlp.experts.119.down_proj", "model.layers.52.mlp.experts.120.down_proj", "model.layers.52.mlp.experts.121.down_proj", "model.layers.52.mlp.experts.122.down_proj", "model.layers.52.mlp.experts.123.down_proj", "model.layers.52.mlp.experts.124.down_proj", "model.layers.52.mlp.experts.125.down_proj", "model.layers.52.mlp.experts.126.down_proj", "model.layers.52.mlp.experts.127.down_proj", "model.layers.52.mlp.experts.128.down_proj", "model.layers.52.mlp.experts.129.down_proj", "model.layers.52.mlp.experts.130.down_proj", "model.layers.52.mlp.experts.131.down_proj", "model.layers.52.mlp.experts.132.down_proj", "model.layers.52.mlp.experts.133.down_proj", "model.layers.52.mlp.experts.134.down_proj", "model.layers.52.mlp.experts.135.down_proj", "model.layers.52.mlp.experts.136.down_proj", "model.layers.52.mlp.experts.137.down_proj", "model.layers.52.mlp.experts.138.down_proj", "model.layers.52.mlp.experts.139.down_proj", "model.layers.52.mlp.experts.140.down_proj", "model.layers.52.mlp.experts.141.down_proj", "model.layers.52.mlp.experts.142.down_proj", "model.layers.52.mlp.experts.143.down_proj", "model.layers.52.mlp.experts.144.down_proj", "model.layers.52.mlp.experts.145.down_proj", "model.layers.52.mlp.experts.146.down_proj", "model.layers.52.mlp.experts.147.down_proj", "model.layers.52.mlp.experts.148.down_proj", "model.layers.52.mlp.experts.149.down_proj", "model.layers.52.mlp.experts.150.down_proj", "model.layers.52.mlp.experts.151.down_proj", "model.layers.52.mlp.experts.152.down_proj", "model.layers.52.mlp.experts.153.down_proj", "model.layers.52.mlp.experts.154.down_proj", "model.layers.52.mlp.experts.155.down_proj", "model.layers.52.mlp.experts.156.down_proj", "model.layers.52.mlp.experts.157.down_proj", "model.layers.52.mlp.experts.158.down_proj", "model.layers.52.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00024020839482546963, "dbits": 1258291200 }, { "dkld": -0.00025058230385184566, "dbits": 2516582400 }, { "dkld": -0.00023125652223826565, "dbits": 3774873600 }, { "dkld": -0.00027366932481527606, "dbits": 6291456000 } ] }, { "idx": 315, "layers": [ "model.layers.53.self_attn.q_proj" ], "candidates": [ { "dkld": -6.51990063488539e-05, "dbits": 62914560 }, { "dkld": -0.00016577579081059057, "dbits": 125829120 }, { "dkld": -0.0001591598615050399, "dbits": 188743680 }, { "dkld": -0.0001839367672801101, "dbits": 314572800 } ] }, { "idx": 316, "layers": [ "model.layers.53.self_attn.k_proj", "model.layers.53.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0006882865913212383, "dbits": 10485760 }, { "dkld": -0.0006731220521032921, "dbits": 20971520 }, { "dkld": -0.0005834045819938238, "dbits": 31457280 }, { "dkld": -0.0006051247008144883, "dbits": 52428800 } ] }, { "idx": 317, "layers": [ "model.layers.53.self_attn.o_proj" ], "candidates": [ { "dkld": 2.624178305267455e-05, "dbits": 62914560 }, { "dkld": 2.4936441332099046e-05, "dbits": 125829120 }, { "dkld": -3.452943637967387e-05, "dbits": 188743680 }, { "dkld": -1.8590036779647656e-05, "dbits": 314572800 } ] }, { "idx": 318, "layers": [ "model.layers.53.mlp.shared_experts.gate_proj", "model.layers.53.mlp.shared_experts.up_proj", "model.layers.53.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0004713681526482133, "dbits": 23592960 }, { "dkld": -0.00029769446700811386, "dbits": 47185920 }, { "dkld": -0.00036124372854828835, "dbits": 70778880 }, { "dkld": -0.00031207334250211993, "dbits": 117964800 } ] }, { "idx": 319, "layers": [ "model.layers.53.mlp.experts.0.gate_proj", "model.layers.53.mlp.experts.1.gate_proj", "model.layers.53.mlp.experts.2.gate_proj", "model.layers.53.mlp.experts.3.gate_proj", "model.layers.53.mlp.experts.4.gate_proj", "model.layers.53.mlp.experts.5.gate_proj", "model.layers.53.mlp.experts.6.gate_proj", "model.layers.53.mlp.experts.7.gate_proj", "model.layers.53.mlp.experts.8.gate_proj", "model.layers.53.mlp.experts.9.gate_proj", "model.layers.53.mlp.experts.10.gate_proj", "model.layers.53.mlp.experts.11.gate_proj", "model.layers.53.mlp.experts.12.gate_proj", "model.layers.53.mlp.experts.13.gate_proj", "model.layers.53.mlp.experts.14.gate_proj", "model.layers.53.mlp.experts.15.gate_proj", "model.layers.53.mlp.experts.16.gate_proj", "model.layers.53.mlp.experts.17.gate_proj", "model.layers.53.mlp.experts.18.gate_proj", "model.layers.53.mlp.experts.19.gate_proj", "model.layers.53.mlp.experts.20.gate_proj", "model.layers.53.mlp.experts.21.gate_proj", "model.layers.53.mlp.experts.22.gate_proj", "model.layers.53.mlp.experts.23.gate_proj", "model.layers.53.mlp.experts.24.gate_proj", "model.layers.53.mlp.experts.25.gate_proj", "model.layers.53.mlp.experts.26.gate_proj", "model.layers.53.mlp.experts.27.gate_proj", "model.layers.53.mlp.experts.28.gate_proj", "model.layers.53.mlp.experts.29.gate_proj", "model.layers.53.mlp.experts.30.gate_proj", "model.layers.53.mlp.experts.31.gate_proj", "model.layers.53.mlp.experts.32.gate_proj", "model.layers.53.mlp.experts.33.gate_proj", "model.layers.53.mlp.experts.34.gate_proj", "model.layers.53.mlp.experts.35.gate_proj", "model.layers.53.mlp.experts.36.gate_proj", "model.layers.53.mlp.experts.37.gate_proj", "model.layers.53.mlp.experts.38.gate_proj", "model.layers.53.mlp.experts.39.gate_proj", "model.layers.53.mlp.experts.40.gate_proj", "model.layers.53.mlp.experts.41.gate_proj", "model.layers.53.mlp.experts.42.gate_proj", "model.layers.53.mlp.experts.43.gate_proj", "model.layers.53.mlp.experts.44.gate_proj", "model.layers.53.mlp.experts.45.gate_proj", "model.layers.53.mlp.experts.46.gate_proj", "model.layers.53.mlp.experts.47.gate_proj", "model.layers.53.mlp.experts.48.gate_proj", "model.layers.53.mlp.experts.49.gate_proj", "model.layers.53.mlp.experts.50.gate_proj", "model.layers.53.mlp.experts.51.gate_proj", "model.layers.53.mlp.experts.52.gate_proj", "model.layers.53.mlp.experts.53.gate_proj", "model.layers.53.mlp.experts.54.gate_proj", "model.layers.53.mlp.experts.55.gate_proj", "model.layers.53.mlp.experts.56.gate_proj", "model.layers.53.mlp.experts.57.gate_proj", "model.layers.53.mlp.experts.58.gate_proj", "model.layers.53.mlp.experts.59.gate_proj", "model.layers.53.mlp.experts.60.gate_proj", "model.layers.53.mlp.experts.61.gate_proj", "model.layers.53.mlp.experts.62.gate_proj", "model.layers.53.mlp.experts.63.gate_proj", "model.layers.53.mlp.experts.64.gate_proj", "model.layers.53.mlp.experts.65.gate_proj", "model.layers.53.mlp.experts.66.gate_proj", "model.layers.53.mlp.experts.67.gate_proj", "model.layers.53.mlp.experts.68.gate_proj", "model.layers.53.mlp.experts.69.gate_proj", "model.layers.53.mlp.experts.70.gate_proj", "model.layers.53.mlp.experts.71.gate_proj", "model.layers.53.mlp.experts.72.gate_proj", "model.layers.53.mlp.experts.73.gate_proj", "model.layers.53.mlp.experts.74.gate_proj", "model.layers.53.mlp.experts.75.gate_proj", "model.layers.53.mlp.experts.76.gate_proj", "model.layers.53.mlp.experts.77.gate_proj", "model.layers.53.mlp.experts.78.gate_proj", "model.layers.53.mlp.experts.79.gate_proj", "model.layers.53.mlp.experts.80.gate_proj", "model.layers.53.mlp.experts.81.gate_proj", "model.layers.53.mlp.experts.82.gate_proj", "model.layers.53.mlp.experts.83.gate_proj", "model.layers.53.mlp.experts.84.gate_proj", "model.layers.53.mlp.experts.85.gate_proj", "model.layers.53.mlp.experts.86.gate_proj", "model.layers.53.mlp.experts.87.gate_proj", "model.layers.53.mlp.experts.88.gate_proj", "model.layers.53.mlp.experts.89.gate_proj", "model.layers.53.mlp.experts.90.gate_proj", "model.layers.53.mlp.experts.91.gate_proj", "model.layers.53.mlp.experts.92.gate_proj", "model.layers.53.mlp.experts.93.gate_proj", "model.layers.53.mlp.experts.94.gate_proj", "model.layers.53.mlp.experts.95.gate_proj", "model.layers.53.mlp.experts.96.gate_proj", "model.layers.53.mlp.experts.97.gate_proj", "model.layers.53.mlp.experts.98.gate_proj", "model.layers.53.mlp.experts.99.gate_proj", "model.layers.53.mlp.experts.100.gate_proj", "model.layers.53.mlp.experts.101.gate_proj", "model.layers.53.mlp.experts.102.gate_proj", "model.layers.53.mlp.experts.103.gate_proj", "model.layers.53.mlp.experts.104.gate_proj", "model.layers.53.mlp.experts.105.gate_proj", "model.layers.53.mlp.experts.106.gate_proj", "model.layers.53.mlp.experts.107.gate_proj", "model.layers.53.mlp.experts.108.gate_proj", "model.layers.53.mlp.experts.109.gate_proj", "model.layers.53.mlp.experts.110.gate_proj", "model.layers.53.mlp.experts.111.gate_proj", "model.layers.53.mlp.experts.112.gate_proj", "model.layers.53.mlp.experts.113.gate_proj", "model.layers.53.mlp.experts.114.gate_proj", "model.layers.53.mlp.experts.115.gate_proj", "model.layers.53.mlp.experts.116.gate_proj", "model.layers.53.mlp.experts.117.gate_proj", "model.layers.53.mlp.experts.118.gate_proj", "model.layers.53.mlp.experts.119.gate_proj", "model.layers.53.mlp.experts.120.gate_proj", "model.layers.53.mlp.experts.121.gate_proj", "model.layers.53.mlp.experts.122.gate_proj", "model.layers.53.mlp.experts.123.gate_proj", "model.layers.53.mlp.experts.124.gate_proj", "model.layers.53.mlp.experts.125.gate_proj", "model.layers.53.mlp.experts.126.gate_proj", "model.layers.53.mlp.experts.127.gate_proj", "model.layers.53.mlp.experts.128.gate_proj", "model.layers.53.mlp.experts.129.gate_proj", "model.layers.53.mlp.experts.130.gate_proj", "model.layers.53.mlp.experts.131.gate_proj", "model.layers.53.mlp.experts.132.gate_proj", "model.layers.53.mlp.experts.133.gate_proj", "model.layers.53.mlp.experts.134.gate_proj", "model.layers.53.mlp.experts.135.gate_proj", "model.layers.53.mlp.experts.136.gate_proj", "model.layers.53.mlp.experts.137.gate_proj", "model.layers.53.mlp.experts.138.gate_proj", "model.layers.53.mlp.experts.139.gate_proj", "model.layers.53.mlp.experts.140.gate_proj", "model.layers.53.mlp.experts.141.gate_proj", "model.layers.53.mlp.experts.142.gate_proj", "model.layers.53.mlp.experts.143.gate_proj", "model.layers.53.mlp.experts.144.gate_proj", "model.layers.53.mlp.experts.145.gate_proj", "model.layers.53.mlp.experts.146.gate_proj", "model.layers.53.mlp.experts.147.gate_proj", "model.layers.53.mlp.experts.148.gate_proj", "model.layers.53.mlp.experts.149.gate_proj", "model.layers.53.mlp.experts.150.gate_proj", "model.layers.53.mlp.experts.151.gate_proj", "model.layers.53.mlp.experts.152.gate_proj", "model.layers.53.mlp.experts.153.gate_proj", "model.layers.53.mlp.experts.154.gate_proj", "model.layers.53.mlp.experts.155.gate_proj", "model.layers.53.mlp.experts.156.gate_proj", "model.layers.53.mlp.experts.157.gate_proj", "model.layers.53.mlp.experts.158.gate_proj", "model.layers.53.mlp.experts.159.gate_proj", "model.layers.53.mlp.experts.0.up_proj", "model.layers.53.mlp.experts.1.up_proj", "model.layers.53.mlp.experts.2.up_proj", "model.layers.53.mlp.experts.3.up_proj", "model.layers.53.mlp.experts.4.up_proj", "model.layers.53.mlp.experts.5.up_proj", "model.layers.53.mlp.experts.6.up_proj", "model.layers.53.mlp.experts.7.up_proj", "model.layers.53.mlp.experts.8.up_proj", "model.layers.53.mlp.experts.9.up_proj", "model.layers.53.mlp.experts.10.up_proj", "model.layers.53.mlp.experts.11.up_proj", "model.layers.53.mlp.experts.12.up_proj", "model.layers.53.mlp.experts.13.up_proj", "model.layers.53.mlp.experts.14.up_proj", "model.layers.53.mlp.experts.15.up_proj", "model.layers.53.mlp.experts.16.up_proj", "model.layers.53.mlp.experts.17.up_proj", "model.layers.53.mlp.experts.18.up_proj", "model.layers.53.mlp.experts.19.up_proj", "model.layers.53.mlp.experts.20.up_proj", "model.layers.53.mlp.experts.21.up_proj", "model.layers.53.mlp.experts.22.up_proj", "model.layers.53.mlp.experts.23.up_proj", "model.layers.53.mlp.experts.24.up_proj", "model.layers.53.mlp.experts.25.up_proj", "model.layers.53.mlp.experts.26.up_proj", "model.layers.53.mlp.experts.27.up_proj", "model.layers.53.mlp.experts.28.up_proj", "model.layers.53.mlp.experts.29.up_proj", "model.layers.53.mlp.experts.30.up_proj", "model.layers.53.mlp.experts.31.up_proj", "model.layers.53.mlp.experts.32.up_proj", "model.layers.53.mlp.experts.33.up_proj", "model.layers.53.mlp.experts.34.up_proj", "model.layers.53.mlp.experts.35.up_proj", "model.layers.53.mlp.experts.36.up_proj", "model.layers.53.mlp.experts.37.up_proj", "model.layers.53.mlp.experts.38.up_proj", "model.layers.53.mlp.experts.39.up_proj", "model.layers.53.mlp.experts.40.up_proj", "model.layers.53.mlp.experts.41.up_proj", "model.layers.53.mlp.experts.42.up_proj", "model.layers.53.mlp.experts.43.up_proj", "model.layers.53.mlp.experts.44.up_proj", "model.layers.53.mlp.experts.45.up_proj", "model.layers.53.mlp.experts.46.up_proj", "model.layers.53.mlp.experts.47.up_proj", "model.layers.53.mlp.experts.48.up_proj", "model.layers.53.mlp.experts.49.up_proj", "model.layers.53.mlp.experts.50.up_proj", "model.layers.53.mlp.experts.51.up_proj", "model.layers.53.mlp.experts.52.up_proj", "model.layers.53.mlp.experts.53.up_proj", "model.layers.53.mlp.experts.54.up_proj", "model.layers.53.mlp.experts.55.up_proj", "model.layers.53.mlp.experts.56.up_proj", "model.layers.53.mlp.experts.57.up_proj", "model.layers.53.mlp.experts.58.up_proj", "model.layers.53.mlp.experts.59.up_proj", "model.layers.53.mlp.experts.60.up_proj", "model.layers.53.mlp.experts.61.up_proj", "model.layers.53.mlp.experts.62.up_proj", "model.layers.53.mlp.experts.63.up_proj", "model.layers.53.mlp.experts.64.up_proj", "model.layers.53.mlp.experts.65.up_proj", "model.layers.53.mlp.experts.66.up_proj", "model.layers.53.mlp.experts.67.up_proj", "model.layers.53.mlp.experts.68.up_proj", "model.layers.53.mlp.experts.69.up_proj", "model.layers.53.mlp.experts.70.up_proj", "model.layers.53.mlp.experts.71.up_proj", "model.layers.53.mlp.experts.72.up_proj", "model.layers.53.mlp.experts.73.up_proj", "model.layers.53.mlp.experts.74.up_proj", "model.layers.53.mlp.experts.75.up_proj", "model.layers.53.mlp.experts.76.up_proj", "model.layers.53.mlp.experts.77.up_proj", "model.layers.53.mlp.experts.78.up_proj", "model.layers.53.mlp.experts.79.up_proj", "model.layers.53.mlp.experts.80.up_proj", "model.layers.53.mlp.experts.81.up_proj", "model.layers.53.mlp.experts.82.up_proj", "model.layers.53.mlp.experts.83.up_proj", "model.layers.53.mlp.experts.84.up_proj", "model.layers.53.mlp.experts.85.up_proj", "model.layers.53.mlp.experts.86.up_proj", "model.layers.53.mlp.experts.87.up_proj", "model.layers.53.mlp.experts.88.up_proj", "model.layers.53.mlp.experts.89.up_proj", "model.layers.53.mlp.experts.90.up_proj", "model.layers.53.mlp.experts.91.up_proj", "model.layers.53.mlp.experts.92.up_proj", "model.layers.53.mlp.experts.93.up_proj", "model.layers.53.mlp.experts.94.up_proj", "model.layers.53.mlp.experts.95.up_proj", "model.layers.53.mlp.experts.96.up_proj", "model.layers.53.mlp.experts.97.up_proj", "model.layers.53.mlp.experts.98.up_proj", "model.layers.53.mlp.experts.99.up_proj", "model.layers.53.mlp.experts.100.up_proj", "model.layers.53.mlp.experts.101.up_proj", "model.layers.53.mlp.experts.102.up_proj", "model.layers.53.mlp.experts.103.up_proj", "model.layers.53.mlp.experts.104.up_proj", "model.layers.53.mlp.experts.105.up_proj", "model.layers.53.mlp.experts.106.up_proj", "model.layers.53.mlp.experts.107.up_proj", "model.layers.53.mlp.experts.108.up_proj", "model.layers.53.mlp.experts.109.up_proj", "model.layers.53.mlp.experts.110.up_proj", "model.layers.53.mlp.experts.111.up_proj", "model.layers.53.mlp.experts.112.up_proj", "model.layers.53.mlp.experts.113.up_proj", "model.layers.53.mlp.experts.114.up_proj", "model.layers.53.mlp.experts.115.up_proj", "model.layers.53.mlp.experts.116.up_proj", "model.layers.53.mlp.experts.117.up_proj", "model.layers.53.mlp.experts.118.up_proj", "model.layers.53.mlp.experts.119.up_proj", "model.layers.53.mlp.experts.120.up_proj", "model.layers.53.mlp.experts.121.up_proj", "model.layers.53.mlp.experts.122.up_proj", "model.layers.53.mlp.experts.123.up_proj", "model.layers.53.mlp.experts.124.up_proj", "model.layers.53.mlp.experts.125.up_proj", "model.layers.53.mlp.experts.126.up_proj", "model.layers.53.mlp.experts.127.up_proj", "model.layers.53.mlp.experts.128.up_proj", "model.layers.53.mlp.experts.129.up_proj", "model.layers.53.mlp.experts.130.up_proj", "model.layers.53.mlp.experts.131.up_proj", "model.layers.53.mlp.experts.132.up_proj", "model.layers.53.mlp.experts.133.up_proj", "model.layers.53.mlp.experts.134.up_proj", "model.layers.53.mlp.experts.135.up_proj", "model.layers.53.mlp.experts.136.up_proj", "model.layers.53.mlp.experts.137.up_proj", "model.layers.53.mlp.experts.138.up_proj", "model.layers.53.mlp.experts.139.up_proj", "model.layers.53.mlp.experts.140.up_proj", "model.layers.53.mlp.experts.141.up_proj", "model.layers.53.mlp.experts.142.up_proj", "model.layers.53.mlp.experts.143.up_proj", "model.layers.53.mlp.experts.144.up_proj", "model.layers.53.mlp.experts.145.up_proj", "model.layers.53.mlp.experts.146.up_proj", "model.layers.53.mlp.experts.147.up_proj", "model.layers.53.mlp.experts.148.up_proj", "model.layers.53.mlp.experts.149.up_proj", "model.layers.53.mlp.experts.150.up_proj", "model.layers.53.mlp.experts.151.up_proj", "model.layers.53.mlp.experts.152.up_proj", "model.layers.53.mlp.experts.153.up_proj", "model.layers.53.mlp.experts.154.up_proj", "model.layers.53.mlp.experts.155.up_proj", "model.layers.53.mlp.experts.156.up_proj", "model.layers.53.mlp.experts.157.up_proj", "model.layers.53.mlp.experts.158.up_proj", "model.layers.53.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00013007549569010735, "dbits": 2516582400 }, { "dkld": -0.00020583318546414653, "dbits": 5033164800 }, { "dkld": -0.00023017153143883862, "dbits": 7549747200 }, { "dkld": -0.00024143634364010014, "dbits": 12582912000 } ] }, { "idx": 320, "layers": [ "model.layers.53.mlp.experts.0.down_proj", "model.layers.53.mlp.experts.1.down_proj", "model.layers.53.mlp.experts.2.down_proj", "model.layers.53.mlp.experts.3.down_proj", "model.layers.53.mlp.experts.4.down_proj", "model.layers.53.mlp.experts.5.down_proj", "model.layers.53.mlp.experts.6.down_proj", "model.layers.53.mlp.experts.7.down_proj", "model.layers.53.mlp.experts.8.down_proj", "model.layers.53.mlp.experts.9.down_proj", "model.layers.53.mlp.experts.10.down_proj", "model.layers.53.mlp.experts.11.down_proj", "model.layers.53.mlp.experts.12.down_proj", "model.layers.53.mlp.experts.13.down_proj", "model.layers.53.mlp.experts.14.down_proj", "model.layers.53.mlp.experts.15.down_proj", "model.layers.53.mlp.experts.16.down_proj", "model.layers.53.mlp.experts.17.down_proj", "model.layers.53.mlp.experts.18.down_proj", "model.layers.53.mlp.experts.19.down_proj", "model.layers.53.mlp.experts.20.down_proj", "model.layers.53.mlp.experts.21.down_proj", "model.layers.53.mlp.experts.22.down_proj", "model.layers.53.mlp.experts.23.down_proj", "model.layers.53.mlp.experts.24.down_proj", "model.layers.53.mlp.experts.25.down_proj", "model.layers.53.mlp.experts.26.down_proj", "model.layers.53.mlp.experts.27.down_proj", "model.layers.53.mlp.experts.28.down_proj", "model.layers.53.mlp.experts.29.down_proj", "model.layers.53.mlp.experts.30.down_proj", "model.layers.53.mlp.experts.31.down_proj", "model.layers.53.mlp.experts.32.down_proj", "model.layers.53.mlp.experts.33.down_proj", "model.layers.53.mlp.experts.34.down_proj", "model.layers.53.mlp.experts.35.down_proj", "model.layers.53.mlp.experts.36.down_proj", "model.layers.53.mlp.experts.37.down_proj", "model.layers.53.mlp.experts.38.down_proj", "model.layers.53.mlp.experts.39.down_proj", "model.layers.53.mlp.experts.40.down_proj", "model.layers.53.mlp.experts.41.down_proj", "model.layers.53.mlp.experts.42.down_proj", "model.layers.53.mlp.experts.43.down_proj", "model.layers.53.mlp.experts.44.down_proj", "model.layers.53.mlp.experts.45.down_proj", "model.layers.53.mlp.experts.46.down_proj", "model.layers.53.mlp.experts.47.down_proj", "model.layers.53.mlp.experts.48.down_proj", "model.layers.53.mlp.experts.49.down_proj", "model.layers.53.mlp.experts.50.down_proj", "model.layers.53.mlp.experts.51.down_proj", "model.layers.53.mlp.experts.52.down_proj", "model.layers.53.mlp.experts.53.down_proj", "model.layers.53.mlp.experts.54.down_proj", "model.layers.53.mlp.experts.55.down_proj", "model.layers.53.mlp.experts.56.down_proj", "model.layers.53.mlp.experts.57.down_proj", "model.layers.53.mlp.experts.58.down_proj", "model.layers.53.mlp.experts.59.down_proj", "model.layers.53.mlp.experts.60.down_proj", "model.layers.53.mlp.experts.61.down_proj", "model.layers.53.mlp.experts.62.down_proj", "model.layers.53.mlp.experts.63.down_proj", "model.layers.53.mlp.experts.64.down_proj", "model.layers.53.mlp.experts.65.down_proj", "model.layers.53.mlp.experts.66.down_proj", "model.layers.53.mlp.experts.67.down_proj", "model.layers.53.mlp.experts.68.down_proj", "model.layers.53.mlp.experts.69.down_proj", "model.layers.53.mlp.experts.70.down_proj", "model.layers.53.mlp.experts.71.down_proj", "model.layers.53.mlp.experts.72.down_proj", "model.layers.53.mlp.experts.73.down_proj", "model.layers.53.mlp.experts.74.down_proj", "model.layers.53.mlp.experts.75.down_proj", "model.layers.53.mlp.experts.76.down_proj", "model.layers.53.mlp.experts.77.down_proj", "model.layers.53.mlp.experts.78.down_proj", "model.layers.53.mlp.experts.79.down_proj", "model.layers.53.mlp.experts.80.down_proj", "model.layers.53.mlp.experts.81.down_proj", "model.layers.53.mlp.experts.82.down_proj", "model.layers.53.mlp.experts.83.down_proj", "model.layers.53.mlp.experts.84.down_proj", "model.layers.53.mlp.experts.85.down_proj", "model.layers.53.mlp.experts.86.down_proj", "model.layers.53.mlp.experts.87.down_proj", "model.layers.53.mlp.experts.88.down_proj", "model.layers.53.mlp.experts.89.down_proj", "model.layers.53.mlp.experts.90.down_proj", "model.layers.53.mlp.experts.91.down_proj", "model.layers.53.mlp.experts.92.down_proj", "model.layers.53.mlp.experts.93.down_proj", "model.layers.53.mlp.experts.94.down_proj", "model.layers.53.mlp.experts.95.down_proj", "model.layers.53.mlp.experts.96.down_proj", "model.layers.53.mlp.experts.97.down_proj", "model.layers.53.mlp.experts.98.down_proj", "model.layers.53.mlp.experts.99.down_proj", "model.layers.53.mlp.experts.100.down_proj", "model.layers.53.mlp.experts.101.down_proj", "model.layers.53.mlp.experts.102.down_proj", "model.layers.53.mlp.experts.103.down_proj", "model.layers.53.mlp.experts.104.down_proj", "model.layers.53.mlp.experts.105.down_proj", "model.layers.53.mlp.experts.106.down_proj", "model.layers.53.mlp.experts.107.down_proj", "model.layers.53.mlp.experts.108.down_proj", "model.layers.53.mlp.experts.109.down_proj", "model.layers.53.mlp.experts.110.down_proj", "model.layers.53.mlp.experts.111.down_proj", "model.layers.53.mlp.experts.112.down_proj", "model.layers.53.mlp.experts.113.down_proj", "model.layers.53.mlp.experts.114.down_proj", "model.layers.53.mlp.experts.115.down_proj", "model.layers.53.mlp.experts.116.down_proj", "model.layers.53.mlp.experts.117.down_proj", "model.layers.53.mlp.experts.118.down_proj", "model.layers.53.mlp.experts.119.down_proj", "model.layers.53.mlp.experts.120.down_proj", "model.layers.53.mlp.experts.121.down_proj", "model.layers.53.mlp.experts.122.down_proj", "model.layers.53.mlp.experts.123.down_proj", "model.layers.53.mlp.experts.124.down_proj", "model.layers.53.mlp.experts.125.down_proj", "model.layers.53.mlp.experts.126.down_proj", "model.layers.53.mlp.experts.127.down_proj", "model.layers.53.mlp.experts.128.down_proj", "model.layers.53.mlp.experts.129.down_proj", "model.layers.53.mlp.experts.130.down_proj", "model.layers.53.mlp.experts.131.down_proj", "model.layers.53.mlp.experts.132.down_proj", "model.layers.53.mlp.experts.133.down_proj", "model.layers.53.mlp.experts.134.down_proj", "model.layers.53.mlp.experts.135.down_proj", "model.layers.53.mlp.experts.136.down_proj", "model.layers.53.mlp.experts.137.down_proj", "model.layers.53.mlp.experts.138.down_proj", "model.layers.53.mlp.experts.139.down_proj", "model.layers.53.mlp.experts.140.down_proj", "model.layers.53.mlp.experts.141.down_proj", "model.layers.53.mlp.experts.142.down_proj", "model.layers.53.mlp.experts.143.down_proj", "model.layers.53.mlp.experts.144.down_proj", "model.layers.53.mlp.experts.145.down_proj", "model.layers.53.mlp.experts.146.down_proj", "model.layers.53.mlp.experts.147.down_proj", "model.layers.53.mlp.experts.148.down_proj", "model.layers.53.mlp.experts.149.down_proj", "model.layers.53.mlp.experts.150.down_proj", "model.layers.53.mlp.experts.151.down_proj", "model.layers.53.mlp.experts.152.down_proj", "model.layers.53.mlp.experts.153.down_proj", "model.layers.53.mlp.experts.154.down_proj", "model.layers.53.mlp.experts.155.down_proj", "model.layers.53.mlp.experts.156.down_proj", "model.layers.53.mlp.experts.157.down_proj", "model.layers.53.mlp.experts.158.down_proj", "model.layers.53.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00016093440353870392, "dbits": 1258291200 }, { "dkld": -0.0001761811785399997, "dbits": 2516582400 }, { "dkld": -0.00021228631958365718, "dbits": 3774873600 }, { "dkld": -0.00019467789679766812, "dbits": 6291456000 } ] }, { "idx": 321, "layers": [ "model.layers.54.self_attn.q_proj" ], "candidates": [ { "dkld": 7.873205468057753e-05, "dbits": 62914560 }, { "dkld": 1.754332333803177e-05, "dbits": 125829120 }, { "dkld": -9.35270451009329e-05, "dbits": 188743680 }, { "dkld": -9.77268442511614e-05, "dbits": 314572800 } ] }, { "idx": 322, "layers": [ "model.layers.54.self_attn.k_proj", "model.layers.54.self_attn.v_proj" ], "candidates": [ { "dkld": 0.00047854725271462284, "dbits": 10485760 }, { "dkld": 0.00017042979598044794, "dbits": 20971520 }, { "dkld": 0.00017257258296012046, "dbits": 31457280 }, { "dkld": 0.00019882284104823233, "dbits": 52428800 } ] }, { "idx": 323, "layers": [ "model.layers.54.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0005420075729489354, "dbits": 62914560 }, { "dkld": -0.0004030522890389049, "dbits": 125829120 }, { "dkld": -0.0004452122375369072, "dbits": 188743680 }, { "dkld": -0.0004535999149084119, "dbits": 314572800 } ] }, { "idx": 324, "layers": [ "model.layers.54.mlp.shared_experts.gate_proj", "model.layers.54.mlp.shared_experts.up_proj", "model.layers.54.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00010577440261841375, "dbits": 23592960 }, { "dkld": -6.327852606774487e-05, "dbits": 47185920 }, { "dkld": 8.40630382298313e-06, "dbits": 70778880 }, { "dkld": -3.595557063817978e-05, "dbits": 117964800 } ] }, { "idx": 325, "layers": [ "model.layers.54.mlp.experts.0.gate_proj", "model.layers.54.mlp.experts.1.gate_proj", "model.layers.54.mlp.experts.2.gate_proj", "model.layers.54.mlp.experts.3.gate_proj", "model.layers.54.mlp.experts.4.gate_proj", "model.layers.54.mlp.experts.5.gate_proj", "model.layers.54.mlp.experts.6.gate_proj", "model.layers.54.mlp.experts.7.gate_proj", "model.layers.54.mlp.experts.8.gate_proj", "model.layers.54.mlp.experts.9.gate_proj", "model.layers.54.mlp.experts.10.gate_proj", "model.layers.54.mlp.experts.11.gate_proj", "model.layers.54.mlp.experts.12.gate_proj", "model.layers.54.mlp.experts.13.gate_proj", "model.layers.54.mlp.experts.14.gate_proj", "model.layers.54.mlp.experts.15.gate_proj", "model.layers.54.mlp.experts.16.gate_proj", "model.layers.54.mlp.experts.17.gate_proj", "model.layers.54.mlp.experts.18.gate_proj", "model.layers.54.mlp.experts.19.gate_proj", "model.layers.54.mlp.experts.20.gate_proj", "model.layers.54.mlp.experts.21.gate_proj", "model.layers.54.mlp.experts.22.gate_proj", "model.layers.54.mlp.experts.23.gate_proj", "model.layers.54.mlp.experts.24.gate_proj", "model.layers.54.mlp.experts.25.gate_proj", "model.layers.54.mlp.experts.26.gate_proj", "model.layers.54.mlp.experts.27.gate_proj", "model.layers.54.mlp.experts.28.gate_proj", "model.layers.54.mlp.experts.29.gate_proj", "model.layers.54.mlp.experts.30.gate_proj", "model.layers.54.mlp.experts.31.gate_proj", "model.layers.54.mlp.experts.32.gate_proj", "model.layers.54.mlp.experts.33.gate_proj", "model.layers.54.mlp.experts.34.gate_proj", "model.layers.54.mlp.experts.35.gate_proj", "model.layers.54.mlp.experts.36.gate_proj", "model.layers.54.mlp.experts.37.gate_proj", "model.layers.54.mlp.experts.38.gate_proj", "model.layers.54.mlp.experts.39.gate_proj", "model.layers.54.mlp.experts.40.gate_proj", "model.layers.54.mlp.experts.41.gate_proj", "model.layers.54.mlp.experts.42.gate_proj", "model.layers.54.mlp.experts.43.gate_proj", "model.layers.54.mlp.experts.44.gate_proj", "model.layers.54.mlp.experts.45.gate_proj", "model.layers.54.mlp.experts.46.gate_proj", "model.layers.54.mlp.experts.47.gate_proj", "model.layers.54.mlp.experts.48.gate_proj", "model.layers.54.mlp.experts.49.gate_proj", "model.layers.54.mlp.experts.50.gate_proj", "model.layers.54.mlp.experts.51.gate_proj", "model.layers.54.mlp.experts.52.gate_proj", "model.layers.54.mlp.experts.53.gate_proj", "model.layers.54.mlp.experts.54.gate_proj", "model.layers.54.mlp.experts.55.gate_proj", "model.layers.54.mlp.experts.56.gate_proj", "model.layers.54.mlp.experts.57.gate_proj", "model.layers.54.mlp.experts.58.gate_proj", "model.layers.54.mlp.experts.59.gate_proj", "model.layers.54.mlp.experts.60.gate_proj", "model.layers.54.mlp.experts.61.gate_proj", "model.layers.54.mlp.experts.62.gate_proj", "model.layers.54.mlp.experts.63.gate_proj", "model.layers.54.mlp.experts.64.gate_proj", "model.layers.54.mlp.experts.65.gate_proj", "model.layers.54.mlp.experts.66.gate_proj", "model.layers.54.mlp.experts.67.gate_proj", "model.layers.54.mlp.experts.68.gate_proj", "model.layers.54.mlp.experts.69.gate_proj", "model.layers.54.mlp.experts.70.gate_proj", "model.layers.54.mlp.experts.71.gate_proj", "model.layers.54.mlp.experts.72.gate_proj", "model.layers.54.mlp.experts.73.gate_proj", "model.layers.54.mlp.experts.74.gate_proj", "model.layers.54.mlp.experts.75.gate_proj", "model.layers.54.mlp.experts.76.gate_proj", "model.layers.54.mlp.experts.77.gate_proj", "model.layers.54.mlp.experts.78.gate_proj", "model.layers.54.mlp.experts.79.gate_proj", "model.layers.54.mlp.experts.80.gate_proj", "model.layers.54.mlp.experts.81.gate_proj", "model.layers.54.mlp.experts.82.gate_proj", "model.layers.54.mlp.experts.83.gate_proj", "model.layers.54.mlp.experts.84.gate_proj", "model.layers.54.mlp.experts.85.gate_proj", "model.layers.54.mlp.experts.86.gate_proj", "model.layers.54.mlp.experts.87.gate_proj", "model.layers.54.mlp.experts.88.gate_proj", "model.layers.54.mlp.experts.89.gate_proj", "model.layers.54.mlp.experts.90.gate_proj", "model.layers.54.mlp.experts.91.gate_proj", "model.layers.54.mlp.experts.92.gate_proj", "model.layers.54.mlp.experts.93.gate_proj", "model.layers.54.mlp.experts.94.gate_proj", "model.layers.54.mlp.experts.95.gate_proj", "model.layers.54.mlp.experts.96.gate_proj", "model.layers.54.mlp.experts.97.gate_proj", "model.layers.54.mlp.experts.98.gate_proj", "model.layers.54.mlp.experts.99.gate_proj", "model.layers.54.mlp.experts.100.gate_proj", "model.layers.54.mlp.experts.101.gate_proj", "model.layers.54.mlp.experts.102.gate_proj", "model.layers.54.mlp.experts.103.gate_proj", "model.layers.54.mlp.experts.104.gate_proj", "model.layers.54.mlp.experts.105.gate_proj", "model.layers.54.mlp.experts.106.gate_proj", "model.layers.54.mlp.experts.107.gate_proj", "model.layers.54.mlp.experts.108.gate_proj", "model.layers.54.mlp.experts.109.gate_proj", "model.layers.54.mlp.experts.110.gate_proj", "model.layers.54.mlp.experts.111.gate_proj", "model.layers.54.mlp.experts.112.gate_proj", "model.layers.54.mlp.experts.113.gate_proj", "model.layers.54.mlp.experts.114.gate_proj", "model.layers.54.mlp.experts.115.gate_proj", "model.layers.54.mlp.experts.116.gate_proj", "model.layers.54.mlp.experts.117.gate_proj", "model.layers.54.mlp.experts.118.gate_proj", "model.layers.54.mlp.experts.119.gate_proj", "model.layers.54.mlp.experts.120.gate_proj", "model.layers.54.mlp.experts.121.gate_proj", "model.layers.54.mlp.experts.122.gate_proj", "model.layers.54.mlp.experts.123.gate_proj", "model.layers.54.mlp.experts.124.gate_proj", "model.layers.54.mlp.experts.125.gate_proj", "model.layers.54.mlp.experts.126.gate_proj", "model.layers.54.mlp.experts.127.gate_proj", "model.layers.54.mlp.experts.128.gate_proj", "model.layers.54.mlp.experts.129.gate_proj", "model.layers.54.mlp.experts.130.gate_proj", "model.layers.54.mlp.experts.131.gate_proj", "model.layers.54.mlp.experts.132.gate_proj", "model.layers.54.mlp.experts.133.gate_proj", "model.layers.54.mlp.experts.134.gate_proj", "model.layers.54.mlp.experts.135.gate_proj", "model.layers.54.mlp.experts.136.gate_proj", "model.layers.54.mlp.experts.137.gate_proj", "model.layers.54.mlp.experts.138.gate_proj", "model.layers.54.mlp.experts.139.gate_proj", "model.layers.54.mlp.experts.140.gate_proj", "model.layers.54.mlp.experts.141.gate_proj", "model.layers.54.mlp.experts.142.gate_proj", "model.layers.54.mlp.experts.143.gate_proj", "model.layers.54.mlp.experts.144.gate_proj", "model.layers.54.mlp.experts.145.gate_proj", "model.layers.54.mlp.experts.146.gate_proj", "model.layers.54.mlp.experts.147.gate_proj", "model.layers.54.mlp.experts.148.gate_proj", "model.layers.54.mlp.experts.149.gate_proj", "model.layers.54.mlp.experts.150.gate_proj", "model.layers.54.mlp.experts.151.gate_proj", "model.layers.54.mlp.experts.152.gate_proj", "model.layers.54.mlp.experts.153.gate_proj", "model.layers.54.mlp.experts.154.gate_proj", "model.layers.54.mlp.experts.155.gate_proj", "model.layers.54.mlp.experts.156.gate_proj", "model.layers.54.mlp.experts.157.gate_proj", "model.layers.54.mlp.experts.158.gate_proj", "model.layers.54.mlp.experts.159.gate_proj", "model.layers.54.mlp.experts.0.up_proj", "model.layers.54.mlp.experts.1.up_proj", "model.layers.54.mlp.experts.2.up_proj", "model.layers.54.mlp.experts.3.up_proj", "model.layers.54.mlp.experts.4.up_proj", "model.layers.54.mlp.experts.5.up_proj", "model.layers.54.mlp.experts.6.up_proj", "model.layers.54.mlp.experts.7.up_proj", "model.layers.54.mlp.experts.8.up_proj", "model.layers.54.mlp.experts.9.up_proj", "model.layers.54.mlp.experts.10.up_proj", "model.layers.54.mlp.experts.11.up_proj", "model.layers.54.mlp.experts.12.up_proj", "model.layers.54.mlp.experts.13.up_proj", "model.layers.54.mlp.experts.14.up_proj", "model.layers.54.mlp.experts.15.up_proj", "model.layers.54.mlp.experts.16.up_proj", "model.layers.54.mlp.experts.17.up_proj", "model.layers.54.mlp.experts.18.up_proj", "model.layers.54.mlp.experts.19.up_proj", "model.layers.54.mlp.experts.20.up_proj", "model.layers.54.mlp.experts.21.up_proj", "model.layers.54.mlp.experts.22.up_proj", "model.layers.54.mlp.experts.23.up_proj", "model.layers.54.mlp.experts.24.up_proj", "model.layers.54.mlp.experts.25.up_proj", "model.layers.54.mlp.experts.26.up_proj", "model.layers.54.mlp.experts.27.up_proj", "model.layers.54.mlp.experts.28.up_proj", "model.layers.54.mlp.experts.29.up_proj", "model.layers.54.mlp.experts.30.up_proj", "model.layers.54.mlp.experts.31.up_proj", "model.layers.54.mlp.experts.32.up_proj", "model.layers.54.mlp.experts.33.up_proj", "model.layers.54.mlp.experts.34.up_proj", "model.layers.54.mlp.experts.35.up_proj", "model.layers.54.mlp.experts.36.up_proj", "model.layers.54.mlp.experts.37.up_proj", "model.layers.54.mlp.experts.38.up_proj", "model.layers.54.mlp.experts.39.up_proj", "model.layers.54.mlp.experts.40.up_proj", "model.layers.54.mlp.experts.41.up_proj", "model.layers.54.mlp.experts.42.up_proj", "model.layers.54.mlp.experts.43.up_proj", "model.layers.54.mlp.experts.44.up_proj", "model.layers.54.mlp.experts.45.up_proj", "model.layers.54.mlp.experts.46.up_proj", "model.layers.54.mlp.experts.47.up_proj", "model.layers.54.mlp.experts.48.up_proj", "model.layers.54.mlp.experts.49.up_proj", "model.layers.54.mlp.experts.50.up_proj", "model.layers.54.mlp.experts.51.up_proj", "model.layers.54.mlp.experts.52.up_proj", "model.layers.54.mlp.experts.53.up_proj", "model.layers.54.mlp.experts.54.up_proj", "model.layers.54.mlp.experts.55.up_proj", "model.layers.54.mlp.experts.56.up_proj", "model.layers.54.mlp.experts.57.up_proj", "model.layers.54.mlp.experts.58.up_proj", "model.layers.54.mlp.experts.59.up_proj", "model.layers.54.mlp.experts.60.up_proj", "model.layers.54.mlp.experts.61.up_proj", "model.layers.54.mlp.experts.62.up_proj", "model.layers.54.mlp.experts.63.up_proj", "model.layers.54.mlp.experts.64.up_proj", "model.layers.54.mlp.experts.65.up_proj", "model.layers.54.mlp.experts.66.up_proj", "model.layers.54.mlp.experts.67.up_proj", "model.layers.54.mlp.experts.68.up_proj", "model.layers.54.mlp.experts.69.up_proj", "model.layers.54.mlp.experts.70.up_proj", "model.layers.54.mlp.experts.71.up_proj", "model.layers.54.mlp.experts.72.up_proj", "model.layers.54.mlp.experts.73.up_proj", "model.layers.54.mlp.experts.74.up_proj", "model.layers.54.mlp.experts.75.up_proj", "model.layers.54.mlp.experts.76.up_proj", "model.layers.54.mlp.experts.77.up_proj", "model.layers.54.mlp.experts.78.up_proj", "model.layers.54.mlp.experts.79.up_proj", "model.layers.54.mlp.experts.80.up_proj", "model.layers.54.mlp.experts.81.up_proj", "model.layers.54.mlp.experts.82.up_proj", "model.layers.54.mlp.experts.83.up_proj", "model.layers.54.mlp.experts.84.up_proj", "model.layers.54.mlp.experts.85.up_proj", "model.layers.54.mlp.experts.86.up_proj", "model.layers.54.mlp.experts.87.up_proj", "model.layers.54.mlp.experts.88.up_proj", "model.layers.54.mlp.experts.89.up_proj", "model.layers.54.mlp.experts.90.up_proj", "model.layers.54.mlp.experts.91.up_proj", "model.layers.54.mlp.experts.92.up_proj", "model.layers.54.mlp.experts.93.up_proj", "model.layers.54.mlp.experts.94.up_proj", "model.layers.54.mlp.experts.95.up_proj", "model.layers.54.mlp.experts.96.up_proj", "model.layers.54.mlp.experts.97.up_proj", "model.layers.54.mlp.experts.98.up_proj", "model.layers.54.mlp.experts.99.up_proj", "model.layers.54.mlp.experts.100.up_proj", "model.layers.54.mlp.experts.101.up_proj", "model.layers.54.mlp.experts.102.up_proj", "model.layers.54.mlp.experts.103.up_proj", "model.layers.54.mlp.experts.104.up_proj", "model.layers.54.mlp.experts.105.up_proj", "model.layers.54.mlp.experts.106.up_proj", "model.layers.54.mlp.experts.107.up_proj", "model.layers.54.mlp.experts.108.up_proj", "model.layers.54.mlp.experts.109.up_proj", "model.layers.54.mlp.experts.110.up_proj", "model.layers.54.mlp.experts.111.up_proj", "model.layers.54.mlp.experts.112.up_proj", "model.layers.54.mlp.experts.113.up_proj", "model.layers.54.mlp.experts.114.up_proj", "model.layers.54.mlp.experts.115.up_proj", "model.layers.54.mlp.experts.116.up_proj", "model.layers.54.mlp.experts.117.up_proj", "model.layers.54.mlp.experts.118.up_proj", "model.layers.54.mlp.experts.119.up_proj", "model.layers.54.mlp.experts.120.up_proj", "model.layers.54.mlp.experts.121.up_proj", "model.layers.54.mlp.experts.122.up_proj", "model.layers.54.mlp.experts.123.up_proj", "model.layers.54.mlp.experts.124.up_proj", "model.layers.54.mlp.experts.125.up_proj", "model.layers.54.mlp.experts.126.up_proj", "model.layers.54.mlp.experts.127.up_proj", "model.layers.54.mlp.experts.128.up_proj", "model.layers.54.mlp.experts.129.up_proj", "model.layers.54.mlp.experts.130.up_proj", "model.layers.54.mlp.experts.131.up_proj", "model.layers.54.mlp.experts.132.up_proj", "model.layers.54.mlp.experts.133.up_proj", "model.layers.54.mlp.experts.134.up_proj", "model.layers.54.mlp.experts.135.up_proj", "model.layers.54.mlp.experts.136.up_proj", "model.layers.54.mlp.experts.137.up_proj", "model.layers.54.mlp.experts.138.up_proj", "model.layers.54.mlp.experts.139.up_proj", "model.layers.54.mlp.experts.140.up_proj", "model.layers.54.mlp.experts.141.up_proj", "model.layers.54.mlp.experts.142.up_proj", "model.layers.54.mlp.experts.143.up_proj", "model.layers.54.mlp.experts.144.up_proj", "model.layers.54.mlp.experts.145.up_proj", "model.layers.54.mlp.experts.146.up_proj", "model.layers.54.mlp.experts.147.up_proj", "model.layers.54.mlp.experts.148.up_proj", "model.layers.54.mlp.experts.149.up_proj", "model.layers.54.mlp.experts.150.up_proj", "model.layers.54.mlp.experts.151.up_proj", "model.layers.54.mlp.experts.152.up_proj", "model.layers.54.mlp.experts.153.up_proj", "model.layers.54.mlp.experts.154.up_proj", "model.layers.54.mlp.experts.155.up_proj", "model.layers.54.mlp.experts.156.up_proj", "model.layers.54.mlp.experts.157.up_proj", "model.layers.54.mlp.experts.158.up_proj", "model.layers.54.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00013593602925539294, "dbits": 2516582400 }, { "dkld": -0.00014613699167967398, "dbits": 5033164800 }, { "dkld": -0.00017573898658157505, "dbits": 7549747200 }, { "dkld": -0.00013609454035759527, "dbits": 12582912000 } ] }, { "idx": 326, "layers": [ "model.layers.54.mlp.experts.0.down_proj", "model.layers.54.mlp.experts.1.down_proj", "model.layers.54.mlp.experts.2.down_proj", "model.layers.54.mlp.experts.3.down_proj", "model.layers.54.mlp.experts.4.down_proj", "model.layers.54.mlp.experts.5.down_proj", "model.layers.54.mlp.experts.6.down_proj", "model.layers.54.mlp.experts.7.down_proj", "model.layers.54.mlp.experts.8.down_proj", "model.layers.54.mlp.experts.9.down_proj", "model.layers.54.mlp.experts.10.down_proj", "model.layers.54.mlp.experts.11.down_proj", "model.layers.54.mlp.experts.12.down_proj", "model.layers.54.mlp.experts.13.down_proj", "model.layers.54.mlp.experts.14.down_proj", "model.layers.54.mlp.experts.15.down_proj", "model.layers.54.mlp.experts.16.down_proj", "model.layers.54.mlp.experts.17.down_proj", "model.layers.54.mlp.experts.18.down_proj", "model.layers.54.mlp.experts.19.down_proj", "model.layers.54.mlp.experts.20.down_proj", "model.layers.54.mlp.experts.21.down_proj", "model.layers.54.mlp.experts.22.down_proj", "model.layers.54.mlp.experts.23.down_proj", "model.layers.54.mlp.experts.24.down_proj", "model.layers.54.mlp.experts.25.down_proj", "model.layers.54.mlp.experts.26.down_proj", "model.layers.54.mlp.experts.27.down_proj", "model.layers.54.mlp.experts.28.down_proj", "model.layers.54.mlp.experts.29.down_proj", "model.layers.54.mlp.experts.30.down_proj", "model.layers.54.mlp.experts.31.down_proj", "model.layers.54.mlp.experts.32.down_proj", "model.layers.54.mlp.experts.33.down_proj", "model.layers.54.mlp.experts.34.down_proj", "model.layers.54.mlp.experts.35.down_proj", "model.layers.54.mlp.experts.36.down_proj", "model.layers.54.mlp.experts.37.down_proj", "model.layers.54.mlp.experts.38.down_proj", "model.layers.54.mlp.experts.39.down_proj", "model.layers.54.mlp.experts.40.down_proj", "model.layers.54.mlp.experts.41.down_proj", "model.layers.54.mlp.experts.42.down_proj", "model.layers.54.mlp.experts.43.down_proj", "model.layers.54.mlp.experts.44.down_proj", "model.layers.54.mlp.experts.45.down_proj", "model.layers.54.mlp.experts.46.down_proj", "model.layers.54.mlp.experts.47.down_proj", "model.layers.54.mlp.experts.48.down_proj", "model.layers.54.mlp.experts.49.down_proj", "model.layers.54.mlp.experts.50.down_proj", "model.layers.54.mlp.experts.51.down_proj", "model.layers.54.mlp.experts.52.down_proj", "model.layers.54.mlp.experts.53.down_proj", "model.layers.54.mlp.experts.54.down_proj", "model.layers.54.mlp.experts.55.down_proj", "model.layers.54.mlp.experts.56.down_proj", "model.layers.54.mlp.experts.57.down_proj", "model.layers.54.mlp.experts.58.down_proj", "model.layers.54.mlp.experts.59.down_proj", "model.layers.54.mlp.experts.60.down_proj", "model.layers.54.mlp.experts.61.down_proj", "model.layers.54.mlp.experts.62.down_proj", "model.layers.54.mlp.experts.63.down_proj", "model.layers.54.mlp.experts.64.down_proj", "model.layers.54.mlp.experts.65.down_proj", "model.layers.54.mlp.experts.66.down_proj", "model.layers.54.mlp.experts.67.down_proj", "model.layers.54.mlp.experts.68.down_proj", "model.layers.54.mlp.experts.69.down_proj", "model.layers.54.mlp.experts.70.down_proj", "model.layers.54.mlp.experts.71.down_proj", "model.layers.54.mlp.experts.72.down_proj", "model.layers.54.mlp.experts.73.down_proj", "model.layers.54.mlp.experts.74.down_proj", "model.layers.54.mlp.experts.75.down_proj", "model.layers.54.mlp.experts.76.down_proj", "model.layers.54.mlp.experts.77.down_proj", "model.layers.54.mlp.experts.78.down_proj", "model.layers.54.mlp.experts.79.down_proj", "model.layers.54.mlp.experts.80.down_proj", "model.layers.54.mlp.experts.81.down_proj", "model.layers.54.mlp.experts.82.down_proj", "model.layers.54.mlp.experts.83.down_proj", "model.layers.54.mlp.experts.84.down_proj", "model.layers.54.mlp.experts.85.down_proj", "model.layers.54.mlp.experts.86.down_proj", "model.layers.54.mlp.experts.87.down_proj", "model.layers.54.mlp.experts.88.down_proj", "model.layers.54.mlp.experts.89.down_proj", "model.layers.54.mlp.experts.90.down_proj", "model.layers.54.mlp.experts.91.down_proj", "model.layers.54.mlp.experts.92.down_proj", "model.layers.54.mlp.experts.93.down_proj", "model.layers.54.mlp.experts.94.down_proj", "model.layers.54.mlp.experts.95.down_proj", "model.layers.54.mlp.experts.96.down_proj", "model.layers.54.mlp.experts.97.down_proj", "model.layers.54.mlp.experts.98.down_proj", "model.layers.54.mlp.experts.99.down_proj", "model.layers.54.mlp.experts.100.down_proj", "model.layers.54.mlp.experts.101.down_proj", "model.layers.54.mlp.experts.102.down_proj", "model.layers.54.mlp.experts.103.down_proj", "model.layers.54.mlp.experts.104.down_proj", "model.layers.54.mlp.experts.105.down_proj", "model.layers.54.mlp.experts.106.down_proj", "model.layers.54.mlp.experts.107.down_proj", "model.layers.54.mlp.experts.108.down_proj", "model.layers.54.mlp.experts.109.down_proj", "model.layers.54.mlp.experts.110.down_proj", "model.layers.54.mlp.experts.111.down_proj", "model.layers.54.mlp.experts.112.down_proj", "model.layers.54.mlp.experts.113.down_proj", "model.layers.54.mlp.experts.114.down_proj", "model.layers.54.mlp.experts.115.down_proj", "model.layers.54.mlp.experts.116.down_proj", "model.layers.54.mlp.experts.117.down_proj", "model.layers.54.mlp.experts.118.down_proj", "model.layers.54.mlp.experts.119.down_proj", "model.layers.54.mlp.experts.120.down_proj", "model.layers.54.mlp.experts.121.down_proj", "model.layers.54.mlp.experts.122.down_proj", "model.layers.54.mlp.experts.123.down_proj", "model.layers.54.mlp.experts.124.down_proj", "model.layers.54.mlp.experts.125.down_proj", "model.layers.54.mlp.experts.126.down_proj", "model.layers.54.mlp.experts.127.down_proj", "model.layers.54.mlp.experts.128.down_proj", "model.layers.54.mlp.experts.129.down_proj", "model.layers.54.mlp.experts.130.down_proj", "model.layers.54.mlp.experts.131.down_proj", "model.layers.54.mlp.experts.132.down_proj", "model.layers.54.mlp.experts.133.down_proj", "model.layers.54.mlp.experts.134.down_proj", "model.layers.54.mlp.experts.135.down_proj", "model.layers.54.mlp.experts.136.down_proj", "model.layers.54.mlp.experts.137.down_proj", "model.layers.54.mlp.experts.138.down_proj", "model.layers.54.mlp.experts.139.down_proj", "model.layers.54.mlp.experts.140.down_proj", "model.layers.54.mlp.experts.141.down_proj", "model.layers.54.mlp.experts.142.down_proj", "model.layers.54.mlp.experts.143.down_proj", "model.layers.54.mlp.experts.144.down_proj", "model.layers.54.mlp.experts.145.down_proj", "model.layers.54.mlp.experts.146.down_proj", "model.layers.54.mlp.experts.147.down_proj", "model.layers.54.mlp.experts.148.down_proj", "model.layers.54.mlp.experts.149.down_proj", "model.layers.54.mlp.experts.150.down_proj", "model.layers.54.mlp.experts.151.down_proj", "model.layers.54.mlp.experts.152.down_proj", "model.layers.54.mlp.experts.153.down_proj", "model.layers.54.mlp.experts.154.down_proj", "model.layers.54.mlp.experts.155.down_proj", "model.layers.54.mlp.experts.156.down_proj", "model.layers.54.mlp.experts.157.down_proj", "model.layers.54.mlp.experts.158.down_proj", "model.layers.54.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -9.117685258389074e-05, "dbits": 1258291200 }, { "dkld": -6.251335144043246e-05, "dbits": 2516582400 }, { "dkld": -7.420582696796019e-05, "dbits": 3774873600 }, { "dkld": -8.30156728625353e-05, "dbits": 6291456000 } ] }, { "idx": 327, "layers": [ "model.layers.55.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00017503621056676189, "dbits": 62914560 }, { "dkld": -9.433748200536607e-05, "dbits": 125829120 }, { "dkld": -7.278211414814273e-05, "dbits": 188743680 }, { "dkld": -0.00010027717798949398, "dbits": 314572800 } ] }, { "idx": 328, "layers": [ "model.layers.55.self_attn.k_proj", "model.layers.55.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00011690547689796604, "dbits": 10485760 }, { "dkld": 6.613824516533973e-05, "dbits": 20971520 }, { "dkld": 5.1425304263830185e-05, "dbits": 31457280 }, { "dkld": 7.146764546632489e-05, "dbits": 52428800 } ] }, { "idx": 329, "layers": [ "model.layers.55.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00044184587895870764, "dbits": 62914560 }, { "dkld": -0.0004554975777864567, "dbits": 125829120 }, { "dkld": -0.000415680930018425, "dbits": 188743680 }, { "dkld": -0.0004241586662828978, "dbits": 314572800 } ] }, { "idx": 330, "layers": [ "model.layers.55.mlp.shared_experts.gate_proj", "model.layers.55.mlp.shared_experts.up_proj", "model.layers.55.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0001873900182545185, "dbits": 23592960 }, { "dkld": -0.0002848700620234068, "dbits": 47185920 }, { "dkld": -0.0002432078123092707, "dbits": 70778880 }, { "dkld": -0.00025705564767122824, "dbits": 117964800 } ] }, { "idx": 331, "layers": [ "model.layers.55.mlp.experts.0.gate_proj", "model.layers.55.mlp.experts.1.gate_proj", "model.layers.55.mlp.experts.2.gate_proj", "model.layers.55.mlp.experts.3.gate_proj", "model.layers.55.mlp.experts.4.gate_proj", "model.layers.55.mlp.experts.5.gate_proj", "model.layers.55.mlp.experts.6.gate_proj", "model.layers.55.mlp.experts.7.gate_proj", "model.layers.55.mlp.experts.8.gate_proj", "model.layers.55.mlp.experts.9.gate_proj", "model.layers.55.mlp.experts.10.gate_proj", "model.layers.55.mlp.experts.11.gate_proj", "model.layers.55.mlp.experts.12.gate_proj", "model.layers.55.mlp.experts.13.gate_proj", "model.layers.55.mlp.experts.14.gate_proj", "model.layers.55.mlp.experts.15.gate_proj", "model.layers.55.mlp.experts.16.gate_proj", "model.layers.55.mlp.experts.17.gate_proj", "model.layers.55.mlp.experts.18.gate_proj", "model.layers.55.mlp.experts.19.gate_proj", "model.layers.55.mlp.experts.20.gate_proj", "model.layers.55.mlp.experts.21.gate_proj", "model.layers.55.mlp.experts.22.gate_proj", "model.layers.55.mlp.experts.23.gate_proj", "model.layers.55.mlp.experts.24.gate_proj", "model.layers.55.mlp.experts.25.gate_proj", "model.layers.55.mlp.experts.26.gate_proj", "model.layers.55.mlp.experts.27.gate_proj", "model.layers.55.mlp.experts.28.gate_proj", "model.layers.55.mlp.experts.29.gate_proj", "model.layers.55.mlp.experts.30.gate_proj", "model.layers.55.mlp.experts.31.gate_proj", "model.layers.55.mlp.experts.32.gate_proj", "model.layers.55.mlp.experts.33.gate_proj", "model.layers.55.mlp.experts.34.gate_proj", "model.layers.55.mlp.experts.35.gate_proj", "model.layers.55.mlp.experts.36.gate_proj", "model.layers.55.mlp.experts.37.gate_proj", "model.layers.55.mlp.experts.38.gate_proj", "model.layers.55.mlp.experts.39.gate_proj", "model.layers.55.mlp.experts.40.gate_proj", "model.layers.55.mlp.experts.41.gate_proj", "model.layers.55.mlp.experts.42.gate_proj", "model.layers.55.mlp.experts.43.gate_proj", "model.layers.55.mlp.experts.44.gate_proj", "model.layers.55.mlp.experts.45.gate_proj", "model.layers.55.mlp.experts.46.gate_proj", "model.layers.55.mlp.experts.47.gate_proj", "model.layers.55.mlp.experts.48.gate_proj", "model.layers.55.mlp.experts.49.gate_proj", "model.layers.55.mlp.experts.50.gate_proj", "model.layers.55.mlp.experts.51.gate_proj", "model.layers.55.mlp.experts.52.gate_proj", "model.layers.55.mlp.experts.53.gate_proj", "model.layers.55.mlp.experts.54.gate_proj", "model.layers.55.mlp.experts.55.gate_proj", "model.layers.55.mlp.experts.56.gate_proj", "model.layers.55.mlp.experts.57.gate_proj", "model.layers.55.mlp.experts.58.gate_proj", "model.layers.55.mlp.experts.59.gate_proj", "model.layers.55.mlp.experts.60.gate_proj", "model.layers.55.mlp.experts.61.gate_proj", "model.layers.55.mlp.experts.62.gate_proj", "model.layers.55.mlp.experts.63.gate_proj", "model.layers.55.mlp.experts.64.gate_proj", "model.layers.55.mlp.experts.65.gate_proj", "model.layers.55.mlp.experts.66.gate_proj", "model.layers.55.mlp.experts.67.gate_proj", "model.layers.55.mlp.experts.68.gate_proj", "model.layers.55.mlp.experts.69.gate_proj", "model.layers.55.mlp.experts.70.gate_proj", "model.layers.55.mlp.experts.71.gate_proj", "model.layers.55.mlp.experts.72.gate_proj", "model.layers.55.mlp.experts.73.gate_proj", "model.layers.55.mlp.experts.74.gate_proj", "model.layers.55.mlp.experts.75.gate_proj", "model.layers.55.mlp.experts.76.gate_proj", "model.layers.55.mlp.experts.77.gate_proj", "model.layers.55.mlp.experts.78.gate_proj", "model.layers.55.mlp.experts.79.gate_proj", "model.layers.55.mlp.experts.80.gate_proj", "model.layers.55.mlp.experts.81.gate_proj", "model.layers.55.mlp.experts.82.gate_proj", "model.layers.55.mlp.experts.83.gate_proj", "model.layers.55.mlp.experts.84.gate_proj", "model.layers.55.mlp.experts.85.gate_proj", "model.layers.55.mlp.experts.86.gate_proj", "model.layers.55.mlp.experts.87.gate_proj", "model.layers.55.mlp.experts.88.gate_proj", "model.layers.55.mlp.experts.89.gate_proj", "model.layers.55.mlp.experts.90.gate_proj", "model.layers.55.mlp.experts.91.gate_proj", "model.layers.55.mlp.experts.92.gate_proj", "model.layers.55.mlp.experts.93.gate_proj", "model.layers.55.mlp.experts.94.gate_proj", "model.layers.55.mlp.experts.95.gate_proj", "model.layers.55.mlp.experts.96.gate_proj", "model.layers.55.mlp.experts.97.gate_proj", "model.layers.55.mlp.experts.98.gate_proj", "model.layers.55.mlp.experts.99.gate_proj", "model.layers.55.mlp.experts.100.gate_proj", "model.layers.55.mlp.experts.101.gate_proj", "model.layers.55.mlp.experts.102.gate_proj", "model.layers.55.mlp.experts.103.gate_proj", "model.layers.55.mlp.experts.104.gate_proj", "model.layers.55.mlp.experts.105.gate_proj", "model.layers.55.mlp.experts.106.gate_proj", "model.layers.55.mlp.experts.107.gate_proj", "model.layers.55.mlp.experts.108.gate_proj", "model.layers.55.mlp.experts.109.gate_proj", "model.layers.55.mlp.experts.110.gate_proj", "model.layers.55.mlp.experts.111.gate_proj", "model.layers.55.mlp.experts.112.gate_proj", "model.layers.55.mlp.experts.113.gate_proj", "model.layers.55.mlp.experts.114.gate_proj", "model.layers.55.mlp.experts.115.gate_proj", "model.layers.55.mlp.experts.116.gate_proj", "model.layers.55.mlp.experts.117.gate_proj", "model.layers.55.mlp.experts.118.gate_proj", "model.layers.55.mlp.experts.119.gate_proj", "model.layers.55.mlp.experts.120.gate_proj", "model.layers.55.mlp.experts.121.gate_proj", "model.layers.55.mlp.experts.122.gate_proj", "model.layers.55.mlp.experts.123.gate_proj", "model.layers.55.mlp.experts.124.gate_proj", "model.layers.55.mlp.experts.125.gate_proj", "model.layers.55.mlp.experts.126.gate_proj", "model.layers.55.mlp.experts.127.gate_proj", "model.layers.55.mlp.experts.128.gate_proj", "model.layers.55.mlp.experts.129.gate_proj", "model.layers.55.mlp.experts.130.gate_proj", "model.layers.55.mlp.experts.131.gate_proj", "model.layers.55.mlp.experts.132.gate_proj", "model.layers.55.mlp.experts.133.gate_proj", "model.layers.55.mlp.experts.134.gate_proj", "model.layers.55.mlp.experts.135.gate_proj", "model.layers.55.mlp.experts.136.gate_proj", "model.layers.55.mlp.experts.137.gate_proj", "model.layers.55.mlp.experts.138.gate_proj", "model.layers.55.mlp.experts.139.gate_proj", "model.layers.55.mlp.experts.140.gate_proj", "model.layers.55.mlp.experts.141.gate_proj", "model.layers.55.mlp.experts.142.gate_proj", "model.layers.55.mlp.experts.143.gate_proj", "model.layers.55.mlp.experts.144.gate_proj", "model.layers.55.mlp.experts.145.gate_proj", "model.layers.55.mlp.experts.146.gate_proj", "model.layers.55.mlp.experts.147.gate_proj", "model.layers.55.mlp.experts.148.gate_proj", "model.layers.55.mlp.experts.149.gate_proj", "model.layers.55.mlp.experts.150.gate_proj", "model.layers.55.mlp.experts.151.gate_proj", "model.layers.55.mlp.experts.152.gate_proj", "model.layers.55.mlp.experts.153.gate_proj", "model.layers.55.mlp.experts.154.gate_proj", "model.layers.55.mlp.experts.155.gate_proj", "model.layers.55.mlp.experts.156.gate_proj", "model.layers.55.mlp.experts.157.gate_proj", "model.layers.55.mlp.experts.158.gate_proj", "model.layers.55.mlp.experts.159.gate_proj", "model.layers.55.mlp.experts.0.up_proj", "model.layers.55.mlp.experts.1.up_proj", "model.layers.55.mlp.experts.2.up_proj", "model.layers.55.mlp.experts.3.up_proj", "model.layers.55.mlp.experts.4.up_proj", "model.layers.55.mlp.experts.5.up_proj", "model.layers.55.mlp.experts.6.up_proj", "model.layers.55.mlp.experts.7.up_proj", "model.layers.55.mlp.experts.8.up_proj", "model.layers.55.mlp.experts.9.up_proj", "model.layers.55.mlp.experts.10.up_proj", "model.layers.55.mlp.experts.11.up_proj", "model.layers.55.mlp.experts.12.up_proj", "model.layers.55.mlp.experts.13.up_proj", "model.layers.55.mlp.experts.14.up_proj", "model.layers.55.mlp.experts.15.up_proj", "model.layers.55.mlp.experts.16.up_proj", "model.layers.55.mlp.experts.17.up_proj", "model.layers.55.mlp.experts.18.up_proj", "model.layers.55.mlp.experts.19.up_proj", "model.layers.55.mlp.experts.20.up_proj", "model.layers.55.mlp.experts.21.up_proj", "model.layers.55.mlp.experts.22.up_proj", "model.layers.55.mlp.experts.23.up_proj", "model.layers.55.mlp.experts.24.up_proj", "model.layers.55.mlp.experts.25.up_proj", "model.layers.55.mlp.experts.26.up_proj", "model.layers.55.mlp.experts.27.up_proj", "model.layers.55.mlp.experts.28.up_proj", "model.layers.55.mlp.experts.29.up_proj", "model.layers.55.mlp.experts.30.up_proj", "model.layers.55.mlp.experts.31.up_proj", "model.layers.55.mlp.experts.32.up_proj", "model.layers.55.mlp.experts.33.up_proj", "model.layers.55.mlp.experts.34.up_proj", "model.layers.55.mlp.experts.35.up_proj", "model.layers.55.mlp.experts.36.up_proj", "model.layers.55.mlp.experts.37.up_proj", "model.layers.55.mlp.experts.38.up_proj", "model.layers.55.mlp.experts.39.up_proj", "model.layers.55.mlp.experts.40.up_proj", "model.layers.55.mlp.experts.41.up_proj", "model.layers.55.mlp.experts.42.up_proj", "model.layers.55.mlp.experts.43.up_proj", "model.layers.55.mlp.experts.44.up_proj", "model.layers.55.mlp.experts.45.up_proj", "model.layers.55.mlp.experts.46.up_proj", "model.layers.55.mlp.experts.47.up_proj", "model.layers.55.mlp.experts.48.up_proj", "model.layers.55.mlp.experts.49.up_proj", "model.layers.55.mlp.experts.50.up_proj", "model.layers.55.mlp.experts.51.up_proj", "model.layers.55.mlp.experts.52.up_proj", "model.layers.55.mlp.experts.53.up_proj", "model.layers.55.mlp.experts.54.up_proj", "model.layers.55.mlp.experts.55.up_proj", "model.layers.55.mlp.experts.56.up_proj", "model.layers.55.mlp.experts.57.up_proj", "model.layers.55.mlp.experts.58.up_proj", "model.layers.55.mlp.experts.59.up_proj", "model.layers.55.mlp.experts.60.up_proj", "model.layers.55.mlp.experts.61.up_proj", "model.layers.55.mlp.experts.62.up_proj", "model.layers.55.mlp.experts.63.up_proj", "model.layers.55.mlp.experts.64.up_proj", "model.layers.55.mlp.experts.65.up_proj", "model.layers.55.mlp.experts.66.up_proj", "model.layers.55.mlp.experts.67.up_proj", "model.layers.55.mlp.experts.68.up_proj", "model.layers.55.mlp.experts.69.up_proj", "model.layers.55.mlp.experts.70.up_proj", "model.layers.55.mlp.experts.71.up_proj", "model.layers.55.mlp.experts.72.up_proj", "model.layers.55.mlp.experts.73.up_proj", "model.layers.55.mlp.experts.74.up_proj", "model.layers.55.mlp.experts.75.up_proj", "model.layers.55.mlp.experts.76.up_proj", "model.layers.55.mlp.experts.77.up_proj", "model.layers.55.mlp.experts.78.up_proj", "model.layers.55.mlp.experts.79.up_proj", "model.layers.55.mlp.experts.80.up_proj", "model.layers.55.mlp.experts.81.up_proj", "model.layers.55.mlp.experts.82.up_proj", "model.layers.55.mlp.experts.83.up_proj", "model.layers.55.mlp.experts.84.up_proj", "model.layers.55.mlp.experts.85.up_proj", "model.layers.55.mlp.experts.86.up_proj", "model.layers.55.mlp.experts.87.up_proj", "model.layers.55.mlp.experts.88.up_proj", "model.layers.55.mlp.experts.89.up_proj", "model.layers.55.mlp.experts.90.up_proj", "model.layers.55.mlp.experts.91.up_proj", "model.layers.55.mlp.experts.92.up_proj", "model.layers.55.mlp.experts.93.up_proj", "model.layers.55.mlp.experts.94.up_proj", "model.layers.55.mlp.experts.95.up_proj", "model.layers.55.mlp.experts.96.up_proj", "model.layers.55.mlp.experts.97.up_proj", "model.layers.55.mlp.experts.98.up_proj", "model.layers.55.mlp.experts.99.up_proj", "model.layers.55.mlp.experts.100.up_proj", "model.layers.55.mlp.experts.101.up_proj", "model.layers.55.mlp.experts.102.up_proj", "model.layers.55.mlp.experts.103.up_proj", "model.layers.55.mlp.experts.104.up_proj", "model.layers.55.mlp.experts.105.up_proj", "model.layers.55.mlp.experts.106.up_proj", "model.layers.55.mlp.experts.107.up_proj", "model.layers.55.mlp.experts.108.up_proj", "model.layers.55.mlp.experts.109.up_proj", "model.layers.55.mlp.experts.110.up_proj", "model.layers.55.mlp.experts.111.up_proj", "model.layers.55.mlp.experts.112.up_proj", "model.layers.55.mlp.experts.113.up_proj", "model.layers.55.mlp.experts.114.up_proj", "model.layers.55.mlp.experts.115.up_proj", "model.layers.55.mlp.experts.116.up_proj", "model.layers.55.mlp.experts.117.up_proj", "model.layers.55.mlp.experts.118.up_proj", "model.layers.55.mlp.experts.119.up_proj", "model.layers.55.mlp.experts.120.up_proj", "model.layers.55.mlp.experts.121.up_proj", "model.layers.55.mlp.experts.122.up_proj", "model.layers.55.mlp.experts.123.up_proj", "model.layers.55.mlp.experts.124.up_proj", "model.layers.55.mlp.experts.125.up_proj", "model.layers.55.mlp.experts.126.up_proj", "model.layers.55.mlp.experts.127.up_proj", "model.layers.55.mlp.experts.128.up_proj", "model.layers.55.mlp.experts.129.up_proj", "model.layers.55.mlp.experts.130.up_proj", "model.layers.55.mlp.experts.131.up_proj", "model.layers.55.mlp.experts.132.up_proj", "model.layers.55.mlp.experts.133.up_proj", "model.layers.55.mlp.experts.134.up_proj", "model.layers.55.mlp.experts.135.up_proj", "model.layers.55.mlp.experts.136.up_proj", "model.layers.55.mlp.experts.137.up_proj", "model.layers.55.mlp.experts.138.up_proj", "model.layers.55.mlp.experts.139.up_proj", "model.layers.55.mlp.experts.140.up_proj", "model.layers.55.mlp.experts.141.up_proj", "model.layers.55.mlp.experts.142.up_proj", "model.layers.55.mlp.experts.143.up_proj", "model.layers.55.mlp.experts.144.up_proj", "model.layers.55.mlp.experts.145.up_proj", "model.layers.55.mlp.experts.146.up_proj", "model.layers.55.mlp.experts.147.up_proj", "model.layers.55.mlp.experts.148.up_proj", "model.layers.55.mlp.experts.149.up_proj", "model.layers.55.mlp.experts.150.up_proj", "model.layers.55.mlp.experts.151.up_proj", "model.layers.55.mlp.experts.152.up_proj", "model.layers.55.mlp.experts.153.up_proj", "model.layers.55.mlp.experts.154.up_proj", "model.layers.55.mlp.experts.155.up_proj", "model.layers.55.mlp.experts.156.up_proj", "model.layers.55.mlp.experts.157.up_proj", "model.layers.55.mlp.experts.158.up_proj", "model.layers.55.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00014166589826346276, "dbits": 2516582400 }, { "dkld": -0.00017840582877398647, "dbits": 5033164800 }, { "dkld": -0.00014180317521095276, "dbits": 7549747200 }, { "dkld": -0.00014358218759298602, "dbits": 12582912000 } ] }, { "idx": 332, "layers": [ "model.layers.55.mlp.experts.0.down_proj", "model.layers.55.mlp.experts.1.down_proj", "model.layers.55.mlp.experts.2.down_proj", "model.layers.55.mlp.experts.3.down_proj", "model.layers.55.mlp.experts.4.down_proj", "model.layers.55.mlp.experts.5.down_proj", "model.layers.55.mlp.experts.6.down_proj", "model.layers.55.mlp.experts.7.down_proj", "model.layers.55.mlp.experts.8.down_proj", "model.layers.55.mlp.experts.9.down_proj", "model.layers.55.mlp.experts.10.down_proj", "model.layers.55.mlp.experts.11.down_proj", "model.layers.55.mlp.experts.12.down_proj", "model.layers.55.mlp.experts.13.down_proj", "model.layers.55.mlp.experts.14.down_proj", "model.layers.55.mlp.experts.15.down_proj", "model.layers.55.mlp.experts.16.down_proj", "model.layers.55.mlp.experts.17.down_proj", "model.layers.55.mlp.experts.18.down_proj", "model.layers.55.mlp.experts.19.down_proj", "model.layers.55.mlp.experts.20.down_proj", "model.layers.55.mlp.experts.21.down_proj", "model.layers.55.mlp.experts.22.down_proj", "model.layers.55.mlp.experts.23.down_proj", "model.layers.55.mlp.experts.24.down_proj", "model.layers.55.mlp.experts.25.down_proj", "model.layers.55.mlp.experts.26.down_proj", "model.layers.55.mlp.experts.27.down_proj", "model.layers.55.mlp.experts.28.down_proj", "model.layers.55.mlp.experts.29.down_proj", "model.layers.55.mlp.experts.30.down_proj", "model.layers.55.mlp.experts.31.down_proj", "model.layers.55.mlp.experts.32.down_proj", "model.layers.55.mlp.experts.33.down_proj", "model.layers.55.mlp.experts.34.down_proj", "model.layers.55.mlp.experts.35.down_proj", "model.layers.55.mlp.experts.36.down_proj", "model.layers.55.mlp.experts.37.down_proj", "model.layers.55.mlp.experts.38.down_proj", "model.layers.55.mlp.experts.39.down_proj", "model.layers.55.mlp.experts.40.down_proj", "model.layers.55.mlp.experts.41.down_proj", "model.layers.55.mlp.experts.42.down_proj", "model.layers.55.mlp.experts.43.down_proj", "model.layers.55.mlp.experts.44.down_proj", "model.layers.55.mlp.experts.45.down_proj", "model.layers.55.mlp.experts.46.down_proj", "model.layers.55.mlp.experts.47.down_proj", "model.layers.55.mlp.experts.48.down_proj", "model.layers.55.mlp.experts.49.down_proj", "model.layers.55.mlp.experts.50.down_proj", "model.layers.55.mlp.experts.51.down_proj", "model.layers.55.mlp.experts.52.down_proj", "model.layers.55.mlp.experts.53.down_proj", "model.layers.55.mlp.experts.54.down_proj", "model.layers.55.mlp.experts.55.down_proj", "model.layers.55.mlp.experts.56.down_proj", "model.layers.55.mlp.experts.57.down_proj", "model.layers.55.mlp.experts.58.down_proj", "model.layers.55.mlp.experts.59.down_proj", "model.layers.55.mlp.experts.60.down_proj", "model.layers.55.mlp.experts.61.down_proj", "model.layers.55.mlp.experts.62.down_proj", "model.layers.55.mlp.experts.63.down_proj", "model.layers.55.mlp.experts.64.down_proj", "model.layers.55.mlp.experts.65.down_proj", "model.layers.55.mlp.experts.66.down_proj", "model.layers.55.mlp.experts.67.down_proj", "model.layers.55.mlp.experts.68.down_proj", "model.layers.55.mlp.experts.69.down_proj", "model.layers.55.mlp.experts.70.down_proj", "model.layers.55.mlp.experts.71.down_proj", "model.layers.55.mlp.experts.72.down_proj", "model.layers.55.mlp.experts.73.down_proj", "model.layers.55.mlp.experts.74.down_proj", "model.layers.55.mlp.experts.75.down_proj", "model.layers.55.mlp.experts.76.down_proj", "model.layers.55.mlp.experts.77.down_proj", "model.layers.55.mlp.experts.78.down_proj", "model.layers.55.mlp.experts.79.down_proj", "model.layers.55.mlp.experts.80.down_proj", "model.layers.55.mlp.experts.81.down_proj", "model.layers.55.mlp.experts.82.down_proj", "model.layers.55.mlp.experts.83.down_proj", "model.layers.55.mlp.experts.84.down_proj", "model.layers.55.mlp.experts.85.down_proj", "model.layers.55.mlp.experts.86.down_proj", "model.layers.55.mlp.experts.87.down_proj", "model.layers.55.mlp.experts.88.down_proj", "model.layers.55.mlp.experts.89.down_proj", "model.layers.55.mlp.experts.90.down_proj", "model.layers.55.mlp.experts.91.down_proj", "model.layers.55.mlp.experts.92.down_proj", "model.layers.55.mlp.experts.93.down_proj", "model.layers.55.mlp.experts.94.down_proj", "model.layers.55.mlp.experts.95.down_proj", "model.layers.55.mlp.experts.96.down_proj", "model.layers.55.mlp.experts.97.down_proj", "model.layers.55.mlp.experts.98.down_proj", "model.layers.55.mlp.experts.99.down_proj", "model.layers.55.mlp.experts.100.down_proj", "model.layers.55.mlp.experts.101.down_proj", "model.layers.55.mlp.experts.102.down_proj", "model.layers.55.mlp.experts.103.down_proj", "model.layers.55.mlp.experts.104.down_proj", "model.layers.55.mlp.experts.105.down_proj", "model.layers.55.mlp.experts.106.down_proj", "model.layers.55.mlp.experts.107.down_proj", "model.layers.55.mlp.experts.108.down_proj", "model.layers.55.mlp.experts.109.down_proj", "model.layers.55.mlp.experts.110.down_proj", "model.layers.55.mlp.experts.111.down_proj", "model.layers.55.mlp.experts.112.down_proj", "model.layers.55.mlp.experts.113.down_proj", "model.layers.55.mlp.experts.114.down_proj", "model.layers.55.mlp.experts.115.down_proj", "model.layers.55.mlp.experts.116.down_proj", "model.layers.55.mlp.experts.117.down_proj", "model.layers.55.mlp.experts.118.down_proj", "model.layers.55.mlp.experts.119.down_proj", "model.layers.55.mlp.experts.120.down_proj", "model.layers.55.mlp.experts.121.down_proj", "model.layers.55.mlp.experts.122.down_proj", "model.layers.55.mlp.experts.123.down_proj", "model.layers.55.mlp.experts.124.down_proj", "model.layers.55.mlp.experts.125.down_proj", "model.layers.55.mlp.experts.126.down_proj", "model.layers.55.mlp.experts.127.down_proj", "model.layers.55.mlp.experts.128.down_proj", "model.layers.55.mlp.experts.129.down_proj", "model.layers.55.mlp.experts.130.down_proj", "model.layers.55.mlp.experts.131.down_proj", "model.layers.55.mlp.experts.132.down_proj", "model.layers.55.mlp.experts.133.down_proj", "model.layers.55.mlp.experts.134.down_proj", "model.layers.55.mlp.experts.135.down_proj", "model.layers.55.mlp.experts.136.down_proj", "model.layers.55.mlp.experts.137.down_proj", "model.layers.55.mlp.experts.138.down_proj", "model.layers.55.mlp.experts.139.down_proj", "model.layers.55.mlp.experts.140.down_proj", "model.layers.55.mlp.experts.141.down_proj", "model.layers.55.mlp.experts.142.down_proj", "model.layers.55.mlp.experts.143.down_proj", "model.layers.55.mlp.experts.144.down_proj", "model.layers.55.mlp.experts.145.down_proj", "model.layers.55.mlp.experts.146.down_proj", "model.layers.55.mlp.experts.147.down_proj", "model.layers.55.mlp.experts.148.down_proj", "model.layers.55.mlp.experts.149.down_proj", "model.layers.55.mlp.experts.150.down_proj", "model.layers.55.mlp.experts.151.down_proj", "model.layers.55.mlp.experts.152.down_proj", "model.layers.55.mlp.experts.153.down_proj", "model.layers.55.mlp.experts.154.down_proj", "model.layers.55.mlp.experts.155.down_proj", "model.layers.55.mlp.experts.156.down_proj", "model.layers.55.mlp.experts.157.down_proj", "model.layers.55.mlp.experts.158.down_proj", "model.layers.55.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -3.4491345286380426e-05, "dbits": 1258291200 }, { "dkld": -0.00010560005903244296, "dbits": 2516582400 }, { "dkld": -9.971391409635544e-05, "dbits": 3774873600 }, { "dkld": -0.00010242369025946496, "dbits": 6291456000 } ] }, { "idx": 333, "layers": [ "model.layers.56.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0002997768111526883, "dbits": 62914560 }, { "dkld": 0.00024653300642966947, "dbits": 125829120 }, { "dkld": 0.00023834723979233863, "dbits": 188743680 }, { "dkld": 0.00023930016905068796, "dbits": 314572800 } ] }, { "idx": 334, "layers": [ "model.layers.56.self_attn.k_proj", "model.layers.56.self_attn.v_proj" ], "candidates": [ { "dkld": -4.476467147469798e-05, "dbits": 10485760 }, { "dkld": 9.64555889367974e-05, "dbits": 20971520 }, { "dkld": 0.00012141317129134299, "dbits": 31457280 }, { "dkld": 0.0001040010713040801, "dbits": 52428800 } ] }, { "idx": 335, "layers": [ "model.layers.56.self_attn.o_proj" ], "candidates": [ { "dkld": 4.9475301057089194e-05, "dbits": 62914560 }, { "dkld": 0.00012911893427371146, "dbits": 125829120 }, { "dkld": 0.00012300387024878623, "dbits": 188743680 }, { "dkld": 8.985847234725675e-05, "dbits": 314572800 } ] }, { "idx": 336, "layers": [ "model.layers.56.mlp.shared_experts.gate_proj", "model.layers.56.mlp.shared_experts.up_proj", "model.layers.56.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00020344015210867206, "dbits": 23592960 }, { "dkld": -0.0001321256160736195, "dbits": 47185920 }, { "dkld": -0.0001406635157764069, "dbits": 70778880 }, { "dkld": -0.00013056164607405663, "dbits": 117964800 } ] }, { "idx": 337, "layers": [ "model.layers.56.mlp.experts.0.gate_proj", "model.layers.56.mlp.experts.1.gate_proj", "model.layers.56.mlp.experts.2.gate_proj", "model.layers.56.mlp.experts.3.gate_proj", "model.layers.56.mlp.experts.4.gate_proj", "model.layers.56.mlp.experts.5.gate_proj", "model.layers.56.mlp.experts.6.gate_proj", "model.layers.56.mlp.experts.7.gate_proj", "model.layers.56.mlp.experts.8.gate_proj", "model.layers.56.mlp.experts.9.gate_proj", "model.layers.56.mlp.experts.10.gate_proj", "model.layers.56.mlp.experts.11.gate_proj", "model.layers.56.mlp.experts.12.gate_proj", "model.layers.56.mlp.experts.13.gate_proj", "model.layers.56.mlp.experts.14.gate_proj", "model.layers.56.mlp.experts.15.gate_proj", "model.layers.56.mlp.experts.16.gate_proj", "model.layers.56.mlp.experts.17.gate_proj", "model.layers.56.mlp.experts.18.gate_proj", "model.layers.56.mlp.experts.19.gate_proj", "model.layers.56.mlp.experts.20.gate_proj", "model.layers.56.mlp.experts.21.gate_proj", "model.layers.56.mlp.experts.22.gate_proj", "model.layers.56.mlp.experts.23.gate_proj", "model.layers.56.mlp.experts.24.gate_proj", "model.layers.56.mlp.experts.25.gate_proj", "model.layers.56.mlp.experts.26.gate_proj", "model.layers.56.mlp.experts.27.gate_proj", "model.layers.56.mlp.experts.28.gate_proj", "model.layers.56.mlp.experts.29.gate_proj", "model.layers.56.mlp.experts.30.gate_proj", "model.layers.56.mlp.experts.31.gate_proj", "model.layers.56.mlp.experts.32.gate_proj", "model.layers.56.mlp.experts.33.gate_proj", "model.layers.56.mlp.experts.34.gate_proj", "model.layers.56.mlp.experts.35.gate_proj", "model.layers.56.mlp.experts.36.gate_proj", "model.layers.56.mlp.experts.37.gate_proj", "model.layers.56.mlp.experts.38.gate_proj", "model.layers.56.mlp.experts.39.gate_proj", "model.layers.56.mlp.experts.40.gate_proj", "model.layers.56.mlp.experts.41.gate_proj", "model.layers.56.mlp.experts.42.gate_proj", "model.layers.56.mlp.experts.43.gate_proj", "model.layers.56.mlp.experts.44.gate_proj", "model.layers.56.mlp.experts.45.gate_proj", "model.layers.56.mlp.experts.46.gate_proj", "model.layers.56.mlp.experts.47.gate_proj", "model.layers.56.mlp.experts.48.gate_proj", "model.layers.56.mlp.experts.49.gate_proj", "model.layers.56.mlp.experts.50.gate_proj", "model.layers.56.mlp.experts.51.gate_proj", "model.layers.56.mlp.experts.52.gate_proj", "model.layers.56.mlp.experts.53.gate_proj", "model.layers.56.mlp.experts.54.gate_proj", "model.layers.56.mlp.experts.55.gate_proj", "model.layers.56.mlp.experts.56.gate_proj", "model.layers.56.mlp.experts.57.gate_proj", "model.layers.56.mlp.experts.58.gate_proj", "model.layers.56.mlp.experts.59.gate_proj", "model.layers.56.mlp.experts.60.gate_proj", "model.layers.56.mlp.experts.61.gate_proj", "model.layers.56.mlp.experts.62.gate_proj", "model.layers.56.mlp.experts.63.gate_proj", "model.layers.56.mlp.experts.64.gate_proj", "model.layers.56.mlp.experts.65.gate_proj", "model.layers.56.mlp.experts.66.gate_proj", "model.layers.56.mlp.experts.67.gate_proj", "model.layers.56.mlp.experts.68.gate_proj", "model.layers.56.mlp.experts.69.gate_proj", "model.layers.56.mlp.experts.70.gate_proj", "model.layers.56.mlp.experts.71.gate_proj", "model.layers.56.mlp.experts.72.gate_proj", "model.layers.56.mlp.experts.73.gate_proj", "model.layers.56.mlp.experts.74.gate_proj", "model.layers.56.mlp.experts.75.gate_proj", "model.layers.56.mlp.experts.76.gate_proj", "model.layers.56.mlp.experts.77.gate_proj", "model.layers.56.mlp.experts.78.gate_proj", "model.layers.56.mlp.experts.79.gate_proj", "model.layers.56.mlp.experts.80.gate_proj", "model.layers.56.mlp.experts.81.gate_proj", "model.layers.56.mlp.experts.82.gate_proj", "model.layers.56.mlp.experts.83.gate_proj", "model.layers.56.mlp.experts.84.gate_proj", "model.layers.56.mlp.experts.85.gate_proj", "model.layers.56.mlp.experts.86.gate_proj", "model.layers.56.mlp.experts.87.gate_proj", "model.layers.56.mlp.experts.88.gate_proj", "model.layers.56.mlp.experts.89.gate_proj", "model.layers.56.mlp.experts.90.gate_proj", "model.layers.56.mlp.experts.91.gate_proj", "model.layers.56.mlp.experts.92.gate_proj", "model.layers.56.mlp.experts.93.gate_proj", "model.layers.56.mlp.experts.94.gate_proj", "model.layers.56.mlp.experts.95.gate_proj", "model.layers.56.mlp.experts.96.gate_proj", "model.layers.56.mlp.experts.97.gate_proj", "model.layers.56.mlp.experts.98.gate_proj", "model.layers.56.mlp.experts.99.gate_proj", "model.layers.56.mlp.experts.100.gate_proj", "model.layers.56.mlp.experts.101.gate_proj", "model.layers.56.mlp.experts.102.gate_proj", "model.layers.56.mlp.experts.103.gate_proj", "model.layers.56.mlp.experts.104.gate_proj", "model.layers.56.mlp.experts.105.gate_proj", "model.layers.56.mlp.experts.106.gate_proj", "model.layers.56.mlp.experts.107.gate_proj", "model.layers.56.mlp.experts.108.gate_proj", "model.layers.56.mlp.experts.109.gate_proj", "model.layers.56.mlp.experts.110.gate_proj", "model.layers.56.mlp.experts.111.gate_proj", "model.layers.56.mlp.experts.112.gate_proj", "model.layers.56.mlp.experts.113.gate_proj", "model.layers.56.mlp.experts.114.gate_proj", "model.layers.56.mlp.experts.115.gate_proj", "model.layers.56.mlp.experts.116.gate_proj", "model.layers.56.mlp.experts.117.gate_proj", "model.layers.56.mlp.experts.118.gate_proj", "model.layers.56.mlp.experts.119.gate_proj", "model.layers.56.mlp.experts.120.gate_proj", "model.layers.56.mlp.experts.121.gate_proj", "model.layers.56.mlp.experts.122.gate_proj", "model.layers.56.mlp.experts.123.gate_proj", "model.layers.56.mlp.experts.124.gate_proj", "model.layers.56.mlp.experts.125.gate_proj", "model.layers.56.mlp.experts.126.gate_proj", "model.layers.56.mlp.experts.127.gate_proj", "model.layers.56.mlp.experts.128.gate_proj", "model.layers.56.mlp.experts.129.gate_proj", "model.layers.56.mlp.experts.130.gate_proj", "model.layers.56.mlp.experts.131.gate_proj", "model.layers.56.mlp.experts.132.gate_proj", "model.layers.56.mlp.experts.133.gate_proj", "model.layers.56.mlp.experts.134.gate_proj", "model.layers.56.mlp.experts.135.gate_proj", "model.layers.56.mlp.experts.136.gate_proj", "model.layers.56.mlp.experts.137.gate_proj", "model.layers.56.mlp.experts.138.gate_proj", "model.layers.56.mlp.experts.139.gate_proj", "model.layers.56.mlp.experts.140.gate_proj", "model.layers.56.mlp.experts.141.gate_proj", "model.layers.56.mlp.experts.142.gate_proj", "model.layers.56.mlp.experts.143.gate_proj", "model.layers.56.mlp.experts.144.gate_proj", "model.layers.56.mlp.experts.145.gate_proj", "model.layers.56.mlp.experts.146.gate_proj", "model.layers.56.mlp.experts.147.gate_proj", "model.layers.56.mlp.experts.148.gate_proj", "model.layers.56.mlp.experts.149.gate_proj", "model.layers.56.mlp.experts.150.gate_proj", "model.layers.56.mlp.experts.151.gate_proj", "model.layers.56.mlp.experts.152.gate_proj", "model.layers.56.mlp.experts.153.gate_proj", "model.layers.56.mlp.experts.154.gate_proj", "model.layers.56.mlp.experts.155.gate_proj", "model.layers.56.mlp.experts.156.gate_proj", "model.layers.56.mlp.experts.157.gate_proj", "model.layers.56.mlp.experts.158.gate_proj", "model.layers.56.mlp.experts.159.gate_proj", "model.layers.56.mlp.experts.0.up_proj", "model.layers.56.mlp.experts.1.up_proj", "model.layers.56.mlp.experts.2.up_proj", "model.layers.56.mlp.experts.3.up_proj", "model.layers.56.mlp.experts.4.up_proj", "model.layers.56.mlp.experts.5.up_proj", "model.layers.56.mlp.experts.6.up_proj", "model.layers.56.mlp.experts.7.up_proj", "model.layers.56.mlp.experts.8.up_proj", "model.layers.56.mlp.experts.9.up_proj", "model.layers.56.mlp.experts.10.up_proj", "model.layers.56.mlp.experts.11.up_proj", "model.layers.56.mlp.experts.12.up_proj", "model.layers.56.mlp.experts.13.up_proj", "model.layers.56.mlp.experts.14.up_proj", "model.layers.56.mlp.experts.15.up_proj", "model.layers.56.mlp.experts.16.up_proj", "model.layers.56.mlp.experts.17.up_proj", "model.layers.56.mlp.experts.18.up_proj", "model.layers.56.mlp.experts.19.up_proj", "model.layers.56.mlp.experts.20.up_proj", "model.layers.56.mlp.experts.21.up_proj", "model.layers.56.mlp.experts.22.up_proj", "model.layers.56.mlp.experts.23.up_proj", "model.layers.56.mlp.experts.24.up_proj", "model.layers.56.mlp.experts.25.up_proj", "model.layers.56.mlp.experts.26.up_proj", "model.layers.56.mlp.experts.27.up_proj", "model.layers.56.mlp.experts.28.up_proj", "model.layers.56.mlp.experts.29.up_proj", "model.layers.56.mlp.experts.30.up_proj", "model.layers.56.mlp.experts.31.up_proj", "model.layers.56.mlp.experts.32.up_proj", "model.layers.56.mlp.experts.33.up_proj", "model.layers.56.mlp.experts.34.up_proj", "model.layers.56.mlp.experts.35.up_proj", "model.layers.56.mlp.experts.36.up_proj", "model.layers.56.mlp.experts.37.up_proj", "model.layers.56.mlp.experts.38.up_proj", "model.layers.56.mlp.experts.39.up_proj", "model.layers.56.mlp.experts.40.up_proj", "model.layers.56.mlp.experts.41.up_proj", "model.layers.56.mlp.experts.42.up_proj", "model.layers.56.mlp.experts.43.up_proj", "model.layers.56.mlp.experts.44.up_proj", "model.layers.56.mlp.experts.45.up_proj", "model.layers.56.mlp.experts.46.up_proj", "model.layers.56.mlp.experts.47.up_proj", "model.layers.56.mlp.experts.48.up_proj", "model.layers.56.mlp.experts.49.up_proj", "model.layers.56.mlp.experts.50.up_proj", "model.layers.56.mlp.experts.51.up_proj", "model.layers.56.mlp.experts.52.up_proj", "model.layers.56.mlp.experts.53.up_proj", "model.layers.56.mlp.experts.54.up_proj", "model.layers.56.mlp.experts.55.up_proj", "model.layers.56.mlp.experts.56.up_proj", "model.layers.56.mlp.experts.57.up_proj", "model.layers.56.mlp.experts.58.up_proj", "model.layers.56.mlp.experts.59.up_proj", "model.layers.56.mlp.experts.60.up_proj", "model.layers.56.mlp.experts.61.up_proj", "model.layers.56.mlp.experts.62.up_proj", "model.layers.56.mlp.experts.63.up_proj", "model.layers.56.mlp.experts.64.up_proj", "model.layers.56.mlp.experts.65.up_proj", "model.layers.56.mlp.experts.66.up_proj", "model.layers.56.mlp.experts.67.up_proj", "model.layers.56.mlp.experts.68.up_proj", "model.layers.56.mlp.experts.69.up_proj", "model.layers.56.mlp.experts.70.up_proj", "model.layers.56.mlp.experts.71.up_proj", "model.layers.56.mlp.experts.72.up_proj", "model.layers.56.mlp.experts.73.up_proj", "model.layers.56.mlp.experts.74.up_proj", "model.layers.56.mlp.experts.75.up_proj", "model.layers.56.mlp.experts.76.up_proj", "model.layers.56.mlp.experts.77.up_proj", "model.layers.56.mlp.experts.78.up_proj", "model.layers.56.mlp.experts.79.up_proj", "model.layers.56.mlp.experts.80.up_proj", "model.layers.56.mlp.experts.81.up_proj", "model.layers.56.mlp.experts.82.up_proj", "model.layers.56.mlp.experts.83.up_proj", "model.layers.56.mlp.experts.84.up_proj", "model.layers.56.mlp.experts.85.up_proj", "model.layers.56.mlp.experts.86.up_proj", "model.layers.56.mlp.experts.87.up_proj", "model.layers.56.mlp.experts.88.up_proj", "model.layers.56.mlp.experts.89.up_proj", "model.layers.56.mlp.experts.90.up_proj", "model.layers.56.mlp.experts.91.up_proj", "model.layers.56.mlp.experts.92.up_proj", "model.layers.56.mlp.experts.93.up_proj", "model.layers.56.mlp.experts.94.up_proj", "model.layers.56.mlp.experts.95.up_proj", "model.layers.56.mlp.experts.96.up_proj", "model.layers.56.mlp.experts.97.up_proj", "model.layers.56.mlp.experts.98.up_proj", "model.layers.56.mlp.experts.99.up_proj", "model.layers.56.mlp.experts.100.up_proj", "model.layers.56.mlp.experts.101.up_proj", "model.layers.56.mlp.experts.102.up_proj", "model.layers.56.mlp.experts.103.up_proj", "model.layers.56.mlp.experts.104.up_proj", "model.layers.56.mlp.experts.105.up_proj", "model.layers.56.mlp.experts.106.up_proj", "model.layers.56.mlp.experts.107.up_proj", "model.layers.56.mlp.experts.108.up_proj", "model.layers.56.mlp.experts.109.up_proj", "model.layers.56.mlp.experts.110.up_proj", "model.layers.56.mlp.experts.111.up_proj", "model.layers.56.mlp.experts.112.up_proj", "model.layers.56.mlp.experts.113.up_proj", "model.layers.56.mlp.experts.114.up_proj", "model.layers.56.mlp.experts.115.up_proj", "model.layers.56.mlp.experts.116.up_proj", "model.layers.56.mlp.experts.117.up_proj", "model.layers.56.mlp.experts.118.up_proj", "model.layers.56.mlp.experts.119.up_proj", "model.layers.56.mlp.experts.120.up_proj", "model.layers.56.mlp.experts.121.up_proj", "model.layers.56.mlp.experts.122.up_proj", "model.layers.56.mlp.experts.123.up_proj", "model.layers.56.mlp.experts.124.up_proj", "model.layers.56.mlp.experts.125.up_proj", "model.layers.56.mlp.experts.126.up_proj", "model.layers.56.mlp.experts.127.up_proj", "model.layers.56.mlp.experts.128.up_proj", "model.layers.56.mlp.experts.129.up_proj", "model.layers.56.mlp.experts.130.up_proj", "model.layers.56.mlp.experts.131.up_proj", "model.layers.56.mlp.experts.132.up_proj", "model.layers.56.mlp.experts.133.up_proj", "model.layers.56.mlp.experts.134.up_proj", "model.layers.56.mlp.experts.135.up_proj", "model.layers.56.mlp.experts.136.up_proj", "model.layers.56.mlp.experts.137.up_proj", "model.layers.56.mlp.experts.138.up_proj", "model.layers.56.mlp.experts.139.up_proj", "model.layers.56.mlp.experts.140.up_proj", "model.layers.56.mlp.experts.141.up_proj", "model.layers.56.mlp.experts.142.up_proj", "model.layers.56.mlp.experts.143.up_proj", "model.layers.56.mlp.experts.144.up_proj", "model.layers.56.mlp.experts.145.up_proj", "model.layers.56.mlp.experts.146.up_proj", "model.layers.56.mlp.experts.147.up_proj", "model.layers.56.mlp.experts.148.up_proj", "model.layers.56.mlp.experts.149.up_proj", "model.layers.56.mlp.experts.150.up_proj", "model.layers.56.mlp.experts.151.up_proj", "model.layers.56.mlp.experts.152.up_proj", "model.layers.56.mlp.experts.153.up_proj", "model.layers.56.mlp.experts.154.up_proj", "model.layers.56.mlp.experts.155.up_proj", "model.layers.56.mlp.experts.156.up_proj", "model.layers.56.mlp.experts.157.up_proj", "model.layers.56.mlp.experts.158.up_proj", "model.layers.56.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00016905805096030513, "dbits": 2516582400 }, { "dkld": -0.00018100971356034556, "dbits": 5033164800 }, { "dkld": -0.0002169745042920196, "dbits": 7549747200 }, { "dkld": -0.0001881601288914736, "dbits": 12582912000 } ] }, { "idx": 338, "layers": [ "model.layers.56.mlp.experts.0.down_proj", "model.layers.56.mlp.experts.1.down_proj", "model.layers.56.mlp.experts.2.down_proj", "model.layers.56.mlp.experts.3.down_proj", "model.layers.56.mlp.experts.4.down_proj", "model.layers.56.mlp.experts.5.down_proj", "model.layers.56.mlp.experts.6.down_proj", "model.layers.56.mlp.experts.7.down_proj", "model.layers.56.mlp.experts.8.down_proj", "model.layers.56.mlp.experts.9.down_proj", "model.layers.56.mlp.experts.10.down_proj", "model.layers.56.mlp.experts.11.down_proj", "model.layers.56.mlp.experts.12.down_proj", "model.layers.56.mlp.experts.13.down_proj", "model.layers.56.mlp.experts.14.down_proj", "model.layers.56.mlp.experts.15.down_proj", "model.layers.56.mlp.experts.16.down_proj", "model.layers.56.mlp.experts.17.down_proj", "model.layers.56.mlp.experts.18.down_proj", "model.layers.56.mlp.experts.19.down_proj", "model.layers.56.mlp.experts.20.down_proj", "model.layers.56.mlp.experts.21.down_proj", "model.layers.56.mlp.experts.22.down_proj", "model.layers.56.mlp.experts.23.down_proj", "model.layers.56.mlp.experts.24.down_proj", "model.layers.56.mlp.experts.25.down_proj", "model.layers.56.mlp.experts.26.down_proj", "model.layers.56.mlp.experts.27.down_proj", "model.layers.56.mlp.experts.28.down_proj", "model.layers.56.mlp.experts.29.down_proj", "model.layers.56.mlp.experts.30.down_proj", "model.layers.56.mlp.experts.31.down_proj", "model.layers.56.mlp.experts.32.down_proj", "model.layers.56.mlp.experts.33.down_proj", "model.layers.56.mlp.experts.34.down_proj", "model.layers.56.mlp.experts.35.down_proj", "model.layers.56.mlp.experts.36.down_proj", "model.layers.56.mlp.experts.37.down_proj", "model.layers.56.mlp.experts.38.down_proj", "model.layers.56.mlp.experts.39.down_proj", "model.layers.56.mlp.experts.40.down_proj", "model.layers.56.mlp.experts.41.down_proj", "model.layers.56.mlp.experts.42.down_proj", "model.layers.56.mlp.experts.43.down_proj", "model.layers.56.mlp.experts.44.down_proj", "model.layers.56.mlp.experts.45.down_proj", "model.layers.56.mlp.experts.46.down_proj", "model.layers.56.mlp.experts.47.down_proj", "model.layers.56.mlp.experts.48.down_proj", "model.layers.56.mlp.experts.49.down_proj", "model.layers.56.mlp.experts.50.down_proj", "model.layers.56.mlp.experts.51.down_proj", "model.layers.56.mlp.experts.52.down_proj", "model.layers.56.mlp.experts.53.down_proj", "model.layers.56.mlp.experts.54.down_proj", "model.layers.56.mlp.experts.55.down_proj", "model.layers.56.mlp.experts.56.down_proj", "model.layers.56.mlp.experts.57.down_proj", "model.layers.56.mlp.experts.58.down_proj", "model.layers.56.mlp.experts.59.down_proj", "model.layers.56.mlp.experts.60.down_proj", "model.layers.56.mlp.experts.61.down_proj", "model.layers.56.mlp.experts.62.down_proj", "model.layers.56.mlp.experts.63.down_proj", "model.layers.56.mlp.experts.64.down_proj", "model.layers.56.mlp.experts.65.down_proj", "model.layers.56.mlp.experts.66.down_proj", "model.layers.56.mlp.experts.67.down_proj", "model.layers.56.mlp.experts.68.down_proj", "model.layers.56.mlp.experts.69.down_proj", "model.layers.56.mlp.experts.70.down_proj", "model.layers.56.mlp.experts.71.down_proj", "model.layers.56.mlp.experts.72.down_proj", "model.layers.56.mlp.experts.73.down_proj", "model.layers.56.mlp.experts.74.down_proj", "model.layers.56.mlp.experts.75.down_proj", "model.layers.56.mlp.experts.76.down_proj", "model.layers.56.mlp.experts.77.down_proj", "model.layers.56.mlp.experts.78.down_proj", "model.layers.56.mlp.experts.79.down_proj", "model.layers.56.mlp.experts.80.down_proj", "model.layers.56.mlp.experts.81.down_proj", "model.layers.56.mlp.experts.82.down_proj", "model.layers.56.mlp.experts.83.down_proj", "model.layers.56.mlp.experts.84.down_proj", "model.layers.56.mlp.experts.85.down_proj", "model.layers.56.mlp.experts.86.down_proj", "model.layers.56.mlp.experts.87.down_proj", "model.layers.56.mlp.experts.88.down_proj", "model.layers.56.mlp.experts.89.down_proj", "model.layers.56.mlp.experts.90.down_proj", "model.layers.56.mlp.experts.91.down_proj", "model.layers.56.mlp.experts.92.down_proj", "model.layers.56.mlp.experts.93.down_proj", "model.layers.56.mlp.experts.94.down_proj", "model.layers.56.mlp.experts.95.down_proj", "model.layers.56.mlp.experts.96.down_proj", "model.layers.56.mlp.experts.97.down_proj", "model.layers.56.mlp.experts.98.down_proj", "model.layers.56.mlp.experts.99.down_proj", "model.layers.56.mlp.experts.100.down_proj", "model.layers.56.mlp.experts.101.down_proj", "model.layers.56.mlp.experts.102.down_proj", "model.layers.56.mlp.experts.103.down_proj", "model.layers.56.mlp.experts.104.down_proj", "model.layers.56.mlp.experts.105.down_proj", "model.layers.56.mlp.experts.106.down_proj", "model.layers.56.mlp.experts.107.down_proj", "model.layers.56.mlp.experts.108.down_proj", "model.layers.56.mlp.experts.109.down_proj", "model.layers.56.mlp.experts.110.down_proj", "model.layers.56.mlp.experts.111.down_proj", "model.layers.56.mlp.experts.112.down_proj", "model.layers.56.mlp.experts.113.down_proj", "model.layers.56.mlp.experts.114.down_proj", "model.layers.56.mlp.experts.115.down_proj", "model.layers.56.mlp.experts.116.down_proj", "model.layers.56.mlp.experts.117.down_proj", "model.layers.56.mlp.experts.118.down_proj", "model.layers.56.mlp.experts.119.down_proj", "model.layers.56.mlp.experts.120.down_proj", "model.layers.56.mlp.experts.121.down_proj", "model.layers.56.mlp.experts.122.down_proj", "model.layers.56.mlp.experts.123.down_proj", "model.layers.56.mlp.experts.124.down_proj", "model.layers.56.mlp.experts.125.down_proj", "model.layers.56.mlp.experts.126.down_proj", "model.layers.56.mlp.experts.127.down_proj", "model.layers.56.mlp.experts.128.down_proj", "model.layers.56.mlp.experts.129.down_proj", "model.layers.56.mlp.experts.130.down_proj", "model.layers.56.mlp.experts.131.down_proj", "model.layers.56.mlp.experts.132.down_proj", "model.layers.56.mlp.experts.133.down_proj", "model.layers.56.mlp.experts.134.down_proj", "model.layers.56.mlp.experts.135.down_proj", "model.layers.56.mlp.experts.136.down_proj", "model.layers.56.mlp.experts.137.down_proj", "model.layers.56.mlp.experts.138.down_proj", "model.layers.56.mlp.experts.139.down_proj", "model.layers.56.mlp.experts.140.down_proj", "model.layers.56.mlp.experts.141.down_proj", "model.layers.56.mlp.experts.142.down_proj", "model.layers.56.mlp.experts.143.down_proj", "model.layers.56.mlp.experts.144.down_proj", "model.layers.56.mlp.experts.145.down_proj", "model.layers.56.mlp.experts.146.down_proj", "model.layers.56.mlp.experts.147.down_proj", "model.layers.56.mlp.experts.148.down_proj", "model.layers.56.mlp.experts.149.down_proj", "model.layers.56.mlp.experts.150.down_proj", "model.layers.56.mlp.experts.151.down_proj", "model.layers.56.mlp.experts.152.down_proj", "model.layers.56.mlp.experts.153.down_proj", "model.layers.56.mlp.experts.154.down_proj", "model.layers.56.mlp.experts.155.down_proj", "model.layers.56.mlp.experts.156.down_proj", "model.layers.56.mlp.experts.157.down_proj", "model.layers.56.mlp.experts.158.down_proj", "model.layers.56.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00014618309214711744, "dbits": 1258291200 }, { "dkld": -0.0001777155324816787, "dbits": 2516582400 }, { "dkld": -0.0001664021983742714, "dbits": 3774873600 }, { "dkld": -0.0001405728980898857, "dbits": 6291456000 } ] }, { "idx": 339, "layers": [ "model.layers.57.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00017418945208191872, "dbits": 62914560 }, { "dkld": -0.0001977515406906688, "dbits": 125829120 }, { "dkld": -0.0001887682825326975, "dbits": 188743680 }, { "dkld": -0.00018093148246407786, "dbits": 314572800 } ] }, { "idx": 340, "layers": [ "model.layers.57.self_attn.k_proj", "model.layers.57.self_attn.v_proj" ], "candidates": [ { "dkld": 6.581367924808623e-05, "dbits": 10485760 }, { "dkld": 4.0786340832710266e-05, "dbits": 20971520 }, { "dkld": 2.111867070198059e-05, "dbits": 31457280 }, { "dkld": 6.04875385761261e-05, "dbits": 52428800 } ] }, { "idx": 341, "layers": [ "model.layers.57.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00015777945518494207, "dbits": 62914560 }, { "dkld": -5.086036399007399e-05, "dbits": 125829120 }, { "dkld": -9.4560906291008e-05, "dbits": 188743680 }, { "dkld": -0.00010561970993877567, "dbits": 314572800 } ] }, { "idx": 342, "layers": [ "model.layers.57.mlp.shared_experts.gate_proj", "model.layers.57.mlp.shared_experts.up_proj", "model.layers.57.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.000181792490184296, "dbits": 23592960 }, { "dkld": 0.0003168288618326104, "dbits": 47185920 }, { "dkld": 0.0002758899703621809, "dbits": 70778880 }, { "dkld": 0.00028397990390657823, "dbits": 117964800 } ] }, { "idx": 343, "layers": [ "model.layers.57.mlp.experts.0.gate_proj", "model.layers.57.mlp.experts.1.gate_proj", "model.layers.57.mlp.experts.2.gate_proj", "model.layers.57.mlp.experts.3.gate_proj", "model.layers.57.mlp.experts.4.gate_proj", "model.layers.57.mlp.experts.5.gate_proj", "model.layers.57.mlp.experts.6.gate_proj", "model.layers.57.mlp.experts.7.gate_proj", "model.layers.57.mlp.experts.8.gate_proj", "model.layers.57.mlp.experts.9.gate_proj", "model.layers.57.mlp.experts.10.gate_proj", "model.layers.57.mlp.experts.11.gate_proj", "model.layers.57.mlp.experts.12.gate_proj", "model.layers.57.mlp.experts.13.gate_proj", "model.layers.57.mlp.experts.14.gate_proj", "model.layers.57.mlp.experts.15.gate_proj", "model.layers.57.mlp.experts.16.gate_proj", "model.layers.57.mlp.experts.17.gate_proj", "model.layers.57.mlp.experts.18.gate_proj", "model.layers.57.mlp.experts.19.gate_proj", "model.layers.57.mlp.experts.20.gate_proj", "model.layers.57.mlp.experts.21.gate_proj", "model.layers.57.mlp.experts.22.gate_proj", "model.layers.57.mlp.experts.23.gate_proj", "model.layers.57.mlp.experts.24.gate_proj", "model.layers.57.mlp.experts.25.gate_proj", "model.layers.57.mlp.experts.26.gate_proj", "model.layers.57.mlp.experts.27.gate_proj", "model.layers.57.mlp.experts.28.gate_proj", "model.layers.57.mlp.experts.29.gate_proj", "model.layers.57.mlp.experts.30.gate_proj", "model.layers.57.mlp.experts.31.gate_proj", "model.layers.57.mlp.experts.32.gate_proj", "model.layers.57.mlp.experts.33.gate_proj", "model.layers.57.mlp.experts.34.gate_proj", "model.layers.57.mlp.experts.35.gate_proj", "model.layers.57.mlp.experts.36.gate_proj", "model.layers.57.mlp.experts.37.gate_proj", "model.layers.57.mlp.experts.38.gate_proj", "model.layers.57.mlp.experts.39.gate_proj", "model.layers.57.mlp.experts.40.gate_proj", "model.layers.57.mlp.experts.41.gate_proj", "model.layers.57.mlp.experts.42.gate_proj", "model.layers.57.mlp.experts.43.gate_proj", "model.layers.57.mlp.experts.44.gate_proj", "model.layers.57.mlp.experts.45.gate_proj", "model.layers.57.mlp.experts.46.gate_proj", "model.layers.57.mlp.experts.47.gate_proj", "model.layers.57.mlp.experts.48.gate_proj", "model.layers.57.mlp.experts.49.gate_proj", "model.layers.57.mlp.experts.50.gate_proj", "model.layers.57.mlp.experts.51.gate_proj", "model.layers.57.mlp.experts.52.gate_proj", "model.layers.57.mlp.experts.53.gate_proj", "model.layers.57.mlp.experts.54.gate_proj", "model.layers.57.mlp.experts.55.gate_proj", "model.layers.57.mlp.experts.56.gate_proj", "model.layers.57.mlp.experts.57.gate_proj", "model.layers.57.mlp.experts.58.gate_proj", "model.layers.57.mlp.experts.59.gate_proj", "model.layers.57.mlp.experts.60.gate_proj", "model.layers.57.mlp.experts.61.gate_proj", "model.layers.57.mlp.experts.62.gate_proj", "model.layers.57.mlp.experts.63.gate_proj", "model.layers.57.mlp.experts.64.gate_proj", "model.layers.57.mlp.experts.65.gate_proj", "model.layers.57.mlp.experts.66.gate_proj", "model.layers.57.mlp.experts.67.gate_proj", "model.layers.57.mlp.experts.68.gate_proj", "model.layers.57.mlp.experts.69.gate_proj", "model.layers.57.mlp.experts.70.gate_proj", "model.layers.57.mlp.experts.71.gate_proj", "model.layers.57.mlp.experts.72.gate_proj", "model.layers.57.mlp.experts.73.gate_proj", "model.layers.57.mlp.experts.74.gate_proj", "model.layers.57.mlp.experts.75.gate_proj", "model.layers.57.mlp.experts.76.gate_proj", "model.layers.57.mlp.experts.77.gate_proj", "model.layers.57.mlp.experts.78.gate_proj", "model.layers.57.mlp.experts.79.gate_proj", "model.layers.57.mlp.experts.80.gate_proj", "model.layers.57.mlp.experts.81.gate_proj", "model.layers.57.mlp.experts.82.gate_proj", "model.layers.57.mlp.experts.83.gate_proj", "model.layers.57.mlp.experts.84.gate_proj", "model.layers.57.mlp.experts.85.gate_proj", "model.layers.57.mlp.experts.86.gate_proj", "model.layers.57.mlp.experts.87.gate_proj", "model.layers.57.mlp.experts.88.gate_proj", "model.layers.57.mlp.experts.89.gate_proj", "model.layers.57.mlp.experts.90.gate_proj", "model.layers.57.mlp.experts.91.gate_proj", "model.layers.57.mlp.experts.92.gate_proj", "model.layers.57.mlp.experts.93.gate_proj", "model.layers.57.mlp.experts.94.gate_proj", "model.layers.57.mlp.experts.95.gate_proj", "model.layers.57.mlp.experts.96.gate_proj", "model.layers.57.mlp.experts.97.gate_proj", "model.layers.57.mlp.experts.98.gate_proj", "model.layers.57.mlp.experts.99.gate_proj", "model.layers.57.mlp.experts.100.gate_proj", "model.layers.57.mlp.experts.101.gate_proj", "model.layers.57.mlp.experts.102.gate_proj", "model.layers.57.mlp.experts.103.gate_proj", "model.layers.57.mlp.experts.104.gate_proj", "model.layers.57.mlp.experts.105.gate_proj", "model.layers.57.mlp.experts.106.gate_proj", "model.layers.57.mlp.experts.107.gate_proj", "model.layers.57.mlp.experts.108.gate_proj", "model.layers.57.mlp.experts.109.gate_proj", "model.layers.57.mlp.experts.110.gate_proj", "model.layers.57.mlp.experts.111.gate_proj", "model.layers.57.mlp.experts.112.gate_proj", "model.layers.57.mlp.experts.113.gate_proj", "model.layers.57.mlp.experts.114.gate_proj", "model.layers.57.mlp.experts.115.gate_proj", "model.layers.57.mlp.experts.116.gate_proj", "model.layers.57.mlp.experts.117.gate_proj", "model.layers.57.mlp.experts.118.gate_proj", "model.layers.57.mlp.experts.119.gate_proj", "model.layers.57.mlp.experts.120.gate_proj", "model.layers.57.mlp.experts.121.gate_proj", "model.layers.57.mlp.experts.122.gate_proj", "model.layers.57.mlp.experts.123.gate_proj", "model.layers.57.mlp.experts.124.gate_proj", "model.layers.57.mlp.experts.125.gate_proj", "model.layers.57.mlp.experts.126.gate_proj", "model.layers.57.mlp.experts.127.gate_proj", "model.layers.57.mlp.experts.128.gate_proj", "model.layers.57.mlp.experts.129.gate_proj", "model.layers.57.mlp.experts.130.gate_proj", "model.layers.57.mlp.experts.131.gate_proj", "model.layers.57.mlp.experts.132.gate_proj", "model.layers.57.mlp.experts.133.gate_proj", "model.layers.57.mlp.experts.134.gate_proj", "model.layers.57.mlp.experts.135.gate_proj", "model.layers.57.mlp.experts.136.gate_proj", "model.layers.57.mlp.experts.137.gate_proj", "model.layers.57.mlp.experts.138.gate_proj", "model.layers.57.mlp.experts.139.gate_proj", "model.layers.57.mlp.experts.140.gate_proj", "model.layers.57.mlp.experts.141.gate_proj", "model.layers.57.mlp.experts.142.gate_proj", "model.layers.57.mlp.experts.143.gate_proj", "model.layers.57.mlp.experts.144.gate_proj", "model.layers.57.mlp.experts.145.gate_proj", "model.layers.57.mlp.experts.146.gate_proj", "model.layers.57.mlp.experts.147.gate_proj", "model.layers.57.mlp.experts.148.gate_proj", "model.layers.57.mlp.experts.149.gate_proj", "model.layers.57.mlp.experts.150.gate_proj", "model.layers.57.mlp.experts.151.gate_proj", "model.layers.57.mlp.experts.152.gate_proj", "model.layers.57.mlp.experts.153.gate_proj", "model.layers.57.mlp.experts.154.gate_proj", "model.layers.57.mlp.experts.155.gate_proj", "model.layers.57.mlp.experts.156.gate_proj", "model.layers.57.mlp.experts.157.gate_proj", "model.layers.57.mlp.experts.158.gate_proj", "model.layers.57.mlp.experts.159.gate_proj", "model.layers.57.mlp.experts.0.up_proj", "model.layers.57.mlp.experts.1.up_proj", "model.layers.57.mlp.experts.2.up_proj", "model.layers.57.mlp.experts.3.up_proj", "model.layers.57.mlp.experts.4.up_proj", "model.layers.57.mlp.experts.5.up_proj", "model.layers.57.mlp.experts.6.up_proj", "model.layers.57.mlp.experts.7.up_proj", "model.layers.57.mlp.experts.8.up_proj", "model.layers.57.mlp.experts.9.up_proj", "model.layers.57.mlp.experts.10.up_proj", "model.layers.57.mlp.experts.11.up_proj", "model.layers.57.mlp.experts.12.up_proj", "model.layers.57.mlp.experts.13.up_proj", "model.layers.57.mlp.experts.14.up_proj", "model.layers.57.mlp.experts.15.up_proj", "model.layers.57.mlp.experts.16.up_proj", "model.layers.57.mlp.experts.17.up_proj", "model.layers.57.mlp.experts.18.up_proj", "model.layers.57.mlp.experts.19.up_proj", "model.layers.57.mlp.experts.20.up_proj", "model.layers.57.mlp.experts.21.up_proj", "model.layers.57.mlp.experts.22.up_proj", "model.layers.57.mlp.experts.23.up_proj", "model.layers.57.mlp.experts.24.up_proj", "model.layers.57.mlp.experts.25.up_proj", "model.layers.57.mlp.experts.26.up_proj", "model.layers.57.mlp.experts.27.up_proj", "model.layers.57.mlp.experts.28.up_proj", "model.layers.57.mlp.experts.29.up_proj", "model.layers.57.mlp.experts.30.up_proj", "model.layers.57.mlp.experts.31.up_proj", "model.layers.57.mlp.experts.32.up_proj", "model.layers.57.mlp.experts.33.up_proj", "model.layers.57.mlp.experts.34.up_proj", "model.layers.57.mlp.experts.35.up_proj", "model.layers.57.mlp.experts.36.up_proj", "model.layers.57.mlp.experts.37.up_proj", "model.layers.57.mlp.experts.38.up_proj", "model.layers.57.mlp.experts.39.up_proj", "model.layers.57.mlp.experts.40.up_proj", "model.layers.57.mlp.experts.41.up_proj", "model.layers.57.mlp.experts.42.up_proj", "model.layers.57.mlp.experts.43.up_proj", "model.layers.57.mlp.experts.44.up_proj", "model.layers.57.mlp.experts.45.up_proj", "model.layers.57.mlp.experts.46.up_proj", "model.layers.57.mlp.experts.47.up_proj", "model.layers.57.mlp.experts.48.up_proj", "model.layers.57.mlp.experts.49.up_proj", "model.layers.57.mlp.experts.50.up_proj", "model.layers.57.mlp.experts.51.up_proj", "model.layers.57.mlp.experts.52.up_proj", "model.layers.57.mlp.experts.53.up_proj", "model.layers.57.mlp.experts.54.up_proj", "model.layers.57.mlp.experts.55.up_proj", "model.layers.57.mlp.experts.56.up_proj", "model.layers.57.mlp.experts.57.up_proj", "model.layers.57.mlp.experts.58.up_proj", "model.layers.57.mlp.experts.59.up_proj", "model.layers.57.mlp.experts.60.up_proj", "model.layers.57.mlp.experts.61.up_proj", "model.layers.57.mlp.experts.62.up_proj", "model.layers.57.mlp.experts.63.up_proj", "model.layers.57.mlp.experts.64.up_proj", "model.layers.57.mlp.experts.65.up_proj", "model.layers.57.mlp.experts.66.up_proj", "model.layers.57.mlp.experts.67.up_proj", "model.layers.57.mlp.experts.68.up_proj", "model.layers.57.mlp.experts.69.up_proj", "model.layers.57.mlp.experts.70.up_proj", "model.layers.57.mlp.experts.71.up_proj", "model.layers.57.mlp.experts.72.up_proj", "model.layers.57.mlp.experts.73.up_proj", "model.layers.57.mlp.experts.74.up_proj", "model.layers.57.mlp.experts.75.up_proj", "model.layers.57.mlp.experts.76.up_proj", "model.layers.57.mlp.experts.77.up_proj", "model.layers.57.mlp.experts.78.up_proj", "model.layers.57.mlp.experts.79.up_proj", "model.layers.57.mlp.experts.80.up_proj", "model.layers.57.mlp.experts.81.up_proj", "model.layers.57.mlp.experts.82.up_proj", "model.layers.57.mlp.experts.83.up_proj", "model.layers.57.mlp.experts.84.up_proj", "model.layers.57.mlp.experts.85.up_proj", "model.layers.57.mlp.experts.86.up_proj", "model.layers.57.mlp.experts.87.up_proj", "model.layers.57.mlp.experts.88.up_proj", "model.layers.57.mlp.experts.89.up_proj", "model.layers.57.mlp.experts.90.up_proj", "model.layers.57.mlp.experts.91.up_proj", "model.layers.57.mlp.experts.92.up_proj", "model.layers.57.mlp.experts.93.up_proj", "model.layers.57.mlp.experts.94.up_proj", "model.layers.57.mlp.experts.95.up_proj", "model.layers.57.mlp.experts.96.up_proj", "model.layers.57.mlp.experts.97.up_proj", "model.layers.57.mlp.experts.98.up_proj", "model.layers.57.mlp.experts.99.up_proj", "model.layers.57.mlp.experts.100.up_proj", "model.layers.57.mlp.experts.101.up_proj", "model.layers.57.mlp.experts.102.up_proj", "model.layers.57.mlp.experts.103.up_proj", "model.layers.57.mlp.experts.104.up_proj", "model.layers.57.mlp.experts.105.up_proj", "model.layers.57.mlp.experts.106.up_proj", "model.layers.57.mlp.experts.107.up_proj", "model.layers.57.mlp.experts.108.up_proj", "model.layers.57.mlp.experts.109.up_proj", "model.layers.57.mlp.experts.110.up_proj", "model.layers.57.mlp.experts.111.up_proj", "model.layers.57.mlp.experts.112.up_proj", "model.layers.57.mlp.experts.113.up_proj", "model.layers.57.mlp.experts.114.up_proj", "model.layers.57.mlp.experts.115.up_proj", "model.layers.57.mlp.experts.116.up_proj", "model.layers.57.mlp.experts.117.up_proj", "model.layers.57.mlp.experts.118.up_proj", "model.layers.57.mlp.experts.119.up_proj", "model.layers.57.mlp.experts.120.up_proj", "model.layers.57.mlp.experts.121.up_proj", "model.layers.57.mlp.experts.122.up_proj", "model.layers.57.mlp.experts.123.up_proj", "model.layers.57.mlp.experts.124.up_proj", "model.layers.57.mlp.experts.125.up_proj", "model.layers.57.mlp.experts.126.up_proj", "model.layers.57.mlp.experts.127.up_proj", "model.layers.57.mlp.experts.128.up_proj", "model.layers.57.mlp.experts.129.up_proj", "model.layers.57.mlp.experts.130.up_proj", "model.layers.57.mlp.experts.131.up_proj", "model.layers.57.mlp.experts.132.up_proj", "model.layers.57.mlp.experts.133.up_proj", "model.layers.57.mlp.experts.134.up_proj", "model.layers.57.mlp.experts.135.up_proj", "model.layers.57.mlp.experts.136.up_proj", "model.layers.57.mlp.experts.137.up_proj", "model.layers.57.mlp.experts.138.up_proj", "model.layers.57.mlp.experts.139.up_proj", "model.layers.57.mlp.experts.140.up_proj", "model.layers.57.mlp.experts.141.up_proj", "model.layers.57.mlp.experts.142.up_proj", "model.layers.57.mlp.experts.143.up_proj", "model.layers.57.mlp.experts.144.up_proj", "model.layers.57.mlp.experts.145.up_proj", "model.layers.57.mlp.experts.146.up_proj", "model.layers.57.mlp.experts.147.up_proj", "model.layers.57.mlp.experts.148.up_proj", "model.layers.57.mlp.experts.149.up_proj", "model.layers.57.mlp.experts.150.up_proj", "model.layers.57.mlp.experts.151.up_proj", "model.layers.57.mlp.experts.152.up_proj", "model.layers.57.mlp.experts.153.up_proj", "model.layers.57.mlp.experts.154.up_proj", "model.layers.57.mlp.experts.155.up_proj", "model.layers.57.mlp.experts.156.up_proj", "model.layers.57.mlp.experts.157.up_proj", "model.layers.57.mlp.experts.158.up_proj", "model.layers.57.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0001827747561037596, "dbits": 2516582400 }, { "dkld": -0.00015212213620544035, "dbits": 5033164800 }, { "dkld": -0.0001380929723382024, "dbits": 7549747200 }, { "dkld": -0.00016897581517696936, "dbits": 12582912000 } ] }, { "idx": 344, "layers": [ "model.layers.57.mlp.experts.0.down_proj", "model.layers.57.mlp.experts.1.down_proj", "model.layers.57.mlp.experts.2.down_proj", "model.layers.57.mlp.experts.3.down_proj", "model.layers.57.mlp.experts.4.down_proj", "model.layers.57.mlp.experts.5.down_proj", "model.layers.57.mlp.experts.6.down_proj", "model.layers.57.mlp.experts.7.down_proj", "model.layers.57.mlp.experts.8.down_proj", "model.layers.57.mlp.experts.9.down_proj", "model.layers.57.mlp.experts.10.down_proj", "model.layers.57.mlp.experts.11.down_proj", "model.layers.57.mlp.experts.12.down_proj", "model.layers.57.mlp.experts.13.down_proj", "model.layers.57.mlp.experts.14.down_proj", "model.layers.57.mlp.experts.15.down_proj", "model.layers.57.mlp.experts.16.down_proj", "model.layers.57.mlp.experts.17.down_proj", "model.layers.57.mlp.experts.18.down_proj", "model.layers.57.mlp.experts.19.down_proj", "model.layers.57.mlp.experts.20.down_proj", "model.layers.57.mlp.experts.21.down_proj", "model.layers.57.mlp.experts.22.down_proj", "model.layers.57.mlp.experts.23.down_proj", "model.layers.57.mlp.experts.24.down_proj", "model.layers.57.mlp.experts.25.down_proj", "model.layers.57.mlp.experts.26.down_proj", "model.layers.57.mlp.experts.27.down_proj", "model.layers.57.mlp.experts.28.down_proj", "model.layers.57.mlp.experts.29.down_proj", "model.layers.57.mlp.experts.30.down_proj", "model.layers.57.mlp.experts.31.down_proj", "model.layers.57.mlp.experts.32.down_proj", "model.layers.57.mlp.experts.33.down_proj", "model.layers.57.mlp.experts.34.down_proj", "model.layers.57.mlp.experts.35.down_proj", "model.layers.57.mlp.experts.36.down_proj", "model.layers.57.mlp.experts.37.down_proj", "model.layers.57.mlp.experts.38.down_proj", "model.layers.57.mlp.experts.39.down_proj", "model.layers.57.mlp.experts.40.down_proj", "model.layers.57.mlp.experts.41.down_proj", "model.layers.57.mlp.experts.42.down_proj", "model.layers.57.mlp.experts.43.down_proj", "model.layers.57.mlp.experts.44.down_proj", "model.layers.57.mlp.experts.45.down_proj", "model.layers.57.mlp.experts.46.down_proj", "model.layers.57.mlp.experts.47.down_proj", "model.layers.57.mlp.experts.48.down_proj", "model.layers.57.mlp.experts.49.down_proj", "model.layers.57.mlp.experts.50.down_proj", "model.layers.57.mlp.experts.51.down_proj", "model.layers.57.mlp.experts.52.down_proj", "model.layers.57.mlp.experts.53.down_proj", "model.layers.57.mlp.experts.54.down_proj", "model.layers.57.mlp.experts.55.down_proj", "model.layers.57.mlp.experts.56.down_proj", "model.layers.57.mlp.experts.57.down_proj", "model.layers.57.mlp.experts.58.down_proj", "model.layers.57.mlp.experts.59.down_proj", "model.layers.57.mlp.experts.60.down_proj", "model.layers.57.mlp.experts.61.down_proj", "model.layers.57.mlp.experts.62.down_proj", "model.layers.57.mlp.experts.63.down_proj", "model.layers.57.mlp.experts.64.down_proj", "model.layers.57.mlp.experts.65.down_proj", "model.layers.57.mlp.experts.66.down_proj", "model.layers.57.mlp.experts.67.down_proj", "model.layers.57.mlp.experts.68.down_proj", "model.layers.57.mlp.experts.69.down_proj", "model.layers.57.mlp.experts.70.down_proj", "model.layers.57.mlp.experts.71.down_proj", "model.layers.57.mlp.experts.72.down_proj", "model.layers.57.mlp.experts.73.down_proj", "model.layers.57.mlp.experts.74.down_proj", "model.layers.57.mlp.experts.75.down_proj", "model.layers.57.mlp.experts.76.down_proj", "model.layers.57.mlp.experts.77.down_proj", "model.layers.57.mlp.experts.78.down_proj", "model.layers.57.mlp.experts.79.down_proj", "model.layers.57.mlp.experts.80.down_proj", "model.layers.57.mlp.experts.81.down_proj", "model.layers.57.mlp.experts.82.down_proj", "model.layers.57.mlp.experts.83.down_proj", "model.layers.57.mlp.experts.84.down_proj", "model.layers.57.mlp.experts.85.down_proj", "model.layers.57.mlp.experts.86.down_proj", "model.layers.57.mlp.experts.87.down_proj", "model.layers.57.mlp.experts.88.down_proj", "model.layers.57.mlp.experts.89.down_proj", "model.layers.57.mlp.experts.90.down_proj", "model.layers.57.mlp.experts.91.down_proj", "model.layers.57.mlp.experts.92.down_proj", "model.layers.57.mlp.experts.93.down_proj", "model.layers.57.mlp.experts.94.down_proj", "model.layers.57.mlp.experts.95.down_proj", "model.layers.57.mlp.experts.96.down_proj", "model.layers.57.mlp.experts.97.down_proj", "model.layers.57.mlp.experts.98.down_proj", "model.layers.57.mlp.experts.99.down_proj", "model.layers.57.mlp.experts.100.down_proj", "model.layers.57.mlp.experts.101.down_proj", "model.layers.57.mlp.experts.102.down_proj", "model.layers.57.mlp.experts.103.down_proj", "model.layers.57.mlp.experts.104.down_proj", "model.layers.57.mlp.experts.105.down_proj", "model.layers.57.mlp.experts.106.down_proj", "model.layers.57.mlp.experts.107.down_proj", "model.layers.57.mlp.experts.108.down_proj", "model.layers.57.mlp.experts.109.down_proj", "model.layers.57.mlp.experts.110.down_proj", "model.layers.57.mlp.experts.111.down_proj", "model.layers.57.mlp.experts.112.down_proj", "model.layers.57.mlp.experts.113.down_proj", "model.layers.57.mlp.experts.114.down_proj", "model.layers.57.mlp.experts.115.down_proj", "model.layers.57.mlp.experts.116.down_proj", "model.layers.57.mlp.experts.117.down_proj", "model.layers.57.mlp.experts.118.down_proj", "model.layers.57.mlp.experts.119.down_proj", "model.layers.57.mlp.experts.120.down_proj", "model.layers.57.mlp.experts.121.down_proj", "model.layers.57.mlp.experts.122.down_proj", "model.layers.57.mlp.experts.123.down_proj", "model.layers.57.mlp.experts.124.down_proj", "model.layers.57.mlp.experts.125.down_proj", "model.layers.57.mlp.experts.126.down_proj", "model.layers.57.mlp.experts.127.down_proj", "model.layers.57.mlp.experts.128.down_proj", "model.layers.57.mlp.experts.129.down_proj", "model.layers.57.mlp.experts.130.down_proj", "model.layers.57.mlp.experts.131.down_proj", "model.layers.57.mlp.experts.132.down_proj", "model.layers.57.mlp.experts.133.down_proj", "model.layers.57.mlp.experts.134.down_proj", "model.layers.57.mlp.experts.135.down_proj", "model.layers.57.mlp.experts.136.down_proj", "model.layers.57.mlp.experts.137.down_proj", "model.layers.57.mlp.experts.138.down_proj", "model.layers.57.mlp.experts.139.down_proj", "model.layers.57.mlp.experts.140.down_proj", "model.layers.57.mlp.experts.141.down_proj", "model.layers.57.mlp.experts.142.down_proj", "model.layers.57.mlp.experts.143.down_proj", "model.layers.57.mlp.experts.144.down_proj", "model.layers.57.mlp.experts.145.down_proj", "model.layers.57.mlp.experts.146.down_proj", "model.layers.57.mlp.experts.147.down_proj", "model.layers.57.mlp.experts.148.down_proj", "model.layers.57.mlp.experts.149.down_proj", "model.layers.57.mlp.experts.150.down_proj", "model.layers.57.mlp.experts.151.down_proj", "model.layers.57.mlp.experts.152.down_proj", "model.layers.57.mlp.experts.153.down_proj", "model.layers.57.mlp.experts.154.down_proj", "model.layers.57.mlp.experts.155.down_proj", "model.layers.57.mlp.experts.156.down_proj", "model.layers.57.mlp.experts.157.down_proj", "model.layers.57.mlp.experts.158.down_proj", "model.layers.57.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -1.064259558916647e-05, "dbits": 1258291200 }, { "dkld": -2.05175019800663e-05, "dbits": 2516582400 }, { "dkld": -3.2750982791193706e-05, "dbits": 3774873600 }, { "dkld": -4.9964338541041986e-05, "dbits": 6291456000 } ] }, { "idx": 345, "layers": [ "model.layers.58.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00012906976044177454, "dbits": 62914560 }, { "dkld": 0.00023069055750965517, "dbits": 125829120 }, { "dkld": 8.137021213769358e-05, "dbits": 188743680 }, { "dkld": 0.0001649101264774716, "dbits": 314572800 } ] }, { "idx": 346, "layers": [ "model.layers.58.self_attn.k_proj", "model.layers.58.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00017306655645371594, "dbits": 10485760 }, { "dkld": -0.0001092424616217641, "dbits": 20971520 }, { "dkld": -0.00018972251564264297, "dbits": 31457280 }, { "dkld": -0.00013197492808103561, "dbits": 52428800 } ] }, { "idx": 347, "layers": [ "model.layers.58.self_attn.o_proj" ], "candidates": [ { "dkld": 2.0363740622988957e-05, "dbits": 62914560 }, { "dkld": 1.803766936063489e-05, "dbits": 125829120 }, { "dkld": -6.845947355033077e-05, "dbits": 188743680 }, { "dkld": -0.00010125692933798947, "dbits": 314572800 } ] }, { "idx": 348, "layers": [ "model.layers.58.mlp.shared_experts.gate_proj", "model.layers.58.mlp.shared_experts.up_proj", "model.layers.58.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00020577032119035998, "dbits": 23592960 }, { "dkld": 4.939744248985689e-05, "dbits": 47185920 }, { "dkld": 9.423121809959412e-06, "dbits": 70778880 }, { "dkld": -1.4571473002433777e-05, "dbits": 117964800 } ] }, { "idx": 349, "layers": [ "model.layers.58.mlp.experts.0.gate_proj", "model.layers.58.mlp.experts.1.gate_proj", "model.layers.58.mlp.experts.2.gate_proj", "model.layers.58.mlp.experts.3.gate_proj", "model.layers.58.mlp.experts.4.gate_proj", "model.layers.58.mlp.experts.5.gate_proj", "model.layers.58.mlp.experts.6.gate_proj", "model.layers.58.mlp.experts.7.gate_proj", "model.layers.58.mlp.experts.8.gate_proj", "model.layers.58.mlp.experts.9.gate_proj", "model.layers.58.mlp.experts.10.gate_proj", "model.layers.58.mlp.experts.11.gate_proj", "model.layers.58.mlp.experts.12.gate_proj", "model.layers.58.mlp.experts.13.gate_proj", "model.layers.58.mlp.experts.14.gate_proj", "model.layers.58.mlp.experts.15.gate_proj", "model.layers.58.mlp.experts.16.gate_proj", "model.layers.58.mlp.experts.17.gate_proj", "model.layers.58.mlp.experts.18.gate_proj", "model.layers.58.mlp.experts.19.gate_proj", "model.layers.58.mlp.experts.20.gate_proj", "model.layers.58.mlp.experts.21.gate_proj", "model.layers.58.mlp.experts.22.gate_proj", "model.layers.58.mlp.experts.23.gate_proj", "model.layers.58.mlp.experts.24.gate_proj", "model.layers.58.mlp.experts.25.gate_proj", "model.layers.58.mlp.experts.26.gate_proj", "model.layers.58.mlp.experts.27.gate_proj", "model.layers.58.mlp.experts.28.gate_proj", "model.layers.58.mlp.experts.29.gate_proj", "model.layers.58.mlp.experts.30.gate_proj", "model.layers.58.mlp.experts.31.gate_proj", "model.layers.58.mlp.experts.32.gate_proj", "model.layers.58.mlp.experts.33.gate_proj", "model.layers.58.mlp.experts.34.gate_proj", "model.layers.58.mlp.experts.35.gate_proj", "model.layers.58.mlp.experts.36.gate_proj", "model.layers.58.mlp.experts.37.gate_proj", "model.layers.58.mlp.experts.38.gate_proj", "model.layers.58.mlp.experts.39.gate_proj", "model.layers.58.mlp.experts.40.gate_proj", "model.layers.58.mlp.experts.41.gate_proj", "model.layers.58.mlp.experts.42.gate_proj", "model.layers.58.mlp.experts.43.gate_proj", "model.layers.58.mlp.experts.44.gate_proj", "model.layers.58.mlp.experts.45.gate_proj", "model.layers.58.mlp.experts.46.gate_proj", "model.layers.58.mlp.experts.47.gate_proj", "model.layers.58.mlp.experts.48.gate_proj", "model.layers.58.mlp.experts.49.gate_proj", "model.layers.58.mlp.experts.50.gate_proj", "model.layers.58.mlp.experts.51.gate_proj", "model.layers.58.mlp.experts.52.gate_proj", "model.layers.58.mlp.experts.53.gate_proj", "model.layers.58.mlp.experts.54.gate_proj", "model.layers.58.mlp.experts.55.gate_proj", "model.layers.58.mlp.experts.56.gate_proj", "model.layers.58.mlp.experts.57.gate_proj", "model.layers.58.mlp.experts.58.gate_proj", "model.layers.58.mlp.experts.59.gate_proj", "model.layers.58.mlp.experts.60.gate_proj", "model.layers.58.mlp.experts.61.gate_proj", "model.layers.58.mlp.experts.62.gate_proj", "model.layers.58.mlp.experts.63.gate_proj", "model.layers.58.mlp.experts.64.gate_proj", "model.layers.58.mlp.experts.65.gate_proj", "model.layers.58.mlp.experts.66.gate_proj", "model.layers.58.mlp.experts.67.gate_proj", "model.layers.58.mlp.experts.68.gate_proj", "model.layers.58.mlp.experts.69.gate_proj", "model.layers.58.mlp.experts.70.gate_proj", "model.layers.58.mlp.experts.71.gate_proj", "model.layers.58.mlp.experts.72.gate_proj", "model.layers.58.mlp.experts.73.gate_proj", "model.layers.58.mlp.experts.74.gate_proj", "model.layers.58.mlp.experts.75.gate_proj", "model.layers.58.mlp.experts.76.gate_proj", "model.layers.58.mlp.experts.77.gate_proj", "model.layers.58.mlp.experts.78.gate_proj", "model.layers.58.mlp.experts.79.gate_proj", "model.layers.58.mlp.experts.80.gate_proj", "model.layers.58.mlp.experts.81.gate_proj", "model.layers.58.mlp.experts.82.gate_proj", "model.layers.58.mlp.experts.83.gate_proj", "model.layers.58.mlp.experts.84.gate_proj", "model.layers.58.mlp.experts.85.gate_proj", "model.layers.58.mlp.experts.86.gate_proj", "model.layers.58.mlp.experts.87.gate_proj", "model.layers.58.mlp.experts.88.gate_proj", "model.layers.58.mlp.experts.89.gate_proj", "model.layers.58.mlp.experts.90.gate_proj", "model.layers.58.mlp.experts.91.gate_proj", "model.layers.58.mlp.experts.92.gate_proj", "model.layers.58.mlp.experts.93.gate_proj", "model.layers.58.mlp.experts.94.gate_proj", "model.layers.58.mlp.experts.95.gate_proj", "model.layers.58.mlp.experts.96.gate_proj", "model.layers.58.mlp.experts.97.gate_proj", "model.layers.58.mlp.experts.98.gate_proj", "model.layers.58.mlp.experts.99.gate_proj", "model.layers.58.mlp.experts.100.gate_proj", "model.layers.58.mlp.experts.101.gate_proj", "model.layers.58.mlp.experts.102.gate_proj", "model.layers.58.mlp.experts.103.gate_proj", "model.layers.58.mlp.experts.104.gate_proj", "model.layers.58.mlp.experts.105.gate_proj", "model.layers.58.mlp.experts.106.gate_proj", "model.layers.58.mlp.experts.107.gate_proj", "model.layers.58.mlp.experts.108.gate_proj", "model.layers.58.mlp.experts.109.gate_proj", "model.layers.58.mlp.experts.110.gate_proj", "model.layers.58.mlp.experts.111.gate_proj", "model.layers.58.mlp.experts.112.gate_proj", "model.layers.58.mlp.experts.113.gate_proj", "model.layers.58.mlp.experts.114.gate_proj", "model.layers.58.mlp.experts.115.gate_proj", "model.layers.58.mlp.experts.116.gate_proj", "model.layers.58.mlp.experts.117.gate_proj", "model.layers.58.mlp.experts.118.gate_proj", "model.layers.58.mlp.experts.119.gate_proj", "model.layers.58.mlp.experts.120.gate_proj", "model.layers.58.mlp.experts.121.gate_proj", "model.layers.58.mlp.experts.122.gate_proj", "model.layers.58.mlp.experts.123.gate_proj", "model.layers.58.mlp.experts.124.gate_proj", "model.layers.58.mlp.experts.125.gate_proj", "model.layers.58.mlp.experts.126.gate_proj", "model.layers.58.mlp.experts.127.gate_proj", "model.layers.58.mlp.experts.128.gate_proj", "model.layers.58.mlp.experts.129.gate_proj", "model.layers.58.mlp.experts.130.gate_proj", "model.layers.58.mlp.experts.131.gate_proj", "model.layers.58.mlp.experts.132.gate_proj", "model.layers.58.mlp.experts.133.gate_proj", "model.layers.58.mlp.experts.134.gate_proj", "model.layers.58.mlp.experts.135.gate_proj", "model.layers.58.mlp.experts.136.gate_proj", "model.layers.58.mlp.experts.137.gate_proj", "model.layers.58.mlp.experts.138.gate_proj", "model.layers.58.mlp.experts.139.gate_proj", "model.layers.58.mlp.experts.140.gate_proj", "model.layers.58.mlp.experts.141.gate_proj", "model.layers.58.mlp.experts.142.gate_proj", "model.layers.58.mlp.experts.143.gate_proj", "model.layers.58.mlp.experts.144.gate_proj", "model.layers.58.mlp.experts.145.gate_proj", "model.layers.58.mlp.experts.146.gate_proj", "model.layers.58.mlp.experts.147.gate_proj", "model.layers.58.mlp.experts.148.gate_proj", "model.layers.58.mlp.experts.149.gate_proj", "model.layers.58.mlp.experts.150.gate_proj", "model.layers.58.mlp.experts.151.gate_proj", "model.layers.58.mlp.experts.152.gate_proj", "model.layers.58.mlp.experts.153.gate_proj", "model.layers.58.mlp.experts.154.gate_proj", "model.layers.58.mlp.experts.155.gate_proj", "model.layers.58.mlp.experts.156.gate_proj", "model.layers.58.mlp.experts.157.gate_proj", "model.layers.58.mlp.experts.158.gate_proj", "model.layers.58.mlp.experts.159.gate_proj", "model.layers.58.mlp.experts.0.up_proj", "model.layers.58.mlp.experts.1.up_proj", "model.layers.58.mlp.experts.2.up_proj", "model.layers.58.mlp.experts.3.up_proj", "model.layers.58.mlp.experts.4.up_proj", "model.layers.58.mlp.experts.5.up_proj", "model.layers.58.mlp.experts.6.up_proj", "model.layers.58.mlp.experts.7.up_proj", "model.layers.58.mlp.experts.8.up_proj", "model.layers.58.mlp.experts.9.up_proj", "model.layers.58.mlp.experts.10.up_proj", "model.layers.58.mlp.experts.11.up_proj", "model.layers.58.mlp.experts.12.up_proj", "model.layers.58.mlp.experts.13.up_proj", "model.layers.58.mlp.experts.14.up_proj", "model.layers.58.mlp.experts.15.up_proj", "model.layers.58.mlp.experts.16.up_proj", "model.layers.58.mlp.experts.17.up_proj", "model.layers.58.mlp.experts.18.up_proj", "model.layers.58.mlp.experts.19.up_proj", "model.layers.58.mlp.experts.20.up_proj", "model.layers.58.mlp.experts.21.up_proj", "model.layers.58.mlp.experts.22.up_proj", "model.layers.58.mlp.experts.23.up_proj", "model.layers.58.mlp.experts.24.up_proj", "model.layers.58.mlp.experts.25.up_proj", "model.layers.58.mlp.experts.26.up_proj", "model.layers.58.mlp.experts.27.up_proj", "model.layers.58.mlp.experts.28.up_proj", "model.layers.58.mlp.experts.29.up_proj", "model.layers.58.mlp.experts.30.up_proj", "model.layers.58.mlp.experts.31.up_proj", "model.layers.58.mlp.experts.32.up_proj", "model.layers.58.mlp.experts.33.up_proj", "model.layers.58.mlp.experts.34.up_proj", "model.layers.58.mlp.experts.35.up_proj", "model.layers.58.mlp.experts.36.up_proj", "model.layers.58.mlp.experts.37.up_proj", "model.layers.58.mlp.experts.38.up_proj", "model.layers.58.mlp.experts.39.up_proj", "model.layers.58.mlp.experts.40.up_proj", "model.layers.58.mlp.experts.41.up_proj", "model.layers.58.mlp.experts.42.up_proj", "model.layers.58.mlp.experts.43.up_proj", "model.layers.58.mlp.experts.44.up_proj", "model.layers.58.mlp.experts.45.up_proj", "model.layers.58.mlp.experts.46.up_proj", "model.layers.58.mlp.experts.47.up_proj", "model.layers.58.mlp.experts.48.up_proj", "model.layers.58.mlp.experts.49.up_proj", "model.layers.58.mlp.experts.50.up_proj", "model.layers.58.mlp.experts.51.up_proj", "model.layers.58.mlp.experts.52.up_proj", "model.layers.58.mlp.experts.53.up_proj", "model.layers.58.mlp.experts.54.up_proj", "model.layers.58.mlp.experts.55.up_proj", "model.layers.58.mlp.experts.56.up_proj", "model.layers.58.mlp.experts.57.up_proj", "model.layers.58.mlp.experts.58.up_proj", "model.layers.58.mlp.experts.59.up_proj", "model.layers.58.mlp.experts.60.up_proj", "model.layers.58.mlp.experts.61.up_proj", "model.layers.58.mlp.experts.62.up_proj", "model.layers.58.mlp.experts.63.up_proj", "model.layers.58.mlp.experts.64.up_proj", "model.layers.58.mlp.experts.65.up_proj", "model.layers.58.mlp.experts.66.up_proj", "model.layers.58.mlp.experts.67.up_proj", "model.layers.58.mlp.experts.68.up_proj", "model.layers.58.mlp.experts.69.up_proj", "model.layers.58.mlp.experts.70.up_proj", "model.layers.58.mlp.experts.71.up_proj", "model.layers.58.mlp.experts.72.up_proj", "model.layers.58.mlp.experts.73.up_proj", "model.layers.58.mlp.experts.74.up_proj", "model.layers.58.mlp.experts.75.up_proj", "model.layers.58.mlp.experts.76.up_proj", "model.layers.58.mlp.experts.77.up_proj", "model.layers.58.mlp.experts.78.up_proj", "model.layers.58.mlp.experts.79.up_proj", "model.layers.58.mlp.experts.80.up_proj", "model.layers.58.mlp.experts.81.up_proj", "model.layers.58.mlp.experts.82.up_proj", "model.layers.58.mlp.experts.83.up_proj", "model.layers.58.mlp.experts.84.up_proj", "model.layers.58.mlp.experts.85.up_proj", "model.layers.58.mlp.experts.86.up_proj", "model.layers.58.mlp.experts.87.up_proj", "model.layers.58.mlp.experts.88.up_proj", "model.layers.58.mlp.experts.89.up_proj", "model.layers.58.mlp.experts.90.up_proj", "model.layers.58.mlp.experts.91.up_proj", "model.layers.58.mlp.experts.92.up_proj", "model.layers.58.mlp.experts.93.up_proj", "model.layers.58.mlp.experts.94.up_proj", "model.layers.58.mlp.experts.95.up_proj", "model.layers.58.mlp.experts.96.up_proj", "model.layers.58.mlp.experts.97.up_proj", "model.layers.58.mlp.experts.98.up_proj", "model.layers.58.mlp.experts.99.up_proj", "model.layers.58.mlp.experts.100.up_proj", "model.layers.58.mlp.experts.101.up_proj", "model.layers.58.mlp.experts.102.up_proj", "model.layers.58.mlp.experts.103.up_proj", "model.layers.58.mlp.experts.104.up_proj", "model.layers.58.mlp.experts.105.up_proj", "model.layers.58.mlp.experts.106.up_proj", "model.layers.58.mlp.experts.107.up_proj", "model.layers.58.mlp.experts.108.up_proj", "model.layers.58.mlp.experts.109.up_proj", "model.layers.58.mlp.experts.110.up_proj", "model.layers.58.mlp.experts.111.up_proj", "model.layers.58.mlp.experts.112.up_proj", "model.layers.58.mlp.experts.113.up_proj", "model.layers.58.mlp.experts.114.up_proj", "model.layers.58.mlp.experts.115.up_proj", "model.layers.58.mlp.experts.116.up_proj", "model.layers.58.mlp.experts.117.up_proj", "model.layers.58.mlp.experts.118.up_proj", "model.layers.58.mlp.experts.119.up_proj", "model.layers.58.mlp.experts.120.up_proj", "model.layers.58.mlp.experts.121.up_proj", "model.layers.58.mlp.experts.122.up_proj", "model.layers.58.mlp.experts.123.up_proj", "model.layers.58.mlp.experts.124.up_proj", "model.layers.58.mlp.experts.125.up_proj", "model.layers.58.mlp.experts.126.up_proj", "model.layers.58.mlp.experts.127.up_proj", "model.layers.58.mlp.experts.128.up_proj", "model.layers.58.mlp.experts.129.up_proj", "model.layers.58.mlp.experts.130.up_proj", "model.layers.58.mlp.experts.131.up_proj", "model.layers.58.mlp.experts.132.up_proj", "model.layers.58.mlp.experts.133.up_proj", "model.layers.58.mlp.experts.134.up_proj", "model.layers.58.mlp.experts.135.up_proj", "model.layers.58.mlp.experts.136.up_proj", "model.layers.58.mlp.experts.137.up_proj", "model.layers.58.mlp.experts.138.up_proj", "model.layers.58.mlp.experts.139.up_proj", "model.layers.58.mlp.experts.140.up_proj", "model.layers.58.mlp.experts.141.up_proj", "model.layers.58.mlp.experts.142.up_proj", "model.layers.58.mlp.experts.143.up_proj", "model.layers.58.mlp.experts.144.up_proj", "model.layers.58.mlp.experts.145.up_proj", "model.layers.58.mlp.experts.146.up_proj", "model.layers.58.mlp.experts.147.up_proj", "model.layers.58.mlp.experts.148.up_proj", "model.layers.58.mlp.experts.149.up_proj", "model.layers.58.mlp.experts.150.up_proj", "model.layers.58.mlp.experts.151.up_proj", "model.layers.58.mlp.experts.152.up_proj", "model.layers.58.mlp.experts.153.up_proj", "model.layers.58.mlp.experts.154.up_proj", "model.layers.58.mlp.experts.155.up_proj", "model.layers.58.mlp.experts.156.up_proj", "model.layers.58.mlp.experts.157.up_proj", "model.layers.58.mlp.experts.158.up_proj", "model.layers.58.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0002481114119291389, "dbits": 2516582400 }, { "dkld": -0.00023469971492887098, "dbits": 5033164800 }, { "dkld": -0.00023625800386072038, "dbits": 7549747200 }, { "dkld": -0.000252281222492462, "dbits": 12582912000 } ] }, { "idx": 350, "layers": [ "model.layers.58.mlp.experts.0.down_proj", "model.layers.58.mlp.experts.1.down_proj", "model.layers.58.mlp.experts.2.down_proj", "model.layers.58.mlp.experts.3.down_proj", "model.layers.58.mlp.experts.4.down_proj", "model.layers.58.mlp.experts.5.down_proj", "model.layers.58.mlp.experts.6.down_proj", "model.layers.58.mlp.experts.7.down_proj", "model.layers.58.mlp.experts.8.down_proj", "model.layers.58.mlp.experts.9.down_proj", "model.layers.58.mlp.experts.10.down_proj", "model.layers.58.mlp.experts.11.down_proj", "model.layers.58.mlp.experts.12.down_proj", "model.layers.58.mlp.experts.13.down_proj", "model.layers.58.mlp.experts.14.down_proj", "model.layers.58.mlp.experts.15.down_proj", "model.layers.58.mlp.experts.16.down_proj", "model.layers.58.mlp.experts.17.down_proj", "model.layers.58.mlp.experts.18.down_proj", "model.layers.58.mlp.experts.19.down_proj", "model.layers.58.mlp.experts.20.down_proj", "model.layers.58.mlp.experts.21.down_proj", "model.layers.58.mlp.experts.22.down_proj", "model.layers.58.mlp.experts.23.down_proj", "model.layers.58.mlp.experts.24.down_proj", "model.layers.58.mlp.experts.25.down_proj", "model.layers.58.mlp.experts.26.down_proj", "model.layers.58.mlp.experts.27.down_proj", "model.layers.58.mlp.experts.28.down_proj", "model.layers.58.mlp.experts.29.down_proj", "model.layers.58.mlp.experts.30.down_proj", "model.layers.58.mlp.experts.31.down_proj", "model.layers.58.mlp.experts.32.down_proj", "model.layers.58.mlp.experts.33.down_proj", "model.layers.58.mlp.experts.34.down_proj", "model.layers.58.mlp.experts.35.down_proj", "model.layers.58.mlp.experts.36.down_proj", "model.layers.58.mlp.experts.37.down_proj", "model.layers.58.mlp.experts.38.down_proj", "model.layers.58.mlp.experts.39.down_proj", "model.layers.58.mlp.experts.40.down_proj", "model.layers.58.mlp.experts.41.down_proj", "model.layers.58.mlp.experts.42.down_proj", "model.layers.58.mlp.experts.43.down_proj", "model.layers.58.mlp.experts.44.down_proj", "model.layers.58.mlp.experts.45.down_proj", "model.layers.58.mlp.experts.46.down_proj", "model.layers.58.mlp.experts.47.down_proj", "model.layers.58.mlp.experts.48.down_proj", "model.layers.58.mlp.experts.49.down_proj", "model.layers.58.mlp.experts.50.down_proj", "model.layers.58.mlp.experts.51.down_proj", "model.layers.58.mlp.experts.52.down_proj", "model.layers.58.mlp.experts.53.down_proj", "model.layers.58.mlp.experts.54.down_proj", "model.layers.58.mlp.experts.55.down_proj", "model.layers.58.mlp.experts.56.down_proj", "model.layers.58.mlp.experts.57.down_proj", "model.layers.58.mlp.experts.58.down_proj", "model.layers.58.mlp.experts.59.down_proj", "model.layers.58.mlp.experts.60.down_proj", "model.layers.58.mlp.experts.61.down_proj", "model.layers.58.mlp.experts.62.down_proj", "model.layers.58.mlp.experts.63.down_proj", "model.layers.58.mlp.experts.64.down_proj", "model.layers.58.mlp.experts.65.down_proj", "model.layers.58.mlp.experts.66.down_proj", "model.layers.58.mlp.experts.67.down_proj", "model.layers.58.mlp.experts.68.down_proj", "model.layers.58.mlp.experts.69.down_proj", "model.layers.58.mlp.experts.70.down_proj", "model.layers.58.mlp.experts.71.down_proj", "model.layers.58.mlp.experts.72.down_proj", "model.layers.58.mlp.experts.73.down_proj", "model.layers.58.mlp.experts.74.down_proj", "model.layers.58.mlp.experts.75.down_proj", "model.layers.58.mlp.experts.76.down_proj", "model.layers.58.mlp.experts.77.down_proj", "model.layers.58.mlp.experts.78.down_proj", "model.layers.58.mlp.experts.79.down_proj", "model.layers.58.mlp.experts.80.down_proj", "model.layers.58.mlp.experts.81.down_proj", "model.layers.58.mlp.experts.82.down_proj", "model.layers.58.mlp.experts.83.down_proj", "model.layers.58.mlp.experts.84.down_proj", "model.layers.58.mlp.experts.85.down_proj", "model.layers.58.mlp.experts.86.down_proj", "model.layers.58.mlp.experts.87.down_proj", "model.layers.58.mlp.experts.88.down_proj", "model.layers.58.mlp.experts.89.down_proj", "model.layers.58.mlp.experts.90.down_proj", "model.layers.58.mlp.experts.91.down_proj", "model.layers.58.mlp.experts.92.down_proj", "model.layers.58.mlp.experts.93.down_proj", "model.layers.58.mlp.experts.94.down_proj", "model.layers.58.mlp.experts.95.down_proj", "model.layers.58.mlp.experts.96.down_proj", "model.layers.58.mlp.experts.97.down_proj", "model.layers.58.mlp.experts.98.down_proj", "model.layers.58.mlp.experts.99.down_proj", "model.layers.58.mlp.experts.100.down_proj", "model.layers.58.mlp.experts.101.down_proj", "model.layers.58.mlp.experts.102.down_proj", "model.layers.58.mlp.experts.103.down_proj", "model.layers.58.mlp.experts.104.down_proj", "model.layers.58.mlp.experts.105.down_proj", "model.layers.58.mlp.experts.106.down_proj", "model.layers.58.mlp.experts.107.down_proj", "model.layers.58.mlp.experts.108.down_proj", "model.layers.58.mlp.experts.109.down_proj", "model.layers.58.mlp.experts.110.down_proj", "model.layers.58.mlp.experts.111.down_proj", "model.layers.58.mlp.experts.112.down_proj", "model.layers.58.mlp.experts.113.down_proj", "model.layers.58.mlp.experts.114.down_proj", "model.layers.58.mlp.experts.115.down_proj", "model.layers.58.mlp.experts.116.down_proj", "model.layers.58.mlp.experts.117.down_proj", "model.layers.58.mlp.experts.118.down_proj", "model.layers.58.mlp.experts.119.down_proj", "model.layers.58.mlp.experts.120.down_proj", "model.layers.58.mlp.experts.121.down_proj", "model.layers.58.mlp.experts.122.down_proj", "model.layers.58.mlp.experts.123.down_proj", "model.layers.58.mlp.experts.124.down_proj", "model.layers.58.mlp.experts.125.down_proj", "model.layers.58.mlp.experts.126.down_proj", "model.layers.58.mlp.experts.127.down_proj", "model.layers.58.mlp.experts.128.down_proj", "model.layers.58.mlp.experts.129.down_proj", "model.layers.58.mlp.experts.130.down_proj", "model.layers.58.mlp.experts.131.down_proj", "model.layers.58.mlp.experts.132.down_proj", "model.layers.58.mlp.experts.133.down_proj", "model.layers.58.mlp.experts.134.down_proj", "model.layers.58.mlp.experts.135.down_proj", "model.layers.58.mlp.experts.136.down_proj", "model.layers.58.mlp.experts.137.down_proj", "model.layers.58.mlp.experts.138.down_proj", "model.layers.58.mlp.experts.139.down_proj", "model.layers.58.mlp.experts.140.down_proj", "model.layers.58.mlp.experts.141.down_proj", "model.layers.58.mlp.experts.142.down_proj", "model.layers.58.mlp.experts.143.down_proj", "model.layers.58.mlp.experts.144.down_proj", "model.layers.58.mlp.experts.145.down_proj", "model.layers.58.mlp.experts.146.down_proj", "model.layers.58.mlp.experts.147.down_proj", "model.layers.58.mlp.experts.148.down_proj", "model.layers.58.mlp.experts.149.down_proj", "model.layers.58.mlp.experts.150.down_proj", "model.layers.58.mlp.experts.151.down_proj", "model.layers.58.mlp.experts.152.down_proj", "model.layers.58.mlp.experts.153.down_proj", "model.layers.58.mlp.experts.154.down_proj", "model.layers.58.mlp.experts.155.down_proj", "model.layers.58.mlp.experts.156.down_proj", "model.layers.58.mlp.experts.157.down_proj", "model.layers.58.mlp.experts.158.down_proj", "model.layers.58.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00023800767958165325, "dbits": 1258291200 }, { "dkld": -0.00020231483504176695, "dbits": 2516582400 }, { "dkld": -0.0002470728009939249, "dbits": 3774873600 }, { "dkld": -0.00023345407098532278, "dbits": 6291456000 } ] }, { "idx": 351, "layers": [ "model.layers.59.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00016867946833372394, "dbits": 62914560 }, { "dkld": -0.00017476435750723995, "dbits": 125829120 }, { "dkld": -0.00020356588065624515, "dbits": 188743680 }, { "dkld": -0.00022485861554742692, "dbits": 314572800 } ] }, { "idx": 352, "layers": [ "model.layers.59.self_attn.k_proj", "model.layers.59.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00011188583448529243, "dbits": 10485760 }, { "dkld": -0.0002694314345717541, "dbits": 20971520 }, { "dkld": -0.00023373905569315512, "dbits": 31457280 }, { "dkld": -0.00023334324359894354, "dbits": 52428800 } ] }, { "idx": 353, "layers": [ "model.layers.59.self_attn.o_proj" ], "candidates": [ { "dkld": 1.290440559387207e-05, "dbits": 62914560 }, { "dkld": 9.071109816431722e-05, "dbits": 125829120 }, { "dkld": 3.6440324038267136e-05, "dbits": 188743680 }, { "dkld": 4.952931776641689e-05, "dbits": 314572800 } ] }, { "idx": 354, "layers": [ "model.layers.59.mlp.shared_experts.gate_proj", "model.layers.59.mlp.shared_experts.up_proj", "model.layers.59.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0001817452721297741, "dbits": 23592960 }, { "dkld": -0.00031948806717992106, "dbits": 47185920 }, { "dkld": -0.0003115192987024895, "dbits": 70778880 }, { "dkld": -0.0002999956719577368, "dbits": 117964800 } ] }, { "idx": 355, "layers": [ "model.layers.59.mlp.experts.0.gate_proj", "model.layers.59.mlp.experts.1.gate_proj", "model.layers.59.mlp.experts.2.gate_proj", "model.layers.59.mlp.experts.3.gate_proj", "model.layers.59.mlp.experts.4.gate_proj", "model.layers.59.mlp.experts.5.gate_proj", "model.layers.59.mlp.experts.6.gate_proj", "model.layers.59.mlp.experts.7.gate_proj", "model.layers.59.mlp.experts.8.gate_proj", "model.layers.59.mlp.experts.9.gate_proj", "model.layers.59.mlp.experts.10.gate_proj", "model.layers.59.mlp.experts.11.gate_proj", "model.layers.59.mlp.experts.12.gate_proj", "model.layers.59.mlp.experts.13.gate_proj", "model.layers.59.mlp.experts.14.gate_proj", "model.layers.59.mlp.experts.15.gate_proj", "model.layers.59.mlp.experts.16.gate_proj", "model.layers.59.mlp.experts.17.gate_proj", "model.layers.59.mlp.experts.18.gate_proj", "model.layers.59.mlp.experts.19.gate_proj", "model.layers.59.mlp.experts.20.gate_proj", "model.layers.59.mlp.experts.21.gate_proj", "model.layers.59.mlp.experts.22.gate_proj", "model.layers.59.mlp.experts.23.gate_proj", "model.layers.59.mlp.experts.24.gate_proj", "model.layers.59.mlp.experts.25.gate_proj", "model.layers.59.mlp.experts.26.gate_proj", "model.layers.59.mlp.experts.27.gate_proj", "model.layers.59.mlp.experts.28.gate_proj", "model.layers.59.mlp.experts.29.gate_proj", "model.layers.59.mlp.experts.30.gate_proj", "model.layers.59.mlp.experts.31.gate_proj", "model.layers.59.mlp.experts.32.gate_proj", "model.layers.59.mlp.experts.33.gate_proj", "model.layers.59.mlp.experts.34.gate_proj", "model.layers.59.mlp.experts.35.gate_proj", "model.layers.59.mlp.experts.36.gate_proj", "model.layers.59.mlp.experts.37.gate_proj", "model.layers.59.mlp.experts.38.gate_proj", "model.layers.59.mlp.experts.39.gate_proj", "model.layers.59.mlp.experts.40.gate_proj", "model.layers.59.mlp.experts.41.gate_proj", "model.layers.59.mlp.experts.42.gate_proj", "model.layers.59.mlp.experts.43.gate_proj", "model.layers.59.mlp.experts.44.gate_proj", "model.layers.59.mlp.experts.45.gate_proj", "model.layers.59.mlp.experts.46.gate_proj", "model.layers.59.mlp.experts.47.gate_proj", "model.layers.59.mlp.experts.48.gate_proj", "model.layers.59.mlp.experts.49.gate_proj", "model.layers.59.mlp.experts.50.gate_proj", "model.layers.59.mlp.experts.51.gate_proj", "model.layers.59.mlp.experts.52.gate_proj", "model.layers.59.mlp.experts.53.gate_proj", "model.layers.59.mlp.experts.54.gate_proj", "model.layers.59.mlp.experts.55.gate_proj", "model.layers.59.mlp.experts.56.gate_proj", "model.layers.59.mlp.experts.57.gate_proj", "model.layers.59.mlp.experts.58.gate_proj", "model.layers.59.mlp.experts.59.gate_proj", "model.layers.59.mlp.experts.60.gate_proj", "model.layers.59.mlp.experts.61.gate_proj", "model.layers.59.mlp.experts.62.gate_proj", "model.layers.59.mlp.experts.63.gate_proj", "model.layers.59.mlp.experts.64.gate_proj", "model.layers.59.mlp.experts.65.gate_proj", "model.layers.59.mlp.experts.66.gate_proj", "model.layers.59.mlp.experts.67.gate_proj", "model.layers.59.mlp.experts.68.gate_proj", "model.layers.59.mlp.experts.69.gate_proj", "model.layers.59.mlp.experts.70.gate_proj", "model.layers.59.mlp.experts.71.gate_proj", "model.layers.59.mlp.experts.72.gate_proj", "model.layers.59.mlp.experts.73.gate_proj", "model.layers.59.mlp.experts.74.gate_proj", "model.layers.59.mlp.experts.75.gate_proj", "model.layers.59.mlp.experts.76.gate_proj", "model.layers.59.mlp.experts.77.gate_proj", "model.layers.59.mlp.experts.78.gate_proj", "model.layers.59.mlp.experts.79.gate_proj", "model.layers.59.mlp.experts.80.gate_proj", "model.layers.59.mlp.experts.81.gate_proj", "model.layers.59.mlp.experts.82.gate_proj", "model.layers.59.mlp.experts.83.gate_proj", "model.layers.59.mlp.experts.84.gate_proj", "model.layers.59.mlp.experts.85.gate_proj", "model.layers.59.mlp.experts.86.gate_proj", "model.layers.59.mlp.experts.87.gate_proj", "model.layers.59.mlp.experts.88.gate_proj", "model.layers.59.mlp.experts.89.gate_proj", "model.layers.59.mlp.experts.90.gate_proj", "model.layers.59.mlp.experts.91.gate_proj", "model.layers.59.mlp.experts.92.gate_proj", "model.layers.59.mlp.experts.93.gate_proj", "model.layers.59.mlp.experts.94.gate_proj", "model.layers.59.mlp.experts.95.gate_proj", "model.layers.59.mlp.experts.96.gate_proj", "model.layers.59.mlp.experts.97.gate_proj", "model.layers.59.mlp.experts.98.gate_proj", "model.layers.59.mlp.experts.99.gate_proj", "model.layers.59.mlp.experts.100.gate_proj", "model.layers.59.mlp.experts.101.gate_proj", "model.layers.59.mlp.experts.102.gate_proj", "model.layers.59.mlp.experts.103.gate_proj", "model.layers.59.mlp.experts.104.gate_proj", "model.layers.59.mlp.experts.105.gate_proj", "model.layers.59.mlp.experts.106.gate_proj", "model.layers.59.mlp.experts.107.gate_proj", "model.layers.59.mlp.experts.108.gate_proj", "model.layers.59.mlp.experts.109.gate_proj", "model.layers.59.mlp.experts.110.gate_proj", "model.layers.59.mlp.experts.111.gate_proj", "model.layers.59.mlp.experts.112.gate_proj", "model.layers.59.mlp.experts.113.gate_proj", "model.layers.59.mlp.experts.114.gate_proj", "model.layers.59.mlp.experts.115.gate_proj", "model.layers.59.mlp.experts.116.gate_proj", "model.layers.59.mlp.experts.117.gate_proj", "model.layers.59.mlp.experts.118.gate_proj", "model.layers.59.mlp.experts.119.gate_proj", "model.layers.59.mlp.experts.120.gate_proj", "model.layers.59.mlp.experts.121.gate_proj", "model.layers.59.mlp.experts.122.gate_proj", "model.layers.59.mlp.experts.123.gate_proj", "model.layers.59.mlp.experts.124.gate_proj", "model.layers.59.mlp.experts.125.gate_proj", "model.layers.59.mlp.experts.126.gate_proj", "model.layers.59.mlp.experts.127.gate_proj", "model.layers.59.mlp.experts.128.gate_proj", "model.layers.59.mlp.experts.129.gate_proj", "model.layers.59.mlp.experts.130.gate_proj", "model.layers.59.mlp.experts.131.gate_proj", "model.layers.59.mlp.experts.132.gate_proj", "model.layers.59.mlp.experts.133.gate_proj", "model.layers.59.mlp.experts.134.gate_proj", "model.layers.59.mlp.experts.135.gate_proj", "model.layers.59.mlp.experts.136.gate_proj", "model.layers.59.mlp.experts.137.gate_proj", "model.layers.59.mlp.experts.138.gate_proj", "model.layers.59.mlp.experts.139.gate_proj", "model.layers.59.mlp.experts.140.gate_proj", "model.layers.59.mlp.experts.141.gate_proj", "model.layers.59.mlp.experts.142.gate_proj", "model.layers.59.mlp.experts.143.gate_proj", "model.layers.59.mlp.experts.144.gate_proj", "model.layers.59.mlp.experts.145.gate_proj", "model.layers.59.mlp.experts.146.gate_proj", "model.layers.59.mlp.experts.147.gate_proj", "model.layers.59.mlp.experts.148.gate_proj", "model.layers.59.mlp.experts.149.gate_proj", "model.layers.59.mlp.experts.150.gate_proj", "model.layers.59.mlp.experts.151.gate_proj", "model.layers.59.mlp.experts.152.gate_proj", "model.layers.59.mlp.experts.153.gate_proj", "model.layers.59.mlp.experts.154.gate_proj", "model.layers.59.mlp.experts.155.gate_proj", "model.layers.59.mlp.experts.156.gate_proj", "model.layers.59.mlp.experts.157.gate_proj", "model.layers.59.mlp.experts.158.gate_proj", "model.layers.59.mlp.experts.159.gate_proj", "model.layers.59.mlp.experts.0.up_proj", "model.layers.59.mlp.experts.1.up_proj", "model.layers.59.mlp.experts.2.up_proj", "model.layers.59.mlp.experts.3.up_proj", "model.layers.59.mlp.experts.4.up_proj", "model.layers.59.mlp.experts.5.up_proj", "model.layers.59.mlp.experts.6.up_proj", "model.layers.59.mlp.experts.7.up_proj", "model.layers.59.mlp.experts.8.up_proj", "model.layers.59.mlp.experts.9.up_proj", "model.layers.59.mlp.experts.10.up_proj", "model.layers.59.mlp.experts.11.up_proj", "model.layers.59.mlp.experts.12.up_proj", "model.layers.59.mlp.experts.13.up_proj", "model.layers.59.mlp.experts.14.up_proj", "model.layers.59.mlp.experts.15.up_proj", "model.layers.59.mlp.experts.16.up_proj", "model.layers.59.mlp.experts.17.up_proj", "model.layers.59.mlp.experts.18.up_proj", "model.layers.59.mlp.experts.19.up_proj", "model.layers.59.mlp.experts.20.up_proj", "model.layers.59.mlp.experts.21.up_proj", "model.layers.59.mlp.experts.22.up_proj", "model.layers.59.mlp.experts.23.up_proj", "model.layers.59.mlp.experts.24.up_proj", "model.layers.59.mlp.experts.25.up_proj", "model.layers.59.mlp.experts.26.up_proj", "model.layers.59.mlp.experts.27.up_proj", "model.layers.59.mlp.experts.28.up_proj", "model.layers.59.mlp.experts.29.up_proj", "model.layers.59.mlp.experts.30.up_proj", "model.layers.59.mlp.experts.31.up_proj", "model.layers.59.mlp.experts.32.up_proj", "model.layers.59.mlp.experts.33.up_proj", "model.layers.59.mlp.experts.34.up_proj", "model.layers.59.mlp.experts.35.up_proj", "model.layers.59.mlp.experts.36.up_proj", "model.layers.59.mlp.experts.37.up_proj", "model.layers.59.mlp.experts.38.up_proj", "model.layers.59.mlp.experts.39.up_proj", "model.layers.59.mlp.experts.40.up_proj", "model.layers.59.mlp.experts.41.up_proj", "model.layers.59.mlp.experts.42.up_proj", "model.layers.59.mlp.experts.43.up_proj", "model.layers.59.mlp.experts.44.up_proj", "model.layers.59.mlp.experts.45.up_proj", "model.layers.59.mlp.experts.46.up_proj", "model.layers.59.mlp.experts.47.up_proj", "model.layers.59.mlp.experts.48.up_proj", "model.layers.59.mlp.experts.49.up_proj", "model.layers.59.mlp.experts.50.up_proj", "model.layers.59.mlp.experts.51.up_proj", "model.layers.59.mlp.experts.52.up_proj", "model.layers.59.mlp.experts.53.up_proj", "model.layers.59.mlp.experts.54.up_proj", "model.layers.59.mlp.experts.55.up_proj", "model.layers.59.mlp.experts.56.up_proj", "model.layers.59.mlp.experts.57.up_proj", "model.layers.59.mlp.experts.58.up_proj", "model.layers.59.mlp.experts.59.up_proj", "model.layers.59.mlp.experts.60.up_proj", "model.layers.59.mlp.experts.61.up_proj", "model.layers.59.mlp.experts.62.up_proj", "model.layers.59.mlp.experts.63.up_proj", "model.layers.59.mlp.experts.64.up_proj", "model.layers.59.mlp.experts.65.up_proj", "model.layers.59.mlp.experts.66.up_proj", "model.layers.59.mlp.experts.67.up_proj", "model.layers.59.mlp.experts.68.up_proj", "model.layers.59.mlp.experts.69.up_proj", "model.layers.59.mlp.experts.70.up_proj", "model.layers.59.mlp.experts.71.up_proj", "model.layers.59.mlp.experts.72.up_proj", "model.layers.59.mlp.experts.73.up_proj", "model.layers.59.mlp.experts.74.up_proj", "model.layers.59.mlp.experts.75.up_proj", "model.layers.59.mlp.experts.76.up_proj", "model.layers.59.mlp.experts.77.up_proj", "model.layers.59.mlp.experts.78.up_proj", "model.layers.59.mlp.experts.79.up_proj", "model.layers.59.mlp.experts.80.up_proj", "model.layers.59.mlp.experts.81.up_proj", "model.layers.59.mlp.experts.82.up_proj", "model.layers.59.mlp.experts.83.up_proj", "model.layers.59.mlp.experts.84.up_proj", "model.layers.59.mlp.experts.85.up_proj", "model.layers.59.mlp.experts.86.up_proj", "model.layers.59.mlp.experts.87.up_proj", "model.layers.59.mlp.experts.88.up_proj", "model.layers.59.mlp.experts.89.up_proj", "model.layers.59.mlp.experts.90.up_proj", "model.layers.59.mlp.experts.91.up_proj", "model.layers.59.mlp.experts.92.up_proj", "model.layers.59.mlp.experts.93.up_proj", "model.layers.59.mlp.experts.94.up_proj", "model.layers.59.mlp.experts.95.up_proj", "model.layers.59.mlp.experts.96.up_proj", "model.layers.59.mlp.experts.97.up_proj", "model.layers.59.mlp.experts.98.up_proj", "model.layers.59.mlp.experts.99.up_proj", "model.layers.59.mlp.experts.100.up_proj", "model.layers.59.mlp.experts.101.up_proj", "model.layers.59.mlp.experts.102.up_proj", "model.layers.59.mlp.experts.103.up_proj", "model.layers.59.mlp.experts.104.up_proj", "model.layers.59.mlp.experts.105.up_proj", "model.layers.59.mlp.experts.106.up_proj", "model.layers.59.mlp.experts.107.up_proj", "model.layers.59.mlp.experts.108.up_proj", "model.layers.59.mlp.experts.109.up_proj", "model.layers.59.mlp.experts.110.up_proj", "model.layers.59.mlp.experts.111.up_proj", "model.layers.59.mlp.experts.112.up_proj", "model.layers.59.mlp.experts.113.up_proj", "model.layers.59.mlp.experts.114.up_proj", "model.layers.59.mlp.experts.115.up_proj", "model.layers.59.mlp.experts.116.up_proj", "model.layers.59.mlp.experts.117.up_proj", "model.layers.59.mlp.experts.118.up_proj", "model.layers.59.mlp.experts.119.up_proj", "model.layers.59.mlp.experts.120.up_proj", "model.layers.59.mlp.experts.121.up_proj", "model.layers.59.mlp.experts.122.up_proj", "model.layers.59.mlp.experts.123.up_proj", "model.layers.59.mlp.experts.124.up_proj", "model.layers.59.mlp.experts.125.up_proj", "model.layers.59.mlp.experts.126.up_proj", "model.layers.59.mlp.experts.127.up_proj", "model.layers.59.mlp.experts.128.up_proj", "model.layers.59.mlp.experts.129.up_proj", "model.layers.59.mlp.experts.130.up_proj", "model.layers.59.mlp.experts.131.up_proj", "model.layers.59.mlp.experts.132.up_proj", "model.layers.59.mlp.experts.133.up_proj", "model.layers.59.mlp.experts.134.up_proj", "model.layers.59.mlp.experts.135.up_proj", "model.layers.59.mlp.experts.136.up_proj", "model.layers.59.mlp.experts.137.up_proj", "model.layers.59.mlp.experts.138.up_proj", "model.layers.59.mlp.experts.139.up_proj", "model.layers.59.mlp.experts.140.up_proj", "model.layers.59.mlp.experts.141.up_proj", "model.layers.59.mlp.experts.142.up_proj", "model.layers.59.mlp.experts.143.up_proj", "model.layers.59.mlp.experts.144.up_proj", "model.layers.59.mlp.experts.145.up_proj", "model.layers.59.mlp.experts.146.up_proj", "model.layers.59.mlp.experts.147.up_proj", "model.layers.59.mlp.experts.148.up_proj", "model.layers.59.mlp.experts.149.up_proj", "model.layers.59.mlp.experts.150.up_proj", "model.layers.59.mlp.experts.151.up_proj", "model.layers.59.mlp.experts.152.up_proj", "model.layers.59.mlp.experts.153.up_proj", "model.layers.59.mlp.experts.154.up_proj", "model.layers.59.mlp.experts.155.up_proj", "model.layers.59.mlp.experts.156.up_proj", "model.layers.59.mlp.experts.157.up_proj", "model.layers.59.mlp.experts.158.up_proj", "model.layers.59.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00010538920760155834, "dbits": 2516582400 }, { "dkld": -0.00010450892150402902, "dbits": 5033164800 }, { "dkld": -0.00013083294034005322, "dbits": 7549747200 }, { "dkld": -0.00012762621045113443, "dbits": 12582912000 } ] }, { "idx": 356, "layers": [ "model.layers.59.mlp.experts.0.down_proj", "model.layers.59.mlp.experts.1.down_proj", "model.layers.59.mlp.experts.2.down_proj", "model.layers.59.mlp.experts.3.down_proj", "model.layers.59.mlp.experts.4.down_proj", "model.layers.59.mlp.experts.5.down_proj", "model.layers.59.mlp.experts.6.down_proj", "model.layers.59.mlp.experts.7.down_proj", "model.layers.59.mlp.experts.8.down_proj", "model.layers.59.mlp.experts.9.down_proj", "model.layers.59.mlp.experts.10.down_proj", "model.layers.59.mlp.experts.11.down_proj", "model.layers.59.mlp.experts.12.down_proj", "model.layers.59.mlp.experts.13.down_proj", "model.layers.59.mlp.experts.14.down_proj", "model.layers.59.mlp.experts.15.down_proj", "model.layers.59.mlp.experts.16.down_proj", "model.layers.59.mlp.experts.17.down_proj", "model.layers.59.mlp.experts.18.down_proj", "model.layers.59.mlp.experts.19.down_proj", "model.layers.59.mlp.experts.20.down_proj", "model.layers.59.mlp.experts.21.down_proj", "model.layers.59.mlp.experts.22.down_proj", "model.layers.59.mlp.experts.23.down_proj", "model.layers.59.mlp.experts.24.down_proj", "model.layers.59.mlp.experts.25.down_proj", "model.layers.59.mlp.experts.26.down_proj", "model.layers.59.mlp.experts.27.down_proj", "model.layers.59.mlp.experts.28.down_proj", "model.layers.59.mlp.experts.29.down_proj", "model.layers.59.mlp.experts.30.down_proj", "model.layers.59.mlp.experts.31.down_proj", "model.layers.59.mlp.experts.32.down_proj", "model.layers.59.mlp.experts.33.down_proj", "model.layers.59.mlp.experts.34.down_proj", "model.layers.59.mlp.experts.35.down_proj", "model.layers.59.mlp.experts.36.down_proj", "model.layers.59.mlp.experts.37.down_proj", "model.layers.59.mlp.experts.38.down_proj", "model.layers.59.mlp.experts.39.down_proj", "model.layers.59.mlp.experts.40.down_proj", "model.layers.59.mlp.experts.41.down_proj", "model.layers.59.mlp.experts.42.down_proj", "model.layers.59.mlp.experts.43.down_proj", "model.layers.59.mlp.experts.44.down_proj", "model.layers.59.mlp.experts.45.down_proj", "model.layers.59.mlp.experts.46.down_proj", "model.layers.59.mlp.experts.47.down_proj", "model.layers.59.mlp.experts.48.down_proj", "model.layers.59.mlp.experts.49.down_proj", "model.layers.59.mlp.experts.50.down_proj", "model.layers.59.mlp.experts.51.down_proj", "model.layers.59.mlp.experts.52.down_proj", "model.layers.59.mlp.experts.53.down_proj", "model.layers.59.mlp.experts.54.down_proj", "model.layers.59.mlp.experts.55.down_proj", "model.layers.59.mlp.experts.56.down_proj", "model.layers.59.mlp.experts.57.down_proj", "model.layers.59.mlp.experts.58.down_proj", "model.layers.59.mlp.experts.59.down_proj", "model.layers.59.mlp.experts.60.down_proj", "model.layers.59.mlp.experts.61.down_proj", "model.layers.59.mlp.experts.62.down_proj", "model.layers.59.mlp.experts.63.down_proj", "model.layers.59.mlp.experts.64.down_proj", "model.layers.59.mlp.experts.65.down_proj", "model.layers.59.mlp.experts.66.down_proj", "model.layers.59.mlp.experts.67.down_proj", "model.layers.59.mlp.experts.68.down_proj", "model.layers.59.mlp.experts.69.down_proj", "model.layers.59.mlp.experts.70.down_proj", "model.layers.59.mlp.experts.71.down_proj", "model.layers.59.mlp.experts.72.down_proj", "model.layers.59.mlp.experts.73.down_proj", "model.layers.59.mlp.experts.74.down_proj", "model.layers.59.mlp.experts.75.down_proj", "model.layers.59.mlp.experts.76.down_proj", "model.layers.59.mlp.experts.77.down_proj", "model.layers.59.mlp.experts.78.down_proj", "model.layers.59.mlp.experts.79.down_proj", "model.layers.59.mlp.experts.80.down_proj", "model.layers.59.mlp.experts.81.down_proj", "model.layers.59.mlp.experts.82.down_proj", "model.layers.59.mlp.experts.83.down_proj", "model.layers.59.mlp.experts.84.down_proj", "model.layers.59.mlp.experts.85.down_proj", "model.layers.59.mlp.experts.86.down_proj", "model.layers.59.mlp.experts.87.down_proj", "model.layers.59.mlp.experts.88.down_proj", "model.layers.59.mlp.experts.89.down_proj", "model.layers.59.mlp.experts.90.down_proj", "model.layers.59.mlp.experts.91.down_proj", "model.layers.59.mlp.experts.92.down_proj", "model.layers.59.mlp.experts.93.down_proj", "model.layers.59.mlp.experts.94.down_proj", "model.layers.59.mlp.experts.95.down_proj", "model.layers.59.mlp.experts.96.down_proj", "model.layers.59.mlp.experts.97.down_proj", "model.layers.59.mlp.experts.98.down_proj", "model.layers.59.mlp.experts.99.down_proj", "model.layers.59.mlp.experts.100.down_proj", "model.layers.59.mlp.experts.101.down_proj", "model.layers.59.mlp.experts.102.down_proj", "model.layers.59.mlp.experts.103.down_proj", "model.layers.59.mlp.experts.104.down_proj", "model.layers.59.mlp.experts.105.down_proj", "model.layers.59.mlp.experts.106.down_proj", "model.layers.59.mlp.experts.107.down_proj", "model.layers.59.mlp.experts.108.down_proj", "model.layers.59.mlp.experts.109.down_proj", "model.layers.59.mlp.experts.110.down_proj", "model.layers.59.mlp.experts.111.down_proj", "model.layers.59.mlp.experts.112.down_proj", "model.layers.59.mlp.experts.113.down_proj", "model.layers.59.mlp.experts.114.down_proj", "model.layers.59.mlp.experts.115.down_proj", "model.layers.59.mlp.experts.116.down_proj", "model.layers.59.mlp.experts.117.down_proj", "model.layers.59.mlp.experts.118.down_proj", "model.layers.59.mlp.experts.119.down_proj", "model.layers.59.mlp.experts.120.down_proj", "model.layers.59.mlp.experts.121.down_proj", "model.layers.59.mlp.experts.122.down_proj", "model.layers.59.mlp.experts.123.down_proj", "model.layers.59.mlp.experts.124.down_proj", "model.layers.59.mlp.experts.125.down_proj", "model.layers.59.mlp.experts.126.down_proj", "model.layers.59.mlp.experts.127.down_proj", "model.layers.59.mlp.experts.128.down_proj", "model.layers.59.mlp.experts.129.down_proj", "model.layers.59.mlp.experts.130.down_proj", "model.layers.59.mlp.experts.131.down_proj", "model.layers.59.mlp.experts.132.down_proj", "model.layers.59.mlp.experts.133.down_proj", "model.layers.59.mlp.experts.134.down_proj", "model.layers.59.mlp.experts.135.down_proj", "model.layers.59.mlp.experts.136.down_proj", "model.layers.59.mlp.experts.137.down_proj", "model.layers.59.mlp.experts.138.down_proj", "model.layers.59.mlp.experts.139.down_proj", "model.layers.59.mlp.experts.140.down_proj", "model.layers.59.mlp.experts.141.down_proj", "model.layers.59.mlp.experts.142.down_proj", "model.layers.59.mlp.experts.143.down_proj", "model.layers.59.mlp.experts.144.down_proj", "model.layers.59.mlp.experts.145.down_proj", "model.layers.59.mlp.experts.146.down_proj", "model.layers.59.mlp.experts.147.down_proj", "model.layers.59.mlp.experts.148.down_proj", "model.layers.59.mlp.experts.149.down_proj", "model.layers.59.mlp.experts.150.down_proj", "model.layers.59.mlp.experts.151.down_proj", "model.layers.59.mlp.experts.152.down_proj", "model.layers.59.mlp.experts.153.down_proj", "model.layers.59.mlp.experts.154.down_proj", "model.layers.59.mlp.experts.155.down_proj", "model.layers.59.mlp.experts.156.down_proj", "model.layers.59.mlp.experts.157.down_proj", "model.layers.59.mlp.experts.158.down_proj", "model.layers.59.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": 0.00015131309628485523, "dbits": 1258291200 }, { "dkld": 0.0001144464127719319, "dbits": 2516582400 }, { "dkld": 0.00011574998497962397, "dbits": 3774873600 }, { "dkld": 0.00013600680977104984, "dbits": 6291456000 } ] }, { "idx": 357, "layers": [ "model.layers.60.self_attn.q_proj" ], "candidates": [ { "dkld": 2.461262047290247e-05, "dbits": 62914560 }, { "dkld": -8.295625448227761e-05, "dbits": 125829120 }, { "dkld": -8.747968822718499e-05, "dbits": 188743680 }, { "dkld": -7.66487792134285e-05, "dbits": 314572800 } ] }, { "idx": 358, "layers": [ "model.layers.60.self_attn.k_proj", "model.layers.60.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00018570236861706335, "dbits": 10485760 }, { "dkld": -0.0002881356514990441, "dbits": 20971520 }, { "dkld": -0.0004107192158699091, "dbits": 31457280 }, { "dkld": -0.00035139936953783313, "dbits": 52428800 } ] }, { "idx": 359, "layers": [ "model.layers.60.self_attn.o_proj" ], "candidates": [ { "dkld": -3.301147371531643e-05, "dbits": 62914560 }, { "dkld": -1.5671923756607753e-05, "dbits": 125829120 }, { "dkld": -2.339025959373353e-05, "dbits": 188743680 }, { "dkld": -3.559309989215054e-05, "dbits": 314572800 } ] }, { "idx": 360, "layers": [ "model.layers.60.mlp.shared_experts.gate_proj", "model.layers.60.mlp.shared_experts.up_proj", "model.layers.60.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0002934621647000396, "dbits": 23592960 }, { "dkld": -0.0003915190696716364, "dbits": 47185920 }, { "dkld": -0.0003467523492872715, "dbits": 70778880 }, { "dkld": -0.00034529538825155137, "dbits": 117964800 } ] }, { "idx": 361, "layers": [ "model.layers.60.mlp.experts.0.gate_proj", "model.layers.60.mlp.experts.1.gate_proj", "model.layers.60.mlp.experts.2.gate_proj", "model.layers.60.mlp.experts.3.gate_proj", "model.layers.60.mlp.experts.4.gate_proj", "model.layers.60.mlp.experts.5.gate_proj", "model.layers.60.mlp.experts.6.gate_proj", "model.layers.60.mlp.experts.7.gate_proj", "model.layers.60.mlp.experts.8.gate_proj", "model.layers.60.mlp.experts.9.gate_proj", "model.layers.60.mlp.experts.10.gate_proj", "model.layers.60.mlp.experts.11.gate_proj", "model.layers.60.mlp.experts.12.gate_proj", "model.layers.60.mlp.experts.13.gate_proj", "model.layers.60.mlp.experts.14.gate_proj", "model.layers.60.mlp.experts.15.gate_proj", "model.layers.60.mlp.experts.16.gate_proj", "model.layers.60.mlp.experts.17.gate_proj", "model.layers.60.mlp.experts.18.gate_proj", "model.layers.60.mlp.experts.19.gate_proj", "model.layers.60.mlp.experts.20.gate_proj", "model.layers.60.mlp.experts.21.gate_proj", "model.layers.60.mlp.experts.22.gate_proj", "model.layers.60.mlp.experts.23.gate_proj", "model.layers.60.mlp.experts.24.gate_proj", "model.layers.60.mlp.experts.25.gate_proj", "model.layers.60.mlp.experts.26.gate_proj", "model.layers.60.mlp.experts.27.gate_proj", "model.layers.60.mlp.experts.28.gate_proj", "model.layers.60.mlp.experts.29.gate_proj", "model.layers.60.mlp.experts.30.gate_proj", "model.layers.60.mlp.experts.31.gate_proj", "model.layers.60.mlp.experts.32.gate_proj", "model.layers.60.mlp.experts.33.gate_proj", "model.layers.60.mlp.experts.34.gate_proj", "model.layers.60.mlp.experts.35.gate_proj", "model.layers.60.mlp.experts.36.gate_proj", "model.layers.60.mlp.experts.37.gate_proj", "model.layers.60.mlp.experts.38.gate_proj", "model.layers.60.mlp.experts.39.gate_proj", "model.layers.60.mlp.experts.40.gate_proj", "model.layers.60.mlp.experts.41.gate_proj", "model.layers.60.mlp.experts.42.gate_proj", "model.layers.60.mlp.experts.43.gate_proj", "model.layers.60.mlp.experts.44.gate_proj", "model.layers.60.mlp.experts.45.gate_proj", "model.layers.60.mlp.experts.46.gate_proj", "model.layers.60.mlp.experts.47.gate_proj", "model.layers.60.mlp.experts.48.gate_proj", "model.layers.60.mlp.experts.49.gate_proj", "model.layers.60.mlp.experts.50.gate_proj", "model.layers.60.mlp.experts.51.gate_proj", "model.layers.60.mlp.experts.52.gate_proj", "model.layers.60.mlp.experts.53.gate_proj", "model.layers.60.mlp.experts.54.gate_proj", "model.layers.60.mlp.experts.55.gate_proj", "model.layers.60.mlp.experts.56.gate_proj", "model.layers.60.mlp.experts.57.gate_proj", "model.layers.60.mlp.experts.58.gate_proj", "model.layers.60.mlp.experts.59.gate_proj", "model.layers.60.mlp.experts.60.gate_proj", "model.layers.60.mlp.experts.61.gate_proj", "model.layers.60.mlp.experts.62.gate_proj", "model.layers.60.mlp.experts.63.gate_proj", "model.layers.60.mlp.experts.64.gate_proj", "model.layers.60.mlp.experts.65.gate_proj", "model.layers.60.mlp.experts.66.gate_proj", "model.layers.60.mlp.experts.67.gate_proj", "model.layers.60.mlp.experts.68.gate_proj", "model.layers.60.mlp.experts.69.gate_proj", "model.layers.60.mlp.experts.70.gate_proj", "model.layers.60.mlp.experts.71.gate_proj", "model.layers.60.mlp.experts.72.gate_proj", "model.layers.60.mlp.experts.73.gate_proj", "model.layers.60.mlp.experts.74.gate_proj", "model.layers.60.mlp.experts.75.gate_proj", "model.layers.60.mlp.experts.76.gate_proj", "model.layers.60.mlp.experts.77.gate_proj", "model.layers.60.mlp.experts.78.gate_proj", "model.layers.60.mlp.experts.79.gate_proj", "model.layers.60.mlp.experts.80.gate_proj", "model.layers.60.mlp.experts.81.gate_proj", "model.layers.60.mlp.experts.82.gate_proj", "model.layers.60.mlp.experts.83.gate_proj", "model.layers.60.mlp.experts.84.gate_proj", "model.layers.60.mlp.experts.85.gate_proj", "model.layers.60.mlp.experts.86.gate_proj", "model.layers.60.mlp.experts.87.gate_proj", "model.layers.60.mlp.experts.88.gate_proj", "model.layers.60.mlp.experts.89.gate_proj", "model.layers.60.mlp.experts.90.gate_proj", "model.layers.60.mlp.experts.91.gate_proj", "model.layers.60.mlp.experts.92.gate_proj", "model.layers.60.mlp.experts.93.gate_proj", "model.layers.60.mlp.experts.94.gate_proj", "model.layers.60.mlp.experts.95.gate_proj", "model.layers.60.mlp.experts.96.gate_proj", "model.layers.60.mlp.experts.97.gate_proj", "model.layers.60.mlp.experts.98.gate_proj", "model.layers.60.mlp.experts.99.gate_proj", "model.layers.60.mlp.experts.100.gate_proj", "model.layers.60.mlp.experts.101.gate_proj", "model.layers.60.mlp.experts.102.gate_proj", "model.layers.60.mlp.experts.103.gate_proj", "model.layers.60.mlp.experts.104.gate_proj", "model.layers.60.mlp.experts.105.gate_proj", "model.layers.60.mlp.experts.106.gate_proj", "model.layers.60.mlp.experts.107.gate_proj", "model.layers.60.mlp.experts.108.gate_proj", "model.layers.60.mlp.experts.109.gate_proj", "model.layers.60.mlp.experts.110.gate_proj", "model.layers.60.mlp.experts.111.gate_proj", "model.layers.60.mlp.experts.112.gate_proj", "model.layers.60.mlp.experts.113.gate_proj", "model.layers.60.mlp.experts.114.gate_proj", "model.layers.60.mlp.experts.115.gate_proj", "model.layers.60.mlp.experts.116.gate_proj", "model.layers.60.mlp.experts.117.gate_proj", "model.layers.60.mlp.experts.118.gate_proj", "model.layers.60.mlp.experts.119.gate_proj", "model.layers.60.mlp.experts.120.gate_proj", "model.layers.60.mlp.experts.121.gate_proj", "model.layers.60.mlp.experts.122.gate_proj", "model.layers.60.mlp.experts.123.gate_proj", "model.layers.60.mlp.experts.124.gate_proj", "model.layers.60.mlp.experts.125.gate_proj", "model.layers.60.mlp.experts.126.gate_proj", "model.layers.60.mlp.experts.127.gate_proj", "model.layers.60.mlp.experts.128.gate_proj", "model.layers.60.mlp.experts.129.gate_proj", "model.layers.60.mlp.experts.130.gate_proj", "model.layers.60.mlp.experts.131.gate_proj", "model.layers.60.mlp.experts.132.gate_proj", "model.layers.60.mlp.experts.133.gate_proj", "model.layers.60.mlp.experts.134.gate_proj", "model.layers.60.mlp.experts.135.gate_proj", "model.layers.60.mlp.experts.136.gate_proj", "model.layers.60.mlp.experts.137.gate_proj", "model.layers.60.mlp.experts.138.gate_proj", "model.layers.60.mlp.experts.139.gate_proj", "model.layers.60.mlp.experts.140.gate_proj", "model.layers.60.mlp.experts.141.gate_proj", "model.layers.60.mlp.experts.142.gate_proj", "model.layers.60.mlp.experts.143.gate_proj", "model.layers.60.mlp.experts.144.gate_proj", "model.layers.60.mlp.experts.145.gate_proj", "model.layers.60.mlp.experts.146.gate_proj", "model.layers.60.mlp.experts.147.gate_proj", "model.layers.60.mlp.experts.148.gate_proj", "model.layers.60.mlp.experts.149.gate_proj", "model.layers.60.mlp.experts.150.gate_proj", "model.layers.60.mlp.experts.151.gate_proj", "model.layers.60.mlp.experts.152.gate_proj", "model.layers.60.mlp.experts.153.gate_proj", "model.layers.60.mlp.experts.154.gate_proj", "model.layers.60.mlp.experts.155.gate_proj", "model.layers.60.mlp.experts.156.gate_proj", "model.layers.60.mlp.experts.157.gate_proj", "model.layers.60.mlp.experts.158.gate_proj", "model.layers.60.mlp.experts.159.gate_proj", "model.layers.60.mlp.experts.0.up_proj", "model.layers.60.mlp.experts.1.up_proj", "model.layers.60.mlp.experts.2.up_proj", "model.layers.60.mlp.experts.3.up_proj", "model.layers.60.mlp.experts.4.up_proj", "model.layers.60.mlp.experts.5.up_proj", "model.layers.60.mlp.experts.6.up_proj", "model.layers.60.mlp.experts.7.up_proj", "model.layers.60.mlp.experts.8.up_proj", "model.layers.60.mlp.experts.9.up_proj", "model.layers.60.mlp.experts.10.up_proj", "model.layers.60.mlp.experts.11.up_proj", "model.layers.60.mlp.experts.12.up_proj", "model.layers.60.mlp.experts.13.up_proj", "model.layers.60.mlp.experts.14.up_proj", "model.layers.60.mlp.experts.15.up_proj", "model.layers.60.mlp.experts.16.up_proj", "model.layers.60.mlp.experts.17.up_proj", "model.layers.60.mlp.experts.18.up_proj", "model.layers.60.mlp.experts.19.up_proj", "model.layers.60.mlp.experts.20.up_proj", "model.layers.60.mlp.experts.21.up_proj", "model.layers.60.mlp.experts.22.up_proj", "model.layers.60.mlp.experts.23.up_proj", "model.layers.60.mlp.experts.24.up_proj", "model.layers.60.mlp.experts.25.up_proj", "model.layers.60.mlp.experts.26.up_proj", "model.layers.60.mlp.experts.27.up_proj", "model.layers.60.mlp.experts.28.up_proj", "model.layers.60.mlp.experts.29.up_proj", "model.layers.60.mlp.experts.30.up_proj", "model.layers.60.mlp.experts.31.up_proj", "model.layers.60.mlp.experts.32.up_proj", "model.layers.60.mlp.experts.33.up_proj", "model.layers.60.mlp.experts.34.up_proj", "model.layers.60.mlp.experts.35.up_proj", "model.layers.60.mlp.experts.36.up_proj", "model.layers.60.mlp.experts.37.up_proj", "model.layers.60.mlp.experts.38.up_proj", "model.layers.60.mlp.experts.39.up_proj", "model.layers.60.mlp.experts.40.up_proj", "model.layers.60.mlp.experts.41.up_proj", "model.layers.60.mlp.experts.42.up_proj", "model.layers.60.mlp.experts.43.up_proj", "model.layers.60.mlp.experts.44.up_proj", "model.layers.60.mlp.experts.45.up_proj", "model.layers.60.mlp.experts.46.up_proj", "model.layers.60.mlp.experts.47.up_proj", "model.layers.60.mlp.experts.48.up_proj", "model.layers.60.mlp.experts.49.up_proj", "model.layers.60.mlp.experts.50.up_proj", "model.layers.60.mlp.experts.51.up_proj", "model.layers.60.mlp.experts.52.up_proj", "model.layers.60.mlp.experts.53.up_proj", "model.layers.60.mlp.experts.54.up_proj", "model.layers.60.mlp.experts.55.up_proj", "model.layers.60.mlp.experts.56.up_proj", "model.layers.60.mlp.experts.57.up_proj", "model.layers.60.mlp.experts.58.up_proj", "model.layers.60.mlp.experts.59.up_proj", "model.layers.60.mlp.experts.60.up_proj", "model.layers.60.mlp.experts.61.up_proj", "model.layers.60.mlp.experts.62.up_proj", "model.layers.60.mlp.experts.63.up_proj", "model.layers.60.mlp.experts.64.up_proj", "model.layers.60.mlp.experts.65.up_proj", "model.layers.60.mlp.experts.66.up_proj", "model.layers.60.mlp.experts.67.up_proj", "model.layers.60.mlp.experts.68.up_proj", "model.layers.60.mlp.experts.69.up_proj", "model.layers.60.mlp.experts.70.up_proj", "model.layers.60.mlp.experts.71.up_proj", "model.layers.60.mlp.experts.72.up_proj", "model.layers.60.mlp.experts.73.up_proj", "model.layers.60.mlp.experts.74.up_proj", "model.layers.60.mlp.experts.75.up_proj", "model.layers.60.mlp.experts.76.up_proj", "model.layers.60.mlp.experts.77.up_proj", "model.layers.60.mlp.experts.78.up_proj", "model.layers.60.mlp.experts.79.up_proj", "model.layers.60.mlp.experts.80.up_proj", "model.layers.60.mlp.experts.81.up_proj", "model.layers.60.mlp.experts.82.up_proj", "model.layers.60.mlp.experts.83.up_proj", "model.layers.60.mlp.experts.84.up_proj", "model.layers.60.mlp.experts.85.up_proj", "model.layers.60.mlp.experts.86.up_proj", "model.layers.60.mlp.experts.87.up_proj", "model.layers.60.mlp.experts.88.up_proj", "model.layers.60.mlp.experts.89.up_proj", "model.layers.60.mlp.experts.90.up_proj", "model.layers.60.mlp.experts.91.up_proj", "model.layers.60.mlp.experts.92.up_proj", "model.layers.60.mlp.experts.93.up_proj", "model.layers.60.mlp.experts.94.up_proj", "model.layers.60.mlp.experts.95.up_proj", "model.layers.60.mlp.experts.96.up_proj", "model.layers.60.mlp.experts.97.up_proj", "model.layers.60.mlp.experts.98.up_proj", "model.layers.60.mlp.experts.99.up_proj", "model.layers.60.mlp.experts.100.up_proj", "model.layers.60.mlp.experts.101.up_proj", "model.layers.60.mlp.experts.102.up_proj", "model.layers.60.mlp.experts.103.up_proj", "model.layers.60.mlp.experts.104.up_proj", "model.layers.60.mlp.experts.105.up_proj", "model.layers.60.mlp.experts.106.up_proj", "model.layers.60.mlp.experts.107.up_proj", "model.layers.60.mlp.experts.108.up_proj", "model.layers.60.mlp.experts.109.up_proj", "model.layers.60.mlp.experts.110.up_proj", "model.layers.60.mlp.experts.111.up_proj", "model.layers.60.mlp.experts.112.up_proj", "model.layers.60.mlp.experts.113.up_proj", "model.layers.60.mlp.experts.114.up_proj", "model.layers.60.mlp.experts.115.up_proj", "model.layers.60.mlp.experts.116.up_proj", "model.layers.60.mlp.experts.117.up_proj", "model.layers.60.mlp.experts.118.up_proj", "model.layers.60.mlp.experts.119.up_proj", "model.layers.60.mlp.experts.120.up_proj", "model.layers.60.mlp.experts.121.up_proj", "model.layers.60.mlp.experts.122.up_proj", "model.layers.60.mlp.experts.123.up_proj", "model.layers.60.mlp.experts.124.up_proj", "model.layers.60.mlp.experts.125.up_proj", "model.layers.60.mlp.experts.126.up_proj", "model.layers.60.mlp.experts.127.up_proj", "model.layers.60.mlp.experts.128.up_proj", "model.layers.60.mlp.experts.129.up_proj", "model.layers.60.mlp.experts.130.up_proj", "model.layers.60.mlp.experts.131.up_proj", "model.layers.60.mlp.experts.132.up_proj", "model.layers.60.mlp.experts.133.up_proj", "model.layers.60.mlp.experts.134.up_proj", "model.layers.60.mlp.experts.135.up_proj", "model.layers.60.mlp.experts.136.up_proj", "model.layers.60.mlp.experts.137.up_proj", "model.layers.60.mlp.experts.138.up_proj", "model.layers.60.mlp.experts.139.up_proj", "model.layers.60.mlp.experts.140.up_proj", "model.layers.60.mlp.experts.141.up_proj", "model.layers.60.mlp.experts.142.up_proj", "model.layers.60.mlp.experts.143.up_proj", "model.layers.60.mlp.experts.144.up_proj", "model.layers.60.mlp.experts.145.up_proj", "model.layers.60.mlp.experts.146.up_proj", "model.layers.60.mlp.experts.147.up_proj", "model.layers.60.mlp.experts.148.up_proj", "model.layers.60.mlp.experts.149.up_proj", "model.layers.60.mlp.experts.150.up_proj", "model.layers.60.mlp.experts.151.up_proj", "model.layers.60.mlp.experts.152.up_proj", "model.layers.60.mlp.experts.153.up_proj", "model.layers.60.mlp.experts.154.up_proj", "model.layers.60.mlp.experts.155.up_proj", "model.layers.60.mlp.experts.156.up_proj", "model.layers.60.mlp.experts.157.up_proj", "model.layers.60.mlp.experts.158.up_proj", "model.layers.60.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00017812624573708413, "dbits": 2516582400 }, { "dkld": -0.00019735237583518028, "dbits": 5033164800 }, { "dkld": -0.00017755106091499329, "dbits": 7549747200 }, { "dkld": -0.00018747206777335046, "dbits": 12582912000 } ] }, { "idx": 362, "layers": [ "model.layers.60.mlp.experts.0.down_proj", "model.layers.60.mlp.experts.1.down_proj", "model.layers.60.mlp.experts.2.down_proj", "model.layers.60.mlp.experts.3.down_proj", "model.layers.60.mlp.experts.4.down_proj", "model.layers.60.mlp.experts.5.down_proj", "model.layers.60.mlp.experts.6.down_proj", "model.layers.60.mlp.experts.7.down_proj", "model.layers.60.mlp.experts.8.down_proj", "model.layers.60.mlp.experts.9.down_proj", "model.layers.60.mlp.experts.10.down_proj", "model.layers.60.mlp.experts.11.down_proj", "model.layers.60.mlp.experts.12.down_proj", "model.layers.60.mlp.experts.13.down_proj", "model.layers.60.mlp.experts.14.down_proj", "model.layers.60.mlp.experts.15.down_proj", "model.layers.60.mlp.experts.16.down_proj", "model.layers.60.mlp.experts.17.down_proj", "model.layers.60.mlp.experts.18.down_proj", "model.layers.60.mlp.experts.19.down_proj", "model.layers.60.mlp.experts.20.down_proj", "model.layers.60.mlp.experts.21.down_proj", "model.layers.60.mlp.experts.22.down_proj", "model.layers.60.mlp.experts.23.down_proj", "model.layers.60.mlp.experts.24.down_proj", "model.layers.60.mlp.experts.25.down_proj", "model.layers.60.mlp.experts.26.down_proj", "model.layers.60.mlp.experts.27.down_proj", "model.layers.60.mlp.experts.28.down_proj", "model.layers.60.mlp.experts.29.down_proj", "model.layers.60.mlp.experts.30.down_proj", "model.layers.60.mlp.experts.31.down_proj", "model.layers.60.mlp.experts.32.down_proj", "model.layers.60.mlp.experts.33.down_proj", "model.layers.60.mlp.experts.34.down_proj", "model.layers.60.mlp.experts.35.down_proj", "model.layers.60.mlp.experts.36.down_proj", "model.layers.60.mlp.experts.37.down_proj", "model.layers.60.mlp.experts.38.down_proj", "model.layers.60.mlp.experts.39.down_proj", "model.layers.60.mlp.experts.40.down_proj", "model.layers.60.mlp.experts.41.down_proj", "model.layers.60.mlp.experts.42.down_proj", "model.layers.60.mlp.experts.43.down_proj", "model.layers.60.mlp.experts.44.down_proj", "model.layers.60.mlp.experts.45.down_proj", "model.layers.60.mlp.experts.46.down_proj", "model.layers.60.mlp.experts.47.down_proj", "model.layers.60.mlp.experts.48.down_proj", "model.layers.60.mlp.experts.49.down_proj", "model.layers.60.mlp.experts.50.down_proj", "model.layers.60.mlp.experts.51.down_proj", "model.layers.60.mlp.experts.52.down_proj", "model.layers.60.mlp.experts.53.down_proj", "model.layers.60.mlp.experts.54.down_proj", "model.layers.60.mlp.experts.55.down_proj", "model.layers.60.mlp.experts.56.down_proj", "model.layers.60.mlp.experts.57.down_proj", "model.layers.60.mlp.experts.58.down_proj", "model.layers.60.mlp.experts.59.down_proj", "model.layers.60.mlp.experts.60.down_proj", "model.layers.60.mlp.experts.61.down_proj", "model.layers.60.mlp.experts.62.down_proj", "model.layers.60.mlp.experts.63.down_proj", "model.layers.60.mlp.experts.64.down_proj", "model.layers.60.mlp.experts.65.down_proj", "model.layers.60.mlp.experts.66.down_proj", "model.layers.60.mlp.experts.67.down_proj", "model.layers.60.mlp.experts.68.down_proj", "model.layers.60.mlp.experts.69.down_proj", "model.layers.60.mlp.experts.70.down_proj", "model.layers.60.mlp.experts.71.down_proj", "model.layers.60.mlp.experts.72.down_proj", "model.layers.60.mlp.experts.73.down_proj", "model.layers.60.mlp.experts.74.down_proj", "model.layers.60.mlp.experts.75.down_proj", "model.layers.60.mlp.experts.76.down_proj", "model.layers.60.mlp.experts.77.down_proj", "model.layers.60.mlp.experts.78.down_proj", "model.layers.60.mlp.experts.79.down_proj", "model.layers.60.mlp.experts.80.down_proj", "model.layers.60.mlp.experts.81.down_proj", "model.layers.60.mlp.experts.82.down_proj", "model.layers.60.mlp.experts.83.down_proj", "model.layers.60.mlp.experts.84.down_proj", "model.layers.60.mlp.experts.85.down_proj", "model.layers.60.mlp.experts.86.down_proj", "model.layers.60.mlp.experts.87.down_proj", "model.layers.60.mlp.experts.88.down_proj", "model.layers.60.mlp.experts.89.down_proj", "model.layers.60.mlp.experts.90.down_proj", "model.layers.60.mlp.experts.91.down_proj", "model.layers.60.mlp.experts.92.down_proj", "model.layers.60.mlp.experts.93.down_proj", "model.layers.60.mlp.experts.94.down_proj", "model.layers.60.mlp.experts.95.down_proj", "model.layers.60.mlp.experts.96.down_proj", "model.layers.60.mlp.experts.97.down_proj", "model.layers.60.mlp.experts.98.down_proj", "model.layers.60.mlp.experts.99.down_proj", "model.layers.60.mlp.experts.100.down_proj", "model.layers.60.mlp.experts.101.down_proj", "model.layers.60.mlp.experts.102.down_proj", "model.layers.60.mlp.experts.103.down_proj", "model.layers.60.mlp.experts.104.down_proj", "model.layers.60.mlp.experts.105.down_proj", "model.layers.60.mlp.experts.106.down_proj", "model.layers.60.mlp.experts.107.down_proj", "model.layers.60.mlp.experts.108.down_proj", "model.layers.60.mlp.experts.109.down_proj", "model.layers.60.mlp.experts.110.down_proj", "model.layers.60.mlp.experts.111.down_proj", "model.layers.60.mlp.experts.112.down_proj", "model.layers.60.mlp.experts.113.down_proj", "model.layers.60.mlp.experts.114.down_proj", "model.layers.60.mlp.experts.115.down_proj", "model.layers.60.mlp.experts.116.down_proj", "model.layers.60.mlp.experts.117.down_proj", "model.layers.60.mlp.experts.118.down_proj", "model.layers.60.mlp.experts.119.down_proj", "model.layers.60.mlp.experts.120.down_proj", "model.layers.60.mlp.experts.121.down_proj", "model.layers.60.mlp.experts.122.down_proj", "model.layers.60.mlp.experts.123.down_proj", "model.layers.60.mlp.experts.124.down_proj", "model.layers.60.mlp.experts.125.down_proj", "model.layers.60.mlp.experts.126.down_proj", "model.layers.60.mlp.experts.127.down_proj", "model.layers.60.mlp.experts.128.down_proj", "model.layers.60.mlp.experts.129.down_proj", "model.layers.60.mlp.experts.130.down_proj", "model.layers.60.mlp.experts.131.down_proj", "model.layers.60.mlp.experts.132.down_proj", "model.layers.60.mlp.experts.133.down_proj", "model.layers.60.mlp.experts.134.down_proj", "model.layers.60.mlp.experts.135.down_proj", "model.layers.60.mlp.experts.136.down_proj", "model.layers.60.mlp.experts.137.down_proj", "model.layers.60.mlp.experts.138.down_proj", "model.layers.60.mlp.experts.139.down_proj", "model.layers.60.mlp.experts.140.down_proj", "model.layers.60.mlp.experts.141.down_proj", "model.layers.60.mlp.experts.142.down_proj", "model.layers.60.mlp.experts.143.down_proj", "model.layers.60.mlp.experts.144.down_proj", "model.layers.60.mlp.experts.145.down_proj", "model.layers.60.mlp.experts.146.down_proj", "model.layers.60.mlp.experts.147.down_proj", "model.layers.60.mlp.experts.148.down_proj", "model.layers.60.mlp.experts.149.down_proj", "model.layers.60.mlp.experts.150.down_proj", "model.layers.60.mlp.experts.151.down_proj", "model.layers.60.mlp.experts.152.down_proj", "model.layers.60.mlp.experts.153.down_proj", "model.layers.60.mlp.experts.154.down_proj", "model.layers.60.mlp.experts.155.down_proj", "model.layers.60.mlp.experts.156.down_proj", "model.layers.60.mlp.experts.157.down_proj", "model.layers.60.mlp.experts.158.down_proj", "model.layers.60.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -6.385073065758029e-05, "dbits": 1258291200 }, { "dkld": -0.0001408616080880165, "dbits": 2516582400 }, { "dkld": -0.00011246688663960058, "dbits": 3774873600 }, { "dkld": -0.00013596266508103527, "dbits": 6291456000 } ] }, { "idx": 363, "layers": [ "model.layers.61.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00022274386137724478, "dbits": 62914560 }, { "dkld": -0.0002276359125971822, "dbits": 125829120 }, { "dkld": -0.00020009446889163174, "dbits": 188743680 }, { "dkld": -0.00018714647740125656, "dbits": 314572800 } ] }, { "idx": 364, "layers": [ "model.layers.61.self_attn.k_proj", "model.layers.61.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0002492995001375786, "dbits": 10485760 }, { "dkld": -0.0002562415786087624, "dbits": 20971520 }, { "dkld": -0.00022448031231761534, "dbits": 31457280 }, { "dkld": -0.00022483682259917537, "dbits": 52428800 } ] }, { "idx": 365, "layers": [ "model.layers.61.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00017002467066050408, "dbits": 62914560 }, { "dkld": -0.00021152645349503674, "dbits": 125829120 }, { "dkld": -0.00020147897303104956, "dbits": 188743680 }, { "dkld": -0.00022636232897639552, "dbits": 314572800 } ] }, { "idx": 366, "layers": [ "model.layers.61.mlp.shared_experts.gate_proj", "model.layers.61.mlp.shared_experts.up_proj", "model.layers.61.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00011938195675612051, "dbits": 23592960 }, { "dkld": -0.00010298024863004962, "dbits": 47185920 }, { "dkld": -0.00016355970874429304, "dbits": 70778880 }, { "dkld": -0.00011778920888901034, "dbits": 117964800 } ] }, { "idx": 367, "layers": [ "model.layers.61.mlp.experts.0.gate_proj", "model.layers.61.mlp.experts.1.gate_proj", "model.layers.61.mlp.experts.2.gate_proj", "model.layers.61.mlp.experts.3.gate_proj", "model.layers.61.mlp.experts.4.gate_proj", "model.layers.61.mlp.experts.5.gate_proj", "model.layers.61.mlp.experts.6.gate_proj", "model.layers.61.mlp.experts.7.gate_proj", "model.layers.61.mlp.experts.8.gate_proj", "model.layers.61.mlp.experts.9.gate_proj", "model.layers.61.mlp.experts.10.gate_proj", "model.layers.61.mlp.experts.11.gate_proj", "model.layers.61.mlp.experts.12.gate_proj", "model.layers.61.mlp.experts.13.gate_proj", "model.layers.61.mlp.experts.14.gate_proj", "model.layers.61.mlp.experts.15.gate_proj", "model.layers.61.mlp.experts.16.gate_proj", "model.layers.61.mlp.experts.17.gate_proj", "model.layers.61.mlp.experts.18.gate_proj", "model.layers.61.mlp.experts.19.gate_proj", "model.layers.61.mlp.experts.20.gate_proj", "model.layers.61.mlp.experts.21.gate_proj", "model.layers.61.mlp.experts.22.gate_proj", "model.layers.61.mlp.experts.23.gate_proj", "model.layers.61.mlp.experts.24.gate_proj", "model.layers.61.mlp.experts.25.gate_proj", "model.layers.61.mlp.experts.26.gate_proj", "model.layers.61.mlp.experts.27.gate_proj", "model.layers.61.mlp.experts.28.gate_proj", "model.layers.61.mlp.experts.29.gate_proj", "model.layers.61.mlp.experts.30.gate_proj", "model.layers.61.mlp.experts.31.gate_proj", "model.layers.61.mlp.experts.32.gate_proj", "model.layers.61.mlp.experts.33.gate_proj", "model.layers.61.mlp.experts.34.gate_proj", "model.layers.61.mlp.experts.35.gate_proj", "model.layers.61.mlp.experts.36.gate_proj", "model.layers.61.mlp.experts.37.gate_proj", "model.layers.61.mlp.experts.38.gate_proj", "model.layers.61.mlp.experts.39.gate_proj", "model.layers.61.mlp.experts.40.gate_proj", "model.layers.61.mlp.experts.41.gate_proj", "model.layers.61.mlp.experts.42.gate_proj", "model.layers.61.mlp.experts.43.gate_proj", "model.layers.61.mlp.experts.44.gate_proj", "model.layers.61.mlp.experts.45.gate_proj", "model.layers.61.mlp.experts.46.gate_proj", "model.layers.61.mlp.experts.47.gate_proj", "model.layers.61.mlp.experts.48.gate_proj", "model.layers.61.mlp.experts.49.gate_proj", "model.layers.61.mlp.experts.50.gate_proj", "model.layers.61.mlp.experts.51.gate_proj", "model.layers.61.mlp.experts.52.gate_proj", "model.layers.61.mlp.experts.53.gate_proj", "model.layers.61.mlp.experts.54.gate_proj", "model.layers.61.mlp.experts.55.gate_proj", "model.layers.61.mlp.experts.56.gate_proj", "model.layers.61.mlp.experts.57.gate_proj", "model.layers.61.mlp.experts.58.gate_proj", "model.layers.61.mlp.experts.59.gate_proj", "model.layers.61.mlp.experts.60.gate_proj", "model.layers.61.mlp.experts.61.gate_proj", "model.layers.61.mlp.experts.62.gate_proj", "model.layers.61.mlp.experts.63.gate_proj", "model.layers.61.mlp.experts.64.gate_proj", "model.layers.61.mlp.experts.65.gate_proj", "model.layers.61.mlp.experts.66.gate_proj", "model.layers.61.mlp.experts.67.gate_proj", "model.layers.61.mlp.experts.68.gate_proj", "model.layers.61.mlp.experts.69.gate_proj", "model.layers.61.mlp.experts.70.gate_proj", "model.layers.61.mlp.experts.71.gate_proj", "model.layers.61.mlp.experts.72.gate_proj", "model.layers.61.mlp.experts.73.gate_proj", "model.layers.61.mlp.experts.74.gate_proj", "model.layers.61.mlp.experts.75.gate_proj", "model.layers.61.mlp.experts.76.gate_proj", "model.layers.61.mlp.experts.77.gate_proj", "model.layers.61.mlp.experts.78.gate_proj", "model.layers.61.mlp.experts.79.gate_proj", "model.layers.61.mlp.experts.80.gate_proj", "model.layers.61.mlp.experts.81.gate_proj", "model.layers.61.mlp.experts.82.gate_proj", "model.layers.61.mlp.experts.83.gate_proj", "model.layers.61.mlp.experts.84.gate_proj", "model.layers.61.mlp.experts.85.gate_proj", "model.layers.61.mlp.experts.86.gate_proj", "model.layers.61.mlp.experts.87.gate_proj", "model.layers.61.mlp.experts.88.gate_proj", "model.layers.61.mlp.experts.89.gate_proj", "model.layers.61.mlp.experts.90.gate_proj", "model.layers.61.mlp.experts.91.gate_proj", "model.layers.61.mlp.experts.92.gate_proj", "model.layers.61.mlp.experts.93.gate_proj", "model.layers.61.mlp.experts.94.gate_proj", "model.layers.61.mlp.experts.95.gate_proj", "model.layers.61.mlp.experts.96.gate_proj", "model.layers.61.mlp.experts.97.gate_proj", "model.layers.61.mlp.experts.98.gate_proj", "model.layers.61.mlp.experts.99.gate_proj", "model.layers.61.mlp.experts.100.gate_proj", "model.layers.61.mlp.experts.101.gate_proj", "model.layers.61.mlp.experts.102.gate_proj", "model.layers.61.mlp.experts.103.gate_proj", "model.layers.61.mlp.experts.104.gate_proj", "model.layers.61.mlp.experts.105.gate_proj", "model.layers.61.mlp.experts.106.gate_proj", "model.layers.61.mlp.experts.107.gate_proj", "model.layers.61.mlp.experts.108.gate_proj", "model.layers.61.mlp.experts.109.gate_proj", "model.layers.61.mlp.experts.110.gate_proj", "model.layers.61.mlp.experts.111.gate_proj", "model.layers.61.mlp.experts.112.gate_proj", "model.layers.61.mlp.experts.113.gate_proj", "model.layers.61.mlp.experts.114.gate_proj", "model.layers.61.mlp.experts.115.gate_proj", "model.layers.61.mlp.experts.116.gate_proj", "model.layers.61.mlp.experts.117.gate_proj", "model.layers.61.mlp.experts.118.gate_proj", "model.layers.61.mlp.experts.119.gate_proj", "model.layers.61.mlp.experts.120.gate_proj", "model.layers.61.mlp.experts.121.gate_proj", "model.layers.61.mlp.experts.122.gate_proj", "model.layers.61.mlp.experts.123.gate_proj", "model.layers.61.mlp.experts.124.gate_proj", "model.layers.61.mlp.experts.125.gate_proj", "model.layers.61.mlp.experts.126.gate_proj", "model.layers.61.mlp.experts.127.gate_proj", "model.layers.61.mlp.experts.128.gate_proj", "model.layers.61.mlp.experts.129.gate_proj", "model.layers.61.mlp.experts.130.gate_proj", "model.layers.61.mlp.experts.131.gate_proj", "model.layers.61.mlp.experts.132.gate_proj", "model.layers.61.mlp.experts.133.gate_proj", "model.layers.61.mlp.experts.134.gate_proj", "model.layers.61.mlp.experts.135.gate_proj", "model.layers.61.mlp.experts.136.gate_proj", "model.layers.61.mlp.experts.137.gate_proj", "model.layers.61.mlp.experts.138.gate_proj", "model.layers.61.mlp.experts.139.gate_proj", "model.layers.61.mlp.experts.140.gate_proj", "model.layers.61.mlp.experts.141.gate_proj", "model.layers.61.mlp.experts.142.gate_proj", "model.layers.61.mlp.experts.143.gate_proj", "model.layers.61.mlp.experts.144.gate_proj", "model.layers.61.mlp.experts.145.gate_proj", "model.layers.61.mlp.experts.146.gate_proj", "model.layers.61.mlp.experts.147.gate_proj", "model.layers.61.mlp.experts.148.gate_proj", "model.layers.61.mlp.experts.149.gate_proj", "model.layers.61.mlp.experts.150.gate_proj", "model.layers.61.mlp.experts.151.gate_proj", "model.layers.61.mlp.experts.152.gate_proj", "model.layers.61.mlp.experts.153.gate_proj", "model.layers.61.mlp.experts.154.gate_proj", "model.layers.61.mlp.experts.155.gate_proj", "model.layers.61.mlp.experts.156.gate_proj", "model.layers.61.mlp.experts.157.gate_proj", "model.layers.61.mlp.experts.158.gate_proj", "model.layers.61.mlp.experts.159.gate_proj", "model.layers.61.mlp.experts.0.up_proj", "model.layers.61.mlp.experts.1.up_proj", "model.layers.61.mlp.experts.2.up_proj", "model.layers.61.mlp.experts.3.up_proj", "model.layers.61.mlp.experts.4.up_proj", "model.layers.61.mlp.experts.5.up_proj", "model.layers.61.mlp.experts.6.up_proj", "model.layers.61.mlp.experts.7.up_proj", "model.layers.61.mlp.experts.8.up_proj", "model.layers.61.mlp.experts.9.up_proj", "model.layers.61.mlp.experts.10.up_proj", "model.layers.61.mlp.experts.11.up_proj", "model.layers.61.mlp.experts.12.up_proj", "model.layers.61.mlp.experts.13.up_proj", "model.layers.61.mlp.experts.14.up_proj", "model.layers.61.mlp.experts.15.up_proj", "model.layers.61.mlp.experts.16.up_proj", "model.layers.61.mlp.experts.17.up_proj", "model.layers.61.mlp.experts.18.up_proj", "model.layers.61.mlp.experts.19.up_proj", "model.layers.61.mlp.experts.20.up_proj", "model.layers.61.mlp.experts.21.up_proj", "model.layers.61.mlp.experts.22.up_proj", "model.layers.61.mlp.experts.23.up_proj", "model.layers.61.mlp.experts.24.up_proj", "model.layers.61.mlp.experts.25.up_proj", "model.layers.61.mlp.experts.26.up_proj", "model.layers.61.mlp.experts.27.up_proj", "model.layers.61.mlp.experts.28.up_proj", "model.layers.61.mlp.experts.29.up_proj", "model.layers.61.mlp.experts.30.up_proj", "model.layers.61.mlp.experts.31.up_proj", "model.layers.61.mlp.experts.32.up_proj", "model.layers.61.mlp.experts.33.up_proj", "model.layers.61.mlp.experts.34.up_proj", "model.layers.61.mlp.experts.35.up_proj", "model.layers.61.mlp.experts.36.up_proj", "model.layers.61.mlp.experts.37.up_proj", "model.layers.61.mlp.experts.38.up_proj", "model.layers.61.mlp.experts.39.up_proj", "model.layers.61.mlp.experts.40.up_proj", "model.layers.61.mlp.experts.41.up_proj", "model.layers.61.mlp.experts.42.up_proj", "model.layers.61.mlp.experts.43.up_proj", "model.layers.61.mlp.experts.44.up_proj", "model.layers.61.mlp.experts.45.up_proj", "model.layers.61.mlp.experts.46.up_proj", "model.layers.61.mlp.experts.47.up_proj", "model.layers.61.mlp.experts.48.up_proj", "model.layers.61.mlp.experts.49.up_proj", "model.layers.61.mlp.experts.50.up_proj", "model.layers.61.mlp.experts.51.up_proj", "model.layers.61.mlp.experts.52.up_proj", "model.layers.61.mlp.experts.53.up_proj", "model.layers.61.mlp.experts.54.up_proj", "model.layers.61.mlp.experts.55.up_proj", "model.layers.61.mlp.experts.56.up_proj", "model.layers.61.mlp.experts.57.up_proj", "model.layers.61.mlp.experts.58.up_proj", "model.layers.61.mlp.experts.59.up_proj", "model.layers.61.mlp.experts.60.up_proj", "model.layers.61.mlp.experts.61.up_proj", "model.layers.61.mlp.experts.62.up_proj", "model.layers.61.mlp.experts.63.up_proj", "model.layers.61.mlp.experts.64.up_proj", "model.layers.61.mlp.experts.65.up_proj", "model.layers.61.mlp.experts.66.up_proj", "model.layers.61.mlp.experts.67.up_proj", "model.layers.61.mlp.experts.68.up_proj", "model.layers.61.mlp.experts.69.up_proj", "model.layers.61.mlp.experts.70.up_proj", "model.layers.61.mlp.experts.71.up_proj", "model.layers.61.mlp.experts.72.up_proj", "model.layers.61.mlp.experts.73.up_proj", "model.layers.61.mlp.experts.74.up_proj", "model.layers.61.mlp.experts.75.up_proj", "model.layers.61.mlp.experts.76.up_proj", "model.layers.61.mlp.experts.77.up_proj", "model.layers.61.mlp.experts.78.up_proj", "model.layers.61.mlp.experts.79.up_proj", "model.layers.61.mlp.experts.80.up_proj", "model.layers.61.mlp.experts.81.up_proj", "model.layers.61.mlp.experts.82.up_proj", "model.layers.61.mlp.experts.83.up_proj", "model.layers.61.mlp.experts.84.up_proj", "model.layers.61.mlp.experts.85.up_proj", "model.layers.61.mlp.experts.86.up_proj", "model.layers.61.mlp.experts.87.up_proj", "model.layers.61.mlp.experts.88.up_proj", "model.layers.61.mlp.experts.89.up_proj", "model.layers.61.mlp.experts.90.up_proj", "model.layers.61.mlp.experts.91.up_proj", "model.layers.61.mlp.experts.92.up_proj", "model.layers.61.mlp.experts.93.up_proj", "model.layers.61.mlp.experts.94.up_proj", "model.layers.61.mlp.experts.95.up_proj", "model.layers.61.mlp.experts.96.up_proj", "model.layers.61.mlp.experts.97.up_proj", "model.layers.61.mlp.experts.98.up_proj", "model.layers.61.mlp.experts.99.up_proj", "model.layers.61.mlp.experts.100.up_proj", "model.layers.61.mlp.experts.101.up_proj", "model.layers.61.mlp.experts.102.up_proj", "model.layers.61.mlp.experts.103.up_proj", "model.layers.61.mlp.experts.104.up_proj", "model.layers.61.mlp.experts.105.up_proj", "model.layers.61.mlp.experts.106.up_proj", "model.layers.61.mlp.experts.107.up_proj", "model.layers.61.mlp.experts.108.up_proj", "model.layers.61.mlp.experts.109.up_proj", "model.layers.61.mlp.experts.110.up_proj", "model.layers.61.mlp.experts.111.up_proj", "model.layers.61.mlp.experts.112.up_proj", "model.layers.61.mlp.experts.113.up_proj", "model.layers.61.mlp.experts.114.up_proj", "model.layers.61.mlp.experts.115.up_proj", "model.layers.61.mlp.experts.116.up_proj", "model.layers.61.mlp.experts.117.up_proj", "model.layers.61.mlp.experts.118.up_proj", "model.layers.61.mlp.experts.119.up_proj", "model.layers.61.mlp.experts.120.up_proj", "model.layers.61.mlp.experts.121.up_proj", "model.layers.61.mlp.experts.122.up_proj", "model.layers.61.mlp.experts.123.up_proj", "model.layers.61.mlp.experts.124.up_proj", "model.layers.61.mlp.experts.125.up_proj", "model.layers.61.mlp.experts.126.up_proj", "model.layers.61.mlp.experts.127.up_proj", "model.layers.61.mlp.experts.128.up_proj", "model.layers.61.mlp.experts.129.up_proj", "model.layers.61.mlp.experts.130.up_proj", "model.layers.61.mlp.experts.131.up_proj", "model.layers.61.mlp.experts.132.up_proj", "model.layers.61.mlp.experts.133.up_proj", "model.layers.61.mlp.experts.134.up_proj", "model.layers.61.mlp.experts.135.up_proj", "model.layers.61.mlp.experts.136.up_proj", "model.layers.61.mlp.experts.137.up_proj", "model.layers.61.mlp.experts.138.up_proj", "model.layers.61.mlp.experts.139.up_proj", "model.layers.61.mlp.experts.140.up_proj", "model.layers.61.mlp.experts.141.up_proj", "model.layers.61.mlp.experts.142.up_proj", "model.layers.61.mlp.experts.143.up_proj", "model.layers.61.mlp.experts.144.up_proj", "model.layers.61.mlp.experts.145.up_proj", "model.layers.61.mlp.experts.146.up_proj", "model.layers.61.mlp.experts.147.up_proj", "model.layers.61.mlp.experts.148.up_proj", "model.layers.61.mlp.experts.149.up_proj", "model.layers.61.mlp.experts.150.up_proj", "model.layers.61.mlp.experts.151.up_proj", "model.layers.61.mlp.experts.152.up_proj", "model.layers.61.mlp.experts.153.up_proj", "model.layers.61.mlp.experts.154.up_proj", "model.layers.61.mlp.experts.155.up_proj", "model.layers.61.mlp.experts.156.up_proj", "model.layers.61.mlp.experts.157.up_proj", "model.layers.61.mlp.experts.158.up_proj", "model.layers.61.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -2.1469220519071408e-05, "dbits": 2516582400 }, { "dkld": -8.86590220034178e-05, "dbits": 5033164800 }, { "dkld": -9.533576667308807e-05, "dbits": 7549747200 }, { "dkld": -8.930498734116554e-05, "dbits": 12582912000 } ] }, { "idx": 368, "layers": [ "model.layers.61.mlp.experts.0.down_proj", "model.layers.61.mlp.experts.1.down_proj", "model.layers.61.mlp.experts.2.down_proj", "model.layers.61.mlp.experts.3.down_proj", "model.layers.61.mlp.experts.4.down_proj", "model.layers.61.mlp.experts.5.down_proj", "model.layers.61.mlp.experts.6.down_proj", "model.layers.61.mlp.experts.7.down_proj", "model.layers.61.mlp.experts.8.down_proj", "model.layers.61.mlp.experts.9.down_proj", "model.layers.61.mlp.experts.10.down_proj", "model.layers.61.mlp.experts.11.down_proj", "model.layers.61.mlp.experts.12.down_proj", "model.layers.61.mlp.experts.13.down_proj", "model.layers.61.mlp.experts.14.down_proj", "model.layers.61.mlp.experts.15.down_proj", "model.layers.61.mlp.experts.16.down_proj", "model.layers.61.mlp.experts.17.down_proj", "model.layers.61.mlp.experts.18.down_proj", "model.layers.61.mlp.experts.19.down_proj", "model.layers.61.mlp.experts.20.down_proj", "model.layers.61.mlp.experts.21.down_proj", "model.layers.61.mlp.experts.22.down_proj", "model.layers.61.mlp.experts.23.down_proj", "model.layers.61.mlp.experts.24.down_proj", "model.layers.61.mlp.experts.25.down_proj", "model.layers.61.mlp.experts.26.down_proj", "model.layers.61.mlp.experts.27.down_proj", "model.layers.61.mlp.experts.28.down_proj", "model.layers.61.mlp.experts.29.down_proj", "model.layers.61.mlp.experts.30.down_proj", "model.layers.61.mlp.experts.31.down_proj", "model.layers.61.mlp.experts.32.down_proj", "model.layers.61.mlp.experts.33.down_proj", "model.layers.61.mlp.experts.34.down_proj", "model.layers.61.mlp.experts.35.down_proj", "model.layers.61.mlp.experts.36.down_proj", "model.layers.61.mlp.experts.37.down_proj", "model.layers.61.mlp.experts.38.down_proj", "model.layers.61.mlp.experts.39.down_proj", "model.layers.61.mlp.experts.40.down_proj", "model.layers.61.mlp.experts.41.down_proj", "model.layers.61.mlp.experts.42.down_proj", "model.layers.61.mlp.experts.43.down_proj", "model.layers.61.mlp.experts.44.down_proj", "model.layers.61.mlp.experts.45.down_proj", "model.layers.61.mlp.experts.46.down_proj", "model.layers.61.mlp.experts.47.down_proj", "model.layers.61.mlp.experts.48.down_proj", "model.layers.61.mlp.experts.49.down_proj", "model.layers.61.mlp.experts.50.down_proj", "model.layers.61.mlp.experts.51.down_proj", "model.layers.61.mlp.experts.52.down_proj", "model.layers.61.mlp.experts.53.down_proj", "model.layers.61.mlp.experts.54.down_proj", "model.layers.61.mlp.experts.55.down_proj", "model.layers.61.mlp.experts.56.down_proj", "model.layers.61.mlp.experts.57.down_proj", "model.layers.61.mlp.experts.58.down_proj", "model.layers.61.mlp.experts.59.down_proj", "model.layers.61.mlp.experts.60.down_proj", "model.layers.61.mlp.experts.61.down_proj", "model.layers.61.mlp.experts.62.down_proj", "model.layers.61.mlp.experts.63.down_proj", "model.layers.61.mlp.experts.64.down_proj", "model.layers.61.mlp.experts.65.down_proj", "model.layers.61.mlp.experts.66.down_proj", "model.layers.61.mlp.experts.67.down_proj", "model.layers.61.mlp.experts.68.down_proj", "model.layers.61.mlp.experts.69.down_proj", "model.layers.61.mlp.experts.70.down_proj", "model.layers.61.mlp.experts.71.down_proj", "model.layers.61.mlp.experts.72.down_proj", "model.layers.61.mlp.experts.73.down_proj", "model.layers.61.mlp.experts.74.down_proj", "model.layers.61.mlp.experts.75.down_proj", "model.layers.61.mlp.experts.76.down_proj", "model.layers.61.mlp.experts.77.down_proj", "model.layers.61.mlp.experts.78.down_proj", "model.layers.61.mlp.experts.79.down_proj", "model.layers.61.mlp.experts.80.down_proj", "model.layers.61.mlp.experts.81.down_proj", "model.layers.61.mlp.experts.82.down_proj", "model.layers.61.mlp.experts.83.down_proj", "model.layers.61.mlp.experts.84.down_proj", "model.layers.61.mlp.experts.85.down_proj", "model.layers.61.mlp.experts.86.down_proj", "model.layers.61.mlp.experts.87.down_proj", "model.layers.61.mlp.experts.88.down_proj", "model.layers.61.mlp.experts.89.down_proj", "model.layers.61.mlp.experts.90.down_proj", "model.layers.61.mlp.experts.91.down_proj", "model.layers.61.mlp.experts.92.down_proj", "model.layers.61.mlp.experts.93.down_proj", "model.layers.61.mlp.experts.94.down_proj", "model.layers.61.mlp.experts.95.down_proj", "model.layers.61.mlp.experts.96.down_proj", "model.layers.61.mlp.experts.97.down_proj", "model.layers.61.mlp.experts.98.down_proj", "model.layers.61.mlp.experts.99.down_proj", "model.layers.61.mlp.experts.100.down_proj", "model.layers.61.mlp.experts.101.down_proj", "model.layers.61.mlp.experts.102.down_proj", "model.layers.61.mlp.experts.103.down_proj", "model.layers.61.mlp.experts.104.down_proj", "model.layers.61.mlp.experts.105.down_proj", "model.layers.61.mlp.experts.106.down_proj", "model.layers.61.mlp.experts.107.down_proj", "model.layers.61.mlp.experts.108.down_proj", "model.layers.61.mlp.experts.109.down_proj", "model.layers.61.mlp.experts.110.down_proj", "model.layers.61.mlp.experts.111.down_proj", "model.layers.61.mlp.experts.112.down_proj", "model.layers.61.mlp.experts.113.down_proj", "model.layers.61.mlp.experts.114.down_proj", "model.layers.61.mlp.experts.115.down_proj", "model.layers.61.mlp.experts.116.down_proj", "model.layers.61.mlp.experts.117.down_proj", "model.layers.61.mlp.experts.118.down_proj", "model.layers.61.mlp.experts.119.down_proj", "model.layers.61.mlp.experts.120.down_proj", "model.layers.61.mlp.experts.121.down_proj", "model.layers.61.mlp.experts.122.down_proj", "model.layers.61.mlp.experts.123.down_proj", "model.layers.61.mlp.experts.124.down_proj", "model.layers.61.mlp.experts.125.down_proj", "model.layers.61.mlp.experts.126.down_proj", "model.layers.61.mlp.experts.127.down_proj", "model.layers.61.mlp.experts.128.down_proj", "model.layers.61.mlp.experts.129.down_proj", "model.layers.61.mlp.experts.130.down_proj", "model.layers.61.mlp.experts.131.down_proj", "model.layers.61.mlp.experts.132.down_proj", "model.layers.61.mlp.experts.133.down_proj", "model.layers.61.mlp.experts.134.down_proj", "model.layers.61.mlp.experts.135.down_proj", "model.layers.61.mlp.experts.136.down_proj", "model.layers.61.mlp.experts.137.down_proj", "model.layers.61.mlp.experts.138.down_proj", "model.layers.61.mlp.experts.139.down_proj", "model.layers.61.mlp.experts.140.down_proj", "model.layers.61.mlp.experts.141.down_proj", "model.layers.61.mlp.experts.142.down_proj", "model.layers.61.mlp.experts.143.down_proj", "model.layers.61.mlp.experts.144.down_proj", "model.layers.61.mlp.experts.145.down_proj", "model.layers.61.mlp.experts.146.down_proj", "model.layers.61.mlp.experts.147.down_proj", "model.layers.61.mlp.experts.148.down_proj", "model.layers.61.mlp.experts.149.down_proj", "model.layers.61.mlp.experts.150.down_proj", "model.layers.61.mlp.experts.151.down_proj", "model.layers.61.mlp.experts.152.down_proj", "model.layers.61.mlp.experts.153.down_proj", "model.layers.61.mlp.experts.154.down_proj", "model.layers.61.mlp.experts.155.down_proj", "model.layers.61.mlp.experts.156.down_proj", "model.layers.61.mlp.experts.157.down_proj", "model.layers.61.mlp.experts.158.down_proj", "model.layers.61.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -7.79248774051694e-05, "dbits": 1258291200 }, { "dkld": -3.593638539314825e-05, "dbits": 2516582400 }, { "dkld": -4.572328180074692e-05, "dbits": 3774873600 }, { "dkld": -7.448438555002213e-05, "dbits": 6291456000 } ] }, { "idx": 369, "layers": [ "model.layers.62.self_attn.q_proj" ], "candidates": [ { "dkld": -7.552858442068378e-05, "dbits": 62914560 }, { "dkld": -5.162600427877084e-06, "dbits": 125829120 }, { "dkld": -2.552829682828106e-05, "dbits": 188743680 }, { "dkld": 6.4073130488367935e-06, "dbits": 314572800 } ] }, { "idx": 370, "layers": [ "model.layers.62.self_attn.k_proj", "model.layers.62.self_attn.v_proj" ], "candidates": [ { "dkld": -8.361637592316229e-05, "dbits": 10485760 }, { "dkld": -0.00013658162206411917, "dbits": 20971520 }, { "dkld": -6.328411400319256e-05, "dbits": 31457280 }, { "dkld": -0.00013106260448694784, "dbits": 52428800 } ] }, { "idx": 371, "layers": [ "model.layers.62.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00011549042537808973, "dbits": 62914560 }, { "dkld": -8.588992059231637e-05, "dbits": 125829120 }, { "dkld": -7.378701120615283e-05, "dbits": 188743680 }, { "dkld": -7.3027424514302e-05, "dbits": 314572800 } ] }, { "idx": 372, "layers": [ "model.layers.62.mlp.shared_experts.gate_proj", "model.layers.62.mlp.shared_experts.up_proj", "model.layers.62.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -5.8302842080595885e-05, "dbits": 23592960 }, { "dkld": 4.5165792107573766e-05, "dbits": 47185920 }, { "dkld": -2.286704257131733e-05, "dbits": 70778880 }, { "dkld": -7.167030125856955e-05, "dbits": 117964800 } ] }, { "idx": 373, "layers": [ "model.layers.62.mlp.experts.0.gate_proj", "model.layers.62.mlp.experts.1.gate_proj", "model.layers.62.mlp.experts.2.gate_proj", "model.layers.62.mlp.experts.3.gate_proj", "model.layers.62.mlp.experts.4.gate_proj", "model.layers.62.mlp.experts.5.gate_proj", "model.layers.62.mlp.experts.6.gate_proj", "model.layers.62.mlp.experts.7.gate_proj", "model.layers.62.mlp.experts.8.gate_proj", "model.layers.62.mlp.experts.9.gate_proj", "model.layers.62.mlp.experts.10.gate_proj", "model.layers.62.mlp.experts.11.gate_proj", "model.layers.62.mlp.experts.12.gate_proj", "model.layers.62.mlp.experts.13.gate_proj", "model.layers.62.mlp.experts.14.gate_proj", "model.layers.62.mlp.experts.15.gate_proj", "model.layers.62.mlp.experts.16.gate_proj", "model.layers.62.mlp.experts.17.gate_proj", "model.layers.62.mlp.experts.18.gate_proj", "model.layers.62.mlp.experts.19.gate_proj", "model.layers.62.mlp.experts.20.gate_proj", "model.layers.62.mlp.experts.21.gate_proj", "model.layers.62.mlp.experts.22.gate_proj", "model.layers.62.mlp.experts.23.gate_proj", "model.layers.62.mlp.experts.24.gate_proj", "model.layers.62.mlp.experts.25.gate_proj", "model.layers.62.mlp.experts.26.gate_proj", "model.layers.62.mlp.experts.27.gate_proj", "model.layers.62.mlp.experts.28.gate_proj", "model.layers.62.mlp.experts.29.gate_proj", "model.layers.62.mlp.experts.30.gate_proj", "model.layers.62.mlp.experts.31.gate_proj", "model.layers.62.mlp.experts.32.gate_proj", "model.layers.62.mlp.experts.33.gate_proj", "model.layers.62.mlp.experts.34.gate_proj", "model.layers.62.mlp.experts.35.gate_proj", "model.layers.62.mlp.experts.36.gate_proj", "model.layers.62.mlp.experts.37.gate_proj", "model.layers.62.mlp.experts.38.gate_proj", "model.layers.62.mlp.experts.39.gate_proj", "model.layers.62.mlp.experts.40.gate_proj", "model.layers.62.mlp.experts.41.gate_proj", "model.layers.62.mlp.experts.42.gate_proj", "model.layers.62.mlp.experts.43.gate_proj", "model.layers.62.mlp.experts.44.gate_proj", "model.layers.62.mlp.experts.45.gate_proj", "model.layers.62.mlp.experts.46.gate_proj", "model.layers.62.mlp.experts.47.gate_proj", "model.layers.62.mlp.experts.48.gate_proj", "model.layers.62.mlp.experts.49.gate_proj", "model.layers.62.mlp.experts.50.gate_proj", "model.layers.62.mlp.experts.51.gate_proj", "model.layers.62.mlp.experts.52.gate_proj", "model.layers.62.mlp.experts.53.gate_proj", "model.layers.62.mlp.experts.54.gate_proj", "model.layers.62.mlp.experts.55.gate_proj", "model.layers.62.mlp.experts.56.gate_proj", "model.layers.62.mlp.experts.57.gate_proj", "model.layers.62.mlp.experts.58.gate_proj", "model.layers.62.mlp.experts.59.gate_proj", "model.layers.62.mlp.experts.60.gate_proj", "model.layers.62.mlp.experts.61.gate_proj", "model.layers.62.mlp.experts.62.gate_proj", "model.layers.62.mlp.experts.63.gate_proj", "model.layers.62.mlp.experts.64.gate_proj", "model.layers.62.mlp.experts.65.gate_proj", "model.layers.62.mlp.experts.66.gate_proj", "model.layers.62.mlp.experts.67.gate_proj", "model.layers.62.mlp.experts.68.gate_proj", "model.layers.62.mlp.experts.69.gate_proj", "model.layers.62.mlp.experts.70.gate_proj", "model.layers.62.mlp.experts.71.gate_proj", "model.layers.62.mlp.experts.72.gate_proj", "model.layers.62.mlp.experts.73.gate_proj", "model.layers.62.mlp.experts.74.gate_proj", "model.layers.62.mlp.experts.75.gate_proj", "model.layers.62.mlp.experts.76.gate_proj", "model.layers.62.mlp.experts.77.gate_proj", "model.layers.62.mlp.experts.78.gate_proj", "model.layers.62.mlp.experts.79.gate_proj", "model.layers.62.mlp.experts.80.gate_proj", "model.layers.62.mlp.experts.81.gate_proj", "model.layers.62.mlp.experts.82.gate_proj", "model.layers.62.mlp.experts.83.gate_proj", "model.layers.62.mlp.experts.84.gate_proj", "model.layers.62.mlp.experts.85.gate_proj", "model.layers.62.mlp.experts.86.gate_proj", "model.layers.62.mlp.experts.87.gate_proj", "model.layers.62.mlp.experts.88.gate_proj", "model.layers.62.mlp.experts.89.gate_proj", "model.layers.62.mlp.experts.90.gate_proj", "model.layers.62.mlp.experts.91.gate_proj", "model.layers.62.mlp.experts.92.gate_proj", "model.layers.62.mlp.experts.93.gate_proj", "model.layers.62.mlp.experts.94.gate_proj", "model.layers.62.mlp.experts.95.gate_proj", "model.layers.62.mlp.experts.96.gate_proj", "model.layers.62.mlp.experts.97.gate_proj", "model.layers.62.mlp.experts.98.gate_proj", "model.layers.62.mlp.experts.99.gate_proj", "model.layers.62.mlp.experts.100.gate_proj", "model.layers.62.mlp.experts.101.gate_proj", "model.layers.62.mlp.experts.102.gate_proj", "model.layers.62.mlp.experts.103.gate_proj", "model.layers.62.mlp.experts.104.gate_proj", "model.layers.62.mlp.experts.105.gate_proj", "model.layers.62.mlp.experts.106.gate_proj", "model.layers.62.mlp.experts.107.gate_proj", "model.layers.62.mlp.experts.108.gate_proj", "model.layers.62.mlp.experts.109.gate_proj", "model.layers.62.mlp.experts.110.gate_proj", "model.layers.62.mlp.experts.111.gate_proj", "model.layers.62.mlp.experts.112.gate_proj", "model.layers.62.mlp.experts.113.gate_proj", "model.layers.62.mlp.experts.114.gate_proj", "model.layers.62.mlp.experts.115.gate_proj", "model.layers.62.mlp.experts.116.gate_proj", "model.layers.62.mlp.experts.117.gate_proj", "model.layers.62.mlp.experts.118.gate_proj", "model.layers.62.mlp.experts.119.gate_proj", "model.layers.62.mlp.experts.120.gate_proj", "model.layers.62.mlp.experts.121.gate_proj", "model.layers.62.mlp.experts.122.gate_proj", "model.layers.62.mlp.experts.123.gate_proj", "model.layers.62.mlp.experts.124.gate_proj", "model.layers.62.mlp.experts.125.gate_proj", "model.layers.62.mlp.experts.126.gate_proj", "model.layers.62.mlp.experts.127.gate_proj", "model.layers.62.mlp.experts.128.gate_proj", "model.layers.62.mlp.experts.129.gate_proj", "model.layers.62.mlp.experts.130.gate_proj", "model.layers.62.mlp.experts.131.gate_proj", "model.layers.62.mlp.experts.132.gate_proj", "model.layers.62.mlp.experts.133.gate_proj", "model.layers.62.mlp.experts.134.gate_proj", "model.layers.62.mlp.experts.135.gate_proj", "model.layers.62.mlp.experts.136.gate_proj", "model.layers.62.mlp.experts.137.gate_proj", "model.layers.62.mlp.experts.138.gate_proj", "model.layers.62.mlp.experts.139.gate_proj", "model.layers.62.mlp.experts.140.gate_proj", "model.layers.62.mlp.experts.141.gate_proj", "model.layers.62.mlp.experts.142.gate_proj", "model.layers.62.mlp.experts.143.gate_proj", "model.layers.62.mlp.experts.144.gate_proj", "model.layers.62.mlp.experts.145.gate_proj", "model.layers.62.mlp.experts.146.gate_proj", "model.layers.62.mlp.experts.147.gate_proj", "model.layers.62.mlp.experts.148.gate_proj", "model.layers.62.mlp.experts.149.gate_proj", "model.layers.62.mlp.experts.150.gate_proj", "model.layers.62.mlp.experts.151.gate_proj", "model.layers.62.mlp.experts.152.gate_proj", "model.layers.62.mlp.experts.153.gate_proj", "model.layers.62.mlp.experts.154.gate_proj", "model.layers.62.mlp.experts.155.gate_proj", "model.layers.62.mlp.experts.156.gate_proj", "model.layers.62.mlp.experts.157.gate_proj", "model.layers.62.mlp.experts.158.gate_proj", "model.layers.62.mlp.experts.159.gate_proj", "model.layers.62.mlp.experts.0.up_proj", "model.layers.62.mlp.experts.1.up_proj", "model.layers.62.mlp.experts.2.up_proj", "model.layers.62.mlp.experts.3.up_proj", "model.layers.62.mlp.experts.4.up_proj", "model.layers.62.mlp.experts.5.up_proj", "model.layers.62.mlp.experts.6.up_proj", "model.layers.62.mlp.experts.7.up_proj", "model.layers.62.mlp.experts.8.up_proj", "model.layers.62.mlp.experts.9.up_proj", "model.layers.62.mlp.experts.10.up_proj", "model.layers.62.mlp.experts.11.up_proj", "model.layers.62.mlp.experts.12.up_proj", "model.layers.62.mlp.experts.13.up_proj", "model.layers.62.mlp.experts.14.up_proj", "model.layers.62.mlp.experts.15.up_proj", "model.layers.62.mlp.experts.16.up_proj", "model.layers.62.mlp.experts.17.up_proj", "model.layers.62.mlp.experts.18.up_proj", "model.layers.62.mlp.experts.19.up_proj", "model.layers.62.mlp.experts.20.up_proj", "model.layers.62.mlp.experts.21.up_proj", "model.layers.62.mlp.experts.22.up_proj", "model.layers.62.mlp.experts.23.up_proj", "model.layers.62.mlp.experts.24.up_proj", "model.layers.62.mlp.experts.25.up_proj", "model.layers.62.mlp.experts.26.up_proj", "model.layers.62.mlp.experts.27.up_proj", "model.layers.62.mlp.experts.28.up_proj", "model.layers.62.mlp.experts.29.up_proj", "model.layers.62.mlp.experts.30.up_proj", "model.layers.62.mlp.experts.31.up_proj", "model.layers.62.mlp.experts.32.up_proj", "model.layers.62.mlp.experts.33.up_proj", "model.layers.62.mlp.experts.34.up_proj", "model.layers.62.mlp.experts.35.up_proj", "model.layers.62.mlp.experts.36.up_proj", "model.layers.62.mlp.experts.37.up_proj", "model.layers.62.mlp.experts.38.up_proj", "model.layers.62.mlp.experts.39.up_proj", "model.layers.62.mlp.experts.40.up_proj", "model.layers.62.mlp.experts.41.up_proj", "model.layers.62.mlp.experts.42.up_proj", "model.layers.62.mlp.experts.43.up_proj", "model.layers.62.mlp.experts.44.up_proj", "model.layers.62.mlp.experts.45.up_proj", "model.layers.62.mlp.experts.46.up_proj", "model.layers.62.mlp.experts.47.up_proj", "model.layers.62.mlp.experts.48.up_proj", "model.layers.62.mlp.experts.49.up_proj", "model.layers.62.mlp.experts.50.up_proj", "model.layers.62.mlp.experts.51.up_proj", "model.layers.62.mlp.experts.52.up_proj", "model.layers.62.mlp.experts.53.up_proj", "model.layers.62.mlp.experts.54.up_proj", "model.layers.62.mlp.experts.55.up_proj", "model.layers.62.mlp.experts.56.up_proj", "model.layers.62.mlp.experts.57.up_proj", "model.layers.62.mlp.experts.58.up_proj", "model.layers.62.mlp.experts.59.up_proj", "model.layers.62.mlp.experts.60.up_proj", "model.layers.62.mlp.experts.61.up_proj", "model.layers.62.mlp.experts.62.up_proj", "model.layers.62.mlp.experts.63.up_proj", "model.layers.62.mlp.experts.64.up_proj", "model.layers.62.mlp.experts.65.up_proj", "model.layers.62.mlp.experts.66.up_proj", "model.layers.62.mlp.experts.67.up_proj", "model.layers.62.mlp.experts.68.up_proj", "model.layers.62.mlp.experts.69.up_proj", "model.layers.62.mlp.experts.70.up_proj", "model.layers.62.mlp.experts.71.up_proj", "model.layers.62.mlp.experts.72.up_proj", "model.layers.62.mlp.experts.73.up_proj", "model.layers.62.mlp.experts.74.up_proj", "model.layers.62.mlp.experts.75.up_proj", "model.layers.62.mlp.experts.76.up_proj", "model.layers.62.mlp.experts.77.up_proj", "model.layers.62.mlp.experts.78.up_proj", "model.layers.62.mlp.experts.79.up_proj", "model.layers.62.mlp.experts.80.up_proj", "model.layers.62.mlp.experts.81.up_proj", "model.layers.62.mlp.experts.82.up_proj", "model.layers.62.mlp.experts.83.up_proj", "model.layers.62.mlp.experts.84.up_proj", "model.layers.62.mlp.experts.85.up_proj", "model.layers.62.mlp.experts.86.up_proj", "model.layers.62.mlp.experts.87.up_proj", "model.layers.62.mlp.experts.88.up_proj", "model.layers.62.mlp.experts.89.up_proj", "model.layers.62.mlp.experts.90.up_proj", "model.layers.62.mlp.experts.91.up_proj", "model.layers.62.mlp.experts.92.up_proj", "model.layers.62.mlp.experts.93.up_proj", "model.layers.62.mlp.experts.94.up_proj", "model.layers.62.mlp.experts.95.up_proj", "model.layers.62.mlp.experts.96.up_proj", "model.layers.62.mlp.experts.97.up_proj", "model.layers.62.mlp.experts.98.up_proj", "model.layers.62.mlp.experts.99.up_proj", "model.layers.62.mlp.experts.100.up_proj", "model.layers.62.mlp.experts.101.up_proj", "model.layers.62.mlp.experts.102.up_proj", "model.layers.62.mlp.experts.103.up_proj", "model.layers.62.mlp.experts.104.up_proj", "model.layers.62.mlp.experts.105.up_proj", "model.layers.62.mlp.experts.106.up_proj", "model.layers.62.mlp.experts.107.up_proj", "model.layers.62.mlp.experts.108.up_proj", "model.layers.62.mlp.experts.109.up_proj", "model.layers.62.mlp.experts.110.up_proj", "model.layers.62.mlp.experts.111.up_proj", "model.layers.62.mlp.experts.112.up_proj", "model.layers.62.mlp.experts.113.up_proj", "model.layers.62.mlp.experts.114.up_proj", "model.layers.62.mlp.experts.115.up_proj", "model.layers.62.mlp.experts.116.up_proj", "model.layers.62.mlp.experts.117.up_proj", "model.layers.62.mlp.experts.118.up_proj", "model.layers.62.mlp.experts.119.up_proj", "model.layers.62.mlp.experts.120.up_proj", "model.layers.62.mlp.experts.121.up_proj", "model.layers.62.mlp.experts.122.up_proj", "model.layers.62.mlp.experts.123.up_proj", "model.layers.62.mlp.experts.124.up_proj", "model.layers.62.mlp.experts.125.up_proj", "model.layers.62.mlp.experts.126.up_proj", "model.layers.62.mlp.experts.127.up_proj", "model.layers.62.mlp.experts.128.up_proj", "model.layers.62.mlp.experts.129.up_proj", "model.layers.62.mlp.experts.130.up_proj", "model.layers.62.mlp.experts.131.up_proj", "model.layers.62.mlp.experts.132.up_proj", "model.layers.62.mlp.experts.133.up_proj", "model.layers.62.mlp.experts.134.up_proj", "model.layers.62.mlp.experts.135.up_proj", "model.layers.62.mlp.experts.136.up_proj", "model.layers.62.mlp.experts.137.up_proj", "model.layers.62.mlp.experts.138.up_proj", "model.layers.62.mlp.experts.139.up_proj", "model.layers.62.mlp.experts.140.up_proj", "model.layers.62.mlp.experts.141.up_proj", "model.layers.62.mlp.experts.142.up_proj", "model.layers.62.mlp.experts.143.up_proj", "model.layers.62.mlp.experts.144.up_proj", "model.layers.62.mlp.experts.145.up_proj", "model.layers.62.mlp.experts.146.up_proj", "model.layers.62.mlp.experts.147.up_proj", "model.layers.62.mlp.experts.148.up_proj", "model.layers.62.mlp.experts.149.up_proj", "model.layers.62.mlp.experts.150.up_proj", "model.layers.62.mlp.experts.151.up_proj", "model.layers.62.mlp.experts.152.up_proj", "model.layers.62.mlp.experts.153.up_proj", "model.layers.62.mlp.experts.154.up_proj", "model.layers.62.mlp.experts.155.up_proj", "model.layers.62.mlp.experts.156.up_proj", "model.layers.62.mlp.experts.157.up_proj", "model.layers.62.mlp.experts.158.up_proj", "model.layers.62.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -9.201560169458389e-05, "dbits": 2516582400 }, { "dkld": -0.00019100196659566082, "dbits": 5033164800 }, { "dkld": -0.00016432655975223698, "dbits": 7549747200 }, { "dkld": -0.0001846218481659917, "dbits": 12582912000 } ] }, { "idx": 374, "layers": [ "model.layers.62.mlp.experts.0.down_proj", "model.layers.62.mlp.experts.1.down_proj", "model.layers.62.mlp.experts.2.down_proj", "model.layers.62.mlp.experts.3.down_proj", "model.layers.62.mlp.experts.4.down_proj", "model.layers.62.mlp.experts.5.down_proj", "model.layers.62.mlp.experts.6.down_proj", "model.layers.62.mlp.experts.7.down_proj", "model.layers.62.mlp.experts.8.down_proj", "model.layers.62.mlp.experts.9.down_proj", "model.layers.62.mlp.experts.10.down_proj", "model.layers.62.mlp.experts.11.down_proj", "model.layers.62.mlp.experts.12.down_proj", "model.layers.62.mlp.experts.13.down_proj", "model.layers.62.mlp.experts.14.down_proj", "model.layers.62.mlp.experts.15.down_proj", "model.layers.62.mlp.experts.16.down_proj", "model.layers.62.mlp.experts.17.down_proj", "model.layers.62.mlp.experts.18.down_proj", "model.layers.62.mlp.experts.19.down_proj", "model.layers.62.mlp.experts.20.down_proj", "model.layers.62.mlp.experts.21.down_proj", "model.layers.62.mlp.experts.22.down_proj", "model.layers.62.mlp.experts.23.down_proj", "model.layers.62.mlp.experts.24.down_proj", "model.layers.62.mlp.experts.25.down_proj", "model.layers.62.mlp.experts.26.down_proj", "model.layers.62.mlp.experts.27.down_proj", "model.layers.62.mlp.experts.28.down_proj", "model.layers.62.mlp.experts.29.down_proj", "model.layers.62.mlp.experts.30.down_proj", "model.layers.62.mlp.experts.31.down_proj", "model.layers.62.mlp.experts.32.down_proj", "model.layers.62.mlp.experts.33.down_proj", "model.layers.62.mlp.experts.34.down_proj", "model.layers.62.mlp.experts.35.down_proj", "model.layers.62.mlp.experts.36.down_proj", "model.layers.62.mlp.experts.37.down_proj", "model.layers.62.mlp.experts.38.down_proj", "model.layers.62.mlp.experts.39.down_proj", "model.layers.62.mlp.experts.40.down_proj", "model.layers.62.mlp.experts.41.down_proj", "model.layers.62.mlp.experts.42.down_proj", "model.layers.62.mlp.experts.43.down_proj", "model.layers.62.mlp.experts.44.down_proj", "model.layers.62.mlp.experts.45.down_proj", "model.layers.62.mlp.experts.46.down_proj", "model.layers.62.mlp.experts.47.down_proj", "model.layers.62.mlp.experts.48.down_proj", "model.layers.62.mlp.experts.49.down_proj", "model.layers.62.mlp.experts.50.down_proj", "model.layers.62.mlp.experts.51.down_proj", "model.layers.62.mlp.experts.52.down_proj", "model.layers.62.mlp.experts.53.down_proj", "model.layers.62.mlp.experts.54.down_proj", "model.layers.62.mlp.experts.55.down_proj", "model.layers.62.mlp.experts.56.down_proj", "model.layers.62.mlp.experts.57.down_proj", "model.layers.62.mlp.experts.58.down_proj", "model.layers.62.mlp.experts.59.down_proj", "model.layers.62.mlp.experts.60.down_proj", "model.layers.62.mlp.experts.61.down_proj", "model.layers.62.mlp.experts.62.down_proj", "model.layers.62.mlp.experts.63.down_proj", "model.layers.62.mlp.experts.64.down_proj", "model.layers.62.mlp.experts.65.down_proj", "model.layers.62.mlp.experts.66.down_proj", "model.layers.62.mlp.experts.67.down_proj", "model.layers.62.mlp.experts.68.down_proj", "model.layers.62.mlp.experts.69.down_proj", "model.layers.62.mlp.experts.70.down_proj", "model.layers.62.mlp.experts.71.down_proj", "model.layers.62.mlp.experts.72.down_proj", "model.layers.62.mlp.experts.73.down_proj", "model.layers.62.mlp.experts.74.down_proj", "model.layers.62.mlp.experts.75.down_proj", "model.layers.62.mlp.experts.76.down_proj", "model.layers.62.mlp.experts.77.down_proj", "model.layers.62.mlp.experts.78.down_proj", "model.layers.62.mlp.experts.79.down_proj", "model.layers.62.mlp.experts.80.down_proj", "model.layers.62.mlp.experts.81.down_proj", "model.layers.62.mlp.experts.82.down_proj", "model.layers.62.mlp.experts.83.down_proj", "model.layers.62.mlp.experts.84.down_proj", "model.layers.62.mlp.experts.85.down_proj", "model.layers.62.mlp.experts.86.down_proj", "model.layers.62.mlp.experts.87.down_proj", "model.layers.62.mlp.experts.88.down_proj", "model.layers.62.mlp.experts.89.down_proj", "model.layers.62.mlp.experts.90.down_proj", "model.layers.62.mlp.experts.91.down_proj", "model.layers.62.mlp.experts.92.down_proj", "model.layers.62.mlp.experts.93.down_proj", "model.layers.62.mlp.experts.94.down_proj", "model.layers.62.mlp.experts.95.down_proj", "model.layers.62.mlp.experts.96.down_proj", "model.layers.62.mlp.experts.97.down_proj", "model.layers.62.mlp.experts.98.down_proj", "model.layers.62.mlp.experts.99.down_proj", "model.layers.62.mlp.experts.100.down_proj", "model.layers.62.mlp.experts.101.down_proj", "model.layers.62.mlp.experts.102.down_proj", "model.layers.62.mlp.experts.103.down_proj", "model.layers.62.mlp.experts.104.down_proj", "model.layers.62.mlp.experts.105.down_proj", "model.layers.62.mlp.experts.106.down_proj", "model.layers.62.mlp.experts.107.down_proj", "model.layers.62.mlp.experts.108.down_proj", "model.layers.62.mlp.experts.109.down_proj", "model.layers.62.mlp.experts.110.down_proj", "model.layers.62.mlp.experts.111.down_proj", "model.layers.62.mlp.experts.112.down_proj", "model.layers.62.mlp.experts.113.down_proj", "model.layers.62.mlp.experts.114.down_proj", "model.layers.62.mlp.experts.115.down_proj", "model.layers.62.mlp.experts.116.down_proj", "model.layers.62.mlp.experts.117.down_proj", "model.layers.62.mlp.experts.118.down_proj", "model.layers.62.mlp.experts.119.down_proj", "model.layers.62.mlp.experts.120.down_proj", "model.layers.62.mlp.experts.121.down_proj", "model.layers.62.mlp.experts.122.down_proj", "model.layers.62.mlp.experts.123.down_proj", "model.layers.62.mlp.experts.124.down_proj", "model.layers.62.mlp.experts.125.down_proj", "model.layers.62.mlp.experts.126.down_proj", "model.layers.62.mlp.experts.127.down_proj", "model.layers.62.mlp.experts.128.down_proj", "model.layers.62.mlp.experts.129.down_proj", "model.layers.62.mlp.experts.130.down_proj", "model.layers.62.mlp.experts.131.down_proj", "model.layers.62.mlp.experts.132.down_proj", "model.layers.62.mlp.experts.133.down_proj", "model.layers.62.mlp.experts.134.down_proj", "model.layers.62.mlp.experts.135.down_proj", "model.layers.62.mlp.experts.136.down_proj", "model.layers.62.mlp.experts.137.down_proj", "model.layers.62.mlp.experts.138.down_proj", "model.layers.62.mlp.experts.139.down_proj", "model.layers.62.mlp.experts.140.down_proj", "model.layers.62.mlp.experts.141.down_proj", "model.layers.62.mlp.experts.142.down_proj", "model.layers.62.mlp.experts.143.down_proj", "model.layers.62.mlp.experts.144.down_proj", "model.layers.62.mlp.experts.145.down_proj", "model.layers.62.mlp.experts.146.down_proj", "model.layers.62.mlp.experts.147.down_proj", "model.layers.62.mlp.experts.148.down_proj", "model.layers.62.mlp.experts.149.down_proj", "model.layers.62.mlp.experts.150.down_proj", "model.layers.62.mlp.experts.151.down_proj", "model.layers.62.mlp.experts.152.down_proj", "model.layers.62.mlp.experts.153.down_proj", "model.layers.62.mlp.experts.154.down_proj", "model.layers.62.mlp.experts.155.down_proj", "model.layers.62.mlp.experts.156.down_proj", "model.layers.62.mlp.experts.157.down_proj", "model.layers.62.mlp.experts.158.down_proj", "model.layers.62.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00017984863370656967, "dbits": 1258291200 }, { "dkld": -0.00019002864137292186, "dbits": 2516582400 }, { "dkld": -0.00014948882162571508, "dbits": 3774873600 }, { "dkld": -0.00013860287144780437, "dbits": 6291456000 } ] }, { "idx": 375, "layers": [ "model.layers.63.self_attn.q_proj" ], "candidates": [ { "dkld": 8.912961930035435e-05, "dbits": 62914560 }, { "dkld": -7.415190339116196e-07, "dbits": 125829120 }, { "dkld": 3.2784417271503052e-06, "dbits": 188743680 }, { "dkld": -1.0797753930091858e-05, "dbits": 314572800 } ] }, { "idx": 376, "layers": [ "model.layers.63.self_attn.k_proj", "model.layers.63.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00012991195544601042, "dbits": 10485760 }, { "dkld": -0.0001039595343172578, "dbits": 20971520 }, { "dkld": -0.0001262594945728862, "dbits": 31457280 }, { "dkld": -0.00014087688177824575, "dbits": 52428800 } ] }, { "idx": 377, "layers": [ "model.layers.63.self_attn.o_proj" ], "candidates": [ { "dkld": -2.522617578507025e-05, "dbits": 62914560 }, { "dkld": -4.6039000153541565e-05, "dbits": 125829120 }, { "dkld": -6.223954260349551e-05, "dbits": 188743680 }, { "dkld": -2.4802237749102507e-05, "dbits": 314572800 } ] }, { "idx": 378, "layers": [ "model.layers.63.mlp.shared_experts.gate_proj", "model.layers.63.mlp.shared_experts.up_proj", "model.layers.63.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -3.3045653253793716e-05, "dbits": 23592960 }, { "dkld": -2.5132112205034085e-05, "dbits": 47185920 }, { "dkld": 3.664940595615729e-06, "dbits": 70778880 }, { "dkld": -7.613003253942319e-06, "dbits": 117964800 } ] }, { "idx": 379, "layers": [ "model.layers.63.mlp.experts.0.gate_proj", "model.layers.63.mlp.experts.1.gate_proj", "model.layers.63.mlp.experts.2.gate_proj", "model.layers.63.mlp.experts.3.gate_proj", "model.layers.63.mlp.experts.4.gate_proj", "model.layers.63.mlp.experts.5.gate_proj", "model.layers.63.mlp.experts.6.gate_proj", "model.layers.63.mlp.experts.7.gate_proj", "model.layers.63.mlp.experts.8.gate_proj", "model.layers.63.mlp.experts.9.gate_proj", "model.layers.63.mlp.experts.10.gate_proj", "model.layers.63.mlp.experts.11.gate_proj", "model.layers.63.mlp.experts.12.gate_proj", "model.layers.63.mlp.experts.13.gate_proj", "model.layers.63.mlp.experts.14.gate_proj", "model.layers.63.mlp.experts.15.gate_proj", "model.layers.63.mlp.experts.16.gate_proj", "model.layers.63.mlp.experts.17.gate_proj", "model.layers.63.mlp.experts.18.gate_proj", "model.layers.63.mlp.experts.19.gate_proj", "model.layers.63.mlp.experts.20.gate_proj", "model.layers.63.mlp.experts.21.gate_proj", "model.layers.63.mlp.experts.22.gate_proj", "model.layers.63.mlp.experts.23.gate_proj", "model.layers.63.mlp.experts.24.gate_proj", "model.layers.63.mlp.experts.25.gate_proj", "model.layers.63.mlp.experts.26.gate_proj", "model.layers.63.mlp.experts.27.gate_proj", "model.layers.63.mlp.experts.28.gate_proj", "model.layers.63.mlp.experts.29.gate_proj", "model.layers.63.mlp.experts.30.gate_proj", "model.layers.63.mlp.experts.31.gate_proj", "model.layers.63.mlp.experts.32.gate_proj", "model.layers.63.mlp.experts.33.gate_proj", "model.layers.63.mlp.experts.34.gate_proj", "model.layers.63.mlp.experts.35.gate_proj", "model.layers.63.mlp.experts.36.gate_proj", "model.layers.63.mlp.experts.37.gate_proj", "model.layers.63.mlp.experts.38.gate_proj", "model.layers.63.mlp.experts.39.gate_proj", "model.layers.63.mlp.experts.40.gate_proj", "model.layers.63.mlp.experts.41.gate_proj", "model.layers.63.mlp.experts.42.gate_proj", "model.layers.63.mlp.experts.43.gate_proj", "model.layers.63.mlp.experts.44.gate_proj", "model.layers.63.mlp.experts.45.gate_proj", "model.layers.63.mlp.experts.46.gate_proj", "model.layers.63.mlp.experts.47.gate_proj", "model.layers.63.mlp.experts.48.gate_proj", "model.layers.63.mlp.experts.49.gate_proj", "model.layers.63.mlp.experts.50.gate_proj", "model.layers.63.mlp.experts.51.gate_proj", "model.layers.63.mlp.experts.52.gate_proj", "model.layers.63.mlp.experts.53.gate_proj", "model.layers.63.mlp.experts.54.gate_proj", "model.layers.63.mlp.experts.55.gate_proj", "model.layers.63.mlp.experts.56.gate_proj", "model.layers.63.mlp.experts.57.gate_proj", "model.layers.63.mlp.experts.58.gate_proj", "model.layers.63.mlp.experts.59.gate_proj", "model.layers.63.mlp.experts.60.gate_proj", "model.layers.63.mlp.experts.61.gate_proj", "model.layers.63.mlp.experts.62.gate_proj", "model.layers.63.mlp.experts.63.gate_proj", "model.layers.63.mlp.experts.64.gate_proj", "model.layers.63.mlp.experts.65.gate_proj", "model.layers.63.mlp.experts.66.gate_proj", "model.layers.63.mlp.experts.67.gate_proj", "model.layers.63.mlp.experts.68.gate_proj", "model.layers.63.mlp.experts.69.gate_proj", "model.layers.63.mlp.experts.70.gate_proj", "model.layers.63.mlp.experts.71.gate_proj", "model.layers.63.mlp.experts.72.gate_proj", "model.layers.63.mlp.experts.73.gate_proj", "model.layers.63.mlp.experts.74.gate_proj", "model.layers.63.mlp.experts.75.gate_proj", "model.layers.63.mlp.experts.76.gate_proj", "model.layers.63.mlp.experts.77.gate_proj", "model.layers.63.mlp.experts.78.gate_proj", "model.layers.63.mlp.experts.79.gate_proj", "model.layers.63.mlp.experts.80.gate_proj", "model.layers.63.mlp.experts.81.gate_proj", "model.layers.63.mlp.experts.82.gate_proj", "model.layers.63.mlp.experts.83.gate_proj", "model.layers.63.mlp.experts.84.gate_proj", "model.layers.63.mlp.experts.85.gate_proj", "model.layers.63.mlp.experts.86.gate_proj", "model.layers.63.mlp.experts.87.gate_proj", "model.layers.63.mlp.experts.88.gate_proj", "model.layers.63.mlp.experts.89.gate_proj", "model.layers.63.mlp.experts.90.gate_proj", "model.layers.63.mlp.experts.91.gate_proj", "model.layers.63.mlp.experts.92.gate_proj", "model.layers.63.mlp.experts.93.gate_proj", "model.layers.63.mlp.experts.94.gate_proj", "model.layers.63.mlp.experts.95.gate_proj", "model.layers.63.mlp.experts.96.gate_proj", "model.layers.63.mlp.experts.97.gate_proj", "model.layers.63.mlp.experts.98.gate_proj", "model.layers.63.mlp.experts.99.gate_proj", "model.layers.63.mlp.experts.100.gate_proj", "model.layers.63.mlp.experts.101.gate_proj", "model.layers.63.mlp.experts.102.gate_proj", "model.layers.63.mlp.experts.103.gate_proj", "model.layers.63.mlp.experts.104.gate_proj", "model.layers.63.mlp.experts.105.gate_proj", "model.layers.63.mlp.experts.106.gate_proj", "model.layers.63.mlp.experts.107.gate_proj", "model.layers.63.mlp.experts.108.gate_proj", "model.layers.63.mlp.experts.109.gate_proj", "model.layers.63.mlp.experts.110.gate_proj", "model.layers.63.mlp.experts.111.gate_proj", "model.layers.63.mlp.experts.112.gate_proj", "model.layers.63.mlp.experts.113.gate_proj", "model.layers.63.mlp.experts.114.gate_proj", "model.layers.63.mlp.experts.115.gate_proj", "model.layers.63.mlp.experts.116.gate_proj", "model.layers.63.mlp.experts.117.gate_proj", "model.layers.63.mlp.experts.118.gate_proj", "model.layers.63.mlp.experts.119.gate_proj", "model.layers.63.mlp.experts.120.gate_proj", "model.layers.63.mlp.experts.121.gate_proj", "model.layers.63.mlp.experts.122.gate_proj", "model.layers.63.mlp.experts.123.gate_proj", "model.layers.63.mlp.experts.124.gate_proj", "model.layers.63.mlp.experts.125.gate_proj", "model.layers.63.mlp.experts.126.gate_proj", "model.layers.63.mlp.experts.127.gate_proj", "model.layers.63.mlp.experts.128.gate_proj", "model.layers.63.mlp.experts.129.gate_proj", "model.layers.63.mlp.experts.130.gate_proj", "model.layers.63.mlp.experts.131.gate_proj", "model.layers.63.mlp.experts.132.gate_proj", "model.layers.63.mlp.experts.133.gate_proj", "model.layers.63.mlp.experts.134.gate_proj", "model.layers.63.mlp.experts.135.gate_proj", "model.layers.63.mlp.experts.136.gate_proj", "model.layers.63.mlp.experts.137.gate_proj", "model.layers.63.mlp.experts.138.gate_proj", "model.layers.63.mlp.experts.139.gate_proj", "model.layers.63.mlp.experts.140.gate_proj", "model.layers.63.mlp.experts.141.gate_proj", "model.layers.63.mlp.experts.142.gate_proj", "model.layers.63.mlp.experts.143.gate_proj", "model.layers.63.mlp.experts.144.gate_proj", "model.layers.63.mlp.experts.145.gate_proj", "model.layers.63.mlp.experts.146.gate_proj", "model.layers.63.mlp.experts.147.gate_proj", "model.layers.63.mlp.experts.148.gate_proj", "model.layers.63.mlp.experts.149.gate_proj", "model.layers.63.mlp.experts.150.gate_proj", "model.layers.63.mlp.experts.151.gate_proj", "model.layers.63.mlp.experts.152.gate_proj", "model.layers.63.mlp.experts.153.gate_proj", "model.layers.63.mlp.experts.154.gate_proj", "model.layers.63.mlp.experts.155.gate_proj", "model.layers.63.mlp.experts.156.gate_proj", "model.layers.63.mlp.experts.157.gate_proj", "model.layers.63.mlp.experts.158.gate_proj", "model.layers.63.mlp.experts.159.gate_proj", "model.layers.63.mlp.experts.0.up_proj", "model.layers.63.mlp.experts.1.up_proj", "model.layers.63.mlp.experts.2.up_proj", "model.layers.63.mlp.experts.3.up_proj", "model.layers.63.mlp.experts.4.up_proj", "model.layers.63.mlp.experts.5.up_proj", "model.layers.63.mlp.experts.6.up_proj", "model.layers.63.mlp.experts.7.up_proj", "model.layers.63.mlp.experts.8.up_proj", "model.layers.63.mlp.experts.9.up_proj", "model.layers.63.mlp.experts.10.up_proj", "model.layers.63.mlp.experts.11.up_proj", "model.layers.63.mlp.experts.12.up_proj", "model.layers.63.mlp.experts.13.up_proj", "model.layers.63.mlp.experts.14.up_proj", "model.layers.63.mlp.experts.15.up_proj", "model.layers.63.mlp.experts.16.up_proj", "model.layers.63.mlp.experts.17.up_proj", "model.layers.63.mlp.experts.18.up_proj", "model.layers.63.mlp.experts.19.up_proj", "model.layers.63.mlp.experts.20.up_proj", "model.layers.63.mlp.experts.21.up_proj", "model.layers.63.mlp.experts.22.up_proj", "model.layers.63.mlp.experts.23.up_proj", "model.layers.63.mlp.experts.24.up_proj", "model.layers.63.mlp.experts.25.up_proj", "model.layers.63.mlp.experts.26.up_proj", "model.layers.63.mlp.experts.27.up_proj", "model.layers.63.mlp.experts.28.up_proj", "model.layers.63.mlp.experts.29.up_proj", "model.layers.63.mlp.experts.30.up_proj", "model.layers.63.mlp.experts.31.up_proj", "model.layers.63.mlp.experts.32.up_proj", "model.layers.63.mlp.experts.33.up_proj", "model.layers.63.mlp.experts.34.up_proj", "model.layers.63.mlp.experts.35.up_proj", "model.layers.63.mlp.experts.36.up_proj", "model.layers.63.mlp.experts.37.up_proj", "model.layers.63.mlp.experts.38.up_proj", "model.layers.63.mlp.experts.39.up_proj", "model.layers.63.mlp.experts.40.up_proj", "model.layers.63.mlp.experts.41.up_proj", "model.layers.63.mlp.experts.42.up_proj", "model.layers.63.mlp.experts.43.up_proj", "model.layers.63.mlp.experts.44.up_proj", "model.layers.63.mlp.experts.45.up_proj", "model.layers.63.mlp.experts.46.up_proj", "model.layers.63.mlp.experts.47.up_proj", "model.layers.63.mlp.experts.48.up_proj", "model.layers.63.mlp.experts.49.up_proj", "model.layers.63.mlp.experts.50.up_proj", "model.layers.63.mlp.experts.51.up_proj", "model.layers.63.mlp.experts.52.up_proj", "model.layers.63.mlp.experts.53.up_proj", "model.layers.63.mlp.experts.54.up_proj", "model.layers.63.mlp.experts.55.up_proj", "model.layers.63.mlp.experts.56.up_proj", "model.layers.63.mlp.experts.57.up_proj", "model.layers.63.mlp.experts.58.up_proj", "model.layers.63.mlp.experts.59.up_proj", "model.layers.63.mlp.experts.60.up_proj", "model.layers.63.mlp.experts.61.up_proj", "model.layers.63.mlp.experts.62.up_proj", "model.layers.63.mlp.experts.63.up_proj", "model.layers.63.mlp.experts.64.up_proj", "model.layers.63.mlp.experts.65.up_proj", "model.layers.63.mlp.experts.66.up_proj", "model.layers.63.mlp.experts.67.up_proj", "model.layers.63.mlp.experts.68.up_proj", "model.layers.63.mlp.experts.69.up_proj", "model.layers.63.mlp.experts.70.up_proj", "model.layers.63.mlp.experts.71.up_proj", "model.layers.63.mlp.experts.72.up_proj", "model.layers.63.mlp.experts.73.up_proj", "model.layers.63.mlp.experts.74.up_proj", "model.layers.63.mlp.experts.75.up_proj", "model.layers.63.mlp.experts.76.up_proj", "model.layers.63.mlp.experts.77.up_proj", "model.layers.63.mlp.experts.78.up_proj", "model.layers.63.mlp.experts.79.up_proj", "model.layers.63.mlp.experts.80.up_proj", "model.layers.63.mlp.experts.81.up_proj", "model.layers.63.mlp.experts.82.up_proj", "model.layers.63.mlp.experts.83.up_proj", "model.layers.63.mlp.experts.84.up_proj", "model.layers.63.mlp.experts.85.up_proj", "model.layers.63.mlp.experts.86.up_proj", "model.layers.63.mlp.experts.87.up_proj", "model.layers.63.mlp.experts.88.up_proj", "model.layers.63.mlp.experts.89.up_proj", "model.layers.63.mlp.experts.90.up_proj", "model.layers.63.mlp.experts.91.up_proj", "model.layers.63.mlp.experts.92.up_proj", "model.layers.63.mlp.experts.93.up_proj", "model.layers.63.mlp.experts.94.up_proj", "model.layers.63.mlp.experts.95.up_proj", "model.layers.63.mlp.experts.96.up_proj", "model.layers.63.mlp.experts.97.up_proj", "model.layers.63.mlp.experts.98.up_proj", "model.layers.63.mlp.experts.99.up_proj", "model.layers.63.mlp.experts.100.up_proj", "model.layers.63.mlp.experts.101.up_proj", "model.layers.63.mlp.experts.102.up_proj", "model.layers.63.mlp.experts.103.up_proj", "model.layers.63.mlp.experts.104.up_proj", "model.layers.63.mlp.experts.105.up_proj", "model.layers.63.mlp.experts.106.up_proj", "model.layers.63.mlp.experts.107.up_proj", "model.layers.63.mlp.experts.108.up_proj", "model.layers.63.mlp.experts.109.up_proj", "model.layers.63.mlp.experts.110.up_proj", "model.layers.63.mlp.experts.111.up_proj", "model.layers.63.mlp.experts.112.up_proj", "model.layers.63.mlp.experts.113.up_proj", "model.layers.63.mlp.experts.114.up_proj", "model.layers.63.mlp.experts.115.up_proj", "model.layers.63.mlp.experts.116.up_proj", "model.layers.63.mlp.experts.117.up_proj", "model.layers.63.mlp.experts.118.up_proj", "model.layers.63.mlp.experts.119.up_proj", "model.layers.63.mlp.experts.120.up_proj", "model.layers.63.mlp.experts.121.up_proj", "model.layers.63.mlp.experts.122.up_proj", "model.layers.63.mlp.experts.123.up_proj", "model.layers.63.mlp.experts.124.up_proj", "model.layers.63.mlp.experts.125.up_proj", "model.layers.63.mlp.experts.126.up_proj", "model.layers.63.mlp.experts.127.up_proj", "model.layers.63.mlp.experts.128.up_proj", "model.layers.63.mlp.experts.129.up_proj", "model.layers.63.mlp.experts.130.up_proj", "model.layers.63.mlp.experts.131.up_proj", "model.layers.63.mlp.experts.132.up_proj", "model.layers.63.mlp.experts.133.up_proj", "model.layers.63.mlp.experts.134.up_proj", "model.layers.63.mlp.experts.135.up_proj", "model.layers.63.mlp.experts.136.up_proj", "model.layers.63.mlp.experts.137.up_proj", "model.layers.63.mlp.experts.138.up_proj", "model.layers.63.mlp.experts.139.up_proj", "model.layers.63.mlp.experts.140.up_proj", "model.layers.63.mlp.experts.141.up_proj", "model.layers.63.mlp.experts.142.up_proj", "model.layers.63.mlp.experts.143.up_proj", "model.layers.63.mlp.experts.144.up_proj", "model.layers.63.mlp.experts.145.up_proj", "model.layers.63.mlp.experts.146.up_proj", "model.layers.63.mlp.experts.147.up_proj", "model.layers.63.mlp.experts.148.up_proj", "model.layers.63.mlp.experts.149.up_proj", "model.layers.63.mlp.experts.150.up_proj", "model.layers.63.mlp.experts.151.up_proj", "model.layers.63.mlp.experts.152.up_proj", "model.layers.63.mlp.experts.153.up_proj", "model.layers.63.mlp.experts.154.up_proj", "model.layers.63.mlp.experts.155.up_proj", "model.layers.63.mlp.experts.156.up_proj", "model.layers.63.mlp.experts.157.up_proj", "model.layers.63.mlp.experts.158.up_proj", "model.layers.63.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 3.8259848952282294e-05, "dbits": 2516582400 }, { "dkld": -3.30757349729649e-05, "dbits": 5033164800 }, { "dkld": -2.9346812516456433e-05, "dbits": 7549747200 }, { "dkld": -2.5556422770023346e-05, "dbits": 12582912000 } ] }, { "idx": 380, "layers": [ "model.layers.63.mlp.experts.0.down_proj", "model.layers.63.mlp.experts.1.down_proj", "model.layers.63.mlp.experts.2.down_proj", "model.layers.63.mlp.experts.3.down_proj", "model.layers.63.mlp.experts.4.down_proj", "model.layers.63.mlp.experts.5.down_proj", "model.layers.63.mlp.experts.6.down_proj", "model.layers.63.mlp.experts.7.down_proj", "model.layers.63.mlp.experts.8.down_proj", "model.layers.63.mlp.experts.9.down_proj", "model.layers.63.mlp.experts.10.down_proj", "model.layers.63.mlp.experts.11.down_proj", "model.layers.63.mlp.experts.12.down_proj", "model.layers.63.mlp.experts.13.down_proj", "model.layers.63.mlp.experts.14.down_proj", "model.layers.63.mlp.experts.15.down_proj", "model.layers.63.mlp.experts.16.down_proj", "model.layers.63.mlp.experts.17.down_proj", "model.layers.63.mlp.experts.18.down_proj", "model.layers.63.mlp.experts.19.down_proj", "model.layers.63.mlp.experts.20.down_proj", "model.layers.63.mlp.experts.21.down_proj", "model.layers.63.mlp.experts.22.down_proj", "model.layers.63.mlp.experts.23.down_proj", "model.layers.63.mlp.experts.24.down_proj", "model.layers.63.mlp.experts.25.down_proj", "model.layers.63.mlp.experts.26.down_proj", "model.layers.63.mlp.experts.27.down_proj", "model.layers.63.mlp.experts.28.down_proj", "model.layers.63.mlp.experts.29.down_proj", "model.layers.63.mlp.experts.30.down_proj", "model.layers.63.mlp.experts.31.down_proj", "model.layers.63.mlp.experts.32.down_proj", "model.layers.63.mlp.experts.33.down_proj", "model.layers.63.mlp.experts.34.down_proj", "model.layers.63.mlp.experts.35.down_proj", "model.layers.63.mlp.experts.36.down_proj", "model.layers.63.mlp.experts.37.down_proj", "model.layers.63.mlp.experts.38.down_proj", "model.layers.63.mlp.experts.39.down_proj", "model.layers.63.mlp.experts.40.down_proj", "model.layers.63.mlp.experts.41.down_proj", "model.layers.63.mlp.experts.42.down_proj", "model.layers.63.mlp.experts.43.down_proj", "model.layers.63.mlp.experts.44.down_proj", "model.layers.63.mlp.experts.45.down_proj", "model.layers.63.mlp.experts.46.down_proj", "model.layers.63.mlp.experts.47.down_proj", "model.layers.63.mlp.experts.48.down_proj", "model.layers.63.mlp.experts.49.down_proj", "model.layers.63.mlp.experts.50.down_proj", "model.layers.63.mlp.experts.51.down_proj", "model.layers.63.mlp.experts.52.down_proj", "model.layers.63.mlp.experts.53.down_proj", "model.layers.63.mlp.experts.54.down_proj", "model.layers.63.mlp.experts.55.down_proj", "model.layers.63.mlp.experts.56.down_proj", "model.layers.63.mlp.experts.57.down_proj", "model.layers.63.mlp.experts.58.down_proj", "model.layers.63.mlp.experts.59.down_proj", "model.layers.63.mlp.experts.60.down_proj", "model.layers.63.mlp.experts.61.down_proj", "model.layers.63.mlp.experts.62.down_proj", "model.layers.63.mlp.experts.63.down_proj", "model.layers.63.mlp.experts.64.down_proj", "model.layers.63.mlp.experts.65.down_proj", "model.layers.63.mlp.experts.66.down_proj", "model.layers.63.mlp.experts.67.down_proj", "model.layers.63.mlp.experts.68.down_proj", "model.layers.63.mlp.experts.69.down_proj", "model.layers.63.mlp.experts.70.down_proj", "model.layers.63.mlp.experts.71.down_proj", "model.layers.63.mlp.experts.72.down_proj", "model.layers.63.mlp.experts.73.down_proj", "model.layers.63.mlp.experts.74.down_proj", "model.layers.63.mlp.experts.75.down_proj", "model.layers.63.mlp.experts.76.down_proj", "model.layers.63.mlp.experts.77.down_proj", "model.layers.63.mlp.experts.78.down_proj", "model.layers.63.mlp.experts.79.down_proj", "model.layers.63.mlp.experts.80.down_proj", "model.layers.63.mlp.experts.81.down_proj", "model.layers.63.mlp.experts.82.down_proj", "model.layers.63.mlp.experts.83.down_proj", "model.layers.63.mlp.experts.84.down_proj", "model.layers.63.mlp.experts.85.down_proj", "model.layers.63.mlp.experts.86.down_proj", "model.layers.63.mlp.experts.87.down_proj", "model.layers.63.mlp.experts.88.down_proj", "model.layers.63.mlp.experts.89.down_proj", "model.layers.63.mlp.experts.90.down_proj", "model.layers.63.mlp.experts.91.down_proj", "model.layers.63.mlp.experts.92.down_proj", "model.layers.63.mlp.experts.93.down_proj", "model.layers.63.mlp.experts.94.down_proj", "model.layers.63.mlp.experts.95.down_proj", "model.layers.63.mlp.experts.96.down_proj", "model.layers.63.mlp.experts.97.down_proj", "model.layers.63.mlp.experts.98.down_proj", "model.layers.63.mlp.experts.99.down_proj", "model.layers.63.mlp.experts.100.down_proj", "model.layers.63.mlp.experts.101.down_proj", "model.layers.63.mlp.experts.102.down_proj", "model.layers.63.mlp.experts.103.down_proj", "model.layers.63.mlp.experts.104.down_proj", "model.layers.63.mlp.experts.105.down_proj", "model.layers.63.mlp.experts.106.down_proj", "model.layers.63.mlp.experts.107.down_proj", "model.layers.63.mlp.experts.108.down_proj", "model.layers.63.mlp.experts.109.down_proj", "model.layers.63.mlp.experts.110.down_proj", "model.layers.63.mlp.experts.111.down_proj", "model.layers.63.mlp.experts.112.down_proj", "model.layers.63.mlp.experts.113.down_proj", "model.layers.63.mlp.experts.114.down_proj", "model.layers.63.mlp.experts.115.down_proj", "model.layers.63.mlp.experts.116.down_proj", "model.layers.63.mlp.experts.117.down_proj", "model.layers.63.mlp.experts.118.down_proj", "model.layers.63.mlp.experts.119.down_proj", "model.layers.63.mlp.experts.120.down_proj", "model.layers.63.mlp.experts.121.down_proj", "model.layers.63.mlp.experts.122.down_proj", "model.layers.63.mlp.experts.123.down_proj", "model.layers.63.mlp.experts.124.down_proj", "model.layers.63.mlp.experts.125.down_proj", "model.layers.63.mlp.experts.126.down_proj", "model.layers.63.mlp.experts.127.down_proj", "model.layers.63.mlp.experts.128.down_proj", "model.layers.63.mlp.experts.129.down_proj", "model.layers.63.mlp.experts.130.down_proj", "model.layers.63.mlp.experts.131.down_proj", "model.layers.63.mlp.experts.132.down_proj", "model.layers.63.mlp.experts.133.down_proj", "model.layers.63.mlp.experts.134.down_proj", "model.layers.63.mlp.experts.135.down_proj", "model.layers.63.mlp.experts.136.down_proj", "model.layers.63.mlp.experts.137.down_proj", "model.layers.63.mlp.experts.138.down_proj", "model.layers.63.mlp.experts.139.down_proj", "model.layers.63.mlp.experts.140.down_proj", "model.layers.63.mlp.experts.141.down_proj", "model.layers.63.mlp.experts.142.down_proj", "model.layers.63.mlp.experts.143.down_proj", "model.layers.63.mlp.experts.144.down_proj", "model.layers.63.mlp.experts.145.down_proj", "model.layers.63.mlp.experts.146.down_proj", "model.layers.63.mlp.experts.147.down_proj", "model.layers.63.mlp.experts.148.down_proj", "model.layers.63.mlp.experts.149.down_proj", "model.layers.63.mlp.experts.150.down_proj", "model.layers.63.mlp.experts.151.down_proj", "model.layers.63.mlp.experts.152.down_proj", "model.layers.63.mlp.experts.153.down_proj", "model.layers.63.mlp.experts.154.down_proj", "model.layers.63.mlp.experts.155.down_proj", "model.layers.63.mlp.experts.156.down_proj", "model.layers.63.mlp.experts.157.down_proj", "model.layers.63.mlp.experts.158.down_proj", "model.layers.63.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -8.073002099991122e-05, "dbits": 1258291200 }, { "dkld": -6.41340389847811e-05, "dbits": 2516582400 }, { "dkld": -7.127281278372644e-05, "dbits": 3774873600 }, { "dkld": -7.056891918183483e-05, "dbits": 6291456000 } ] }, { "idx": 381, "layers": [ "model.layers.64.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00010797213762998858, "dbits": 62914560 }, { "dkld": -5.994196981191913e-05, "dbits": 125829120 }, { "dkld": -8.821394294500351e-05, "dbits": 188743680 }, { "dkld": -7.949024438858032e-05, "dbits": 314572800 } ] }, { "idx": 382, "layers": [ "model.layers.64.self_attn.k_proj", "model.layers.64.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0002774814143776949, "dbits": 10485760 }, { "dkld": -0.0002055540680885315, "dbits": 20971520 }, { "dkld": -0.00020457208156586526, "dbits": 31457280 }, { "dkld": -0.00019201571121812977, "dbits": 52428800 } ] }, { "idx": 383, "layers": [ "model.layers.64.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00011456329375506002, "dbits": 62914560 }, { "dkld": -7.919166237116415e-05, "dbits": 125829120 }, { "dkld": -8.892323821783898e-05, "dbits": 188743680 }, { "dkld": -8.391495794057846e-05, "dbits": 314572800 } ] }, { "idx": 384, "layers": [ "model.layers.64.mlp.shared_experts.gate_proj", "model.layers.64.mlp.shared_experts.up_proj", "model.layers.64.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 8.335653692483069e-05, "dbits": 23592960 }, { "dkld": 4.929490387439728e-05, "dbits": 47185920 }, { "dkld": -1.4056824147706815e-05, "dbits": 70778880 }, { "dkld": -2.414416521788476e-05, "dbits": 117964800 } ] }, { "idx": 385, "layers": [ "model.layers.64.mlp.experts.0.gate_proj", "model.layers.64.mlp.experts.1.gate_proj", "model.layers.64.mlp.experts.2.gate_proj", "model.layers.64.mlp.experts.3.gate_proj", "model.layers.64.mlp.experts.4.gate_proj", "model.layers.64.mlp.experts.5.gate_proj", "model.layers.64.mlp.experts.6.gate_proj", "model.layers.64.mlp.experts.7.gate_proj", "model.layers.64.mlp.experts.8.gate_proj", "model.layers.64.mlp.experts.9.gate_proj", "model.layers.64.mlp.experts.10.gate_proj", "model.layers.64.mlp.experts.11.gate_proj", "model.layers.64.mlp.experts.12.gate_proj", "model.layers.64.mlp.experts.13.gate_proj", "model.layers.64.mlp.experts.14.gate_proj", "model.layers.64.mlp.experts.15.gate_proj", "model.layers.64.mlp.experts.16.gate_proj", "model.layers.64.mlp.experts.17.gate_proj", "model.layers.64.mlp.experts.18.gate_proj", "model.layers.64.mlp.experts.19.gate_proj", "model.layers.64.mlp.experts.20.gate_proj", "model.layers.64.mlp.experts.21.gate_proj", "model.layers.64.mlp.experts.22.gate_proj", "model.layers.64.mlp.experts.23.gate_proj", "model.layers.64.mlp.experts.24.gate_proj", "model.layers.64.mlp.experts.25.gate_proj", "model.layers.64.mlp.experts.26.gate_proj", "model.layers.64.mlp.experts.27.gate_proj", "model.layers.64.mlp.experts.28.gate_proj", "model.layers.64.mlp.experts.29.gate_proj", "model.layers.64.mlp.experts.30.gate_proj", "model.layers.64.mlp.experts.31.gate_proj", "model.layers.64.mlp.experts.32.gate_proj", "model.layers.64.mlp.experts.33.gate_proj", "model.layers.64.mlp.experts.34.gate_proj", "model.layers.64.mlp.experts.35.gate_proj", "model.layers.64.mlp.experts.36.gate_proj", "model.layers.64.mlp.experts.37.gate_proj", "model.layers.64.mlp.experts.38.gate_proj", "model.layers.64.mlp.experts.39.gate_proj", "model.layers.64.mlp.experts.40.gate_proj", "model.layers.64.mlp.experts.41.gate_proj", "model.layers.64.mlp.experts.42.gate_proj", "model.layers.64.mlp.experts.43.gate_proj", "model.layers.64.mlp.experts.44.gate_proj", "model.layers.64.mlp.experts.45.gate_proj", "model.layers.64.mlp.experts.46.gate_proj", "model.layers.64.mlp.experts.47.gate_proj", "model.layers.64.mlp.experts.48.gate_proj", "model.layers.64.mlp.experts.49.gate_proj", "model.layers.64.mlp.experts.50.gate_proj", "model.layers.64.mlp.experts.51.gate_proj", "model.layers.64.mlp.experts.52.gate_proj", "model.layers.64.mlp.experts.53.gate_proj", "model.layers.64.mlp.experts.54.gate_proj", "model.layers.64.mlp.experts.55.gate_proj", "model.layers.64.mlp.experts.56.gate_proj", "model.layers.64.mlp.experts.57.gate_proj", "model.layers.64.mlp.experts.58.gate_proj", "model.layers.64.mlp.experts.59.gate_proj", "model.layers.64.mlp.experts.60.gate_proj", "model.layers.64.mlp.experts.61.gate_proj", "model.layers.64.mlp.experts.62.gate_proj", "model.layers.64.mlp.experts.63.gate_proj", "model.layers.64.mlp.experts.64.gate_proj", "model.layers.64.mlp.experts.65.gate_proj", "model.layers.64.mlp.experts.66.gate_proj", "model.layers.64.mlp.experts.67.gate_proj", "model.layers.64.mlp.experts.68.gate_proj", "model.layers.64.mlp.experts.69.gate_proj", "model.layers.64.mlp.experts.70.gate_proj", "model.layers.64.mlp.experts.71.gate_proj", "model.layers.64.mlp.experts.72.gate_proj", "model.layers.64.mlp.experts.73.gate_proj", "model.layers.64.mlp.experts.74.gate_proj", "model.layers.64.mlp.experts.75.gate_proj", "model.layers.64.mlp.experts.76.gate_proj", "model.layers.64.mlp.experts.77.gate_proj", "model.layers.64.mlp.experts.78.gate_proj", "model.layers.64.mlp.experts.79.gate_proj", "model.layers.64.mlp.experts.80.gate_proj", "model.layers.64.mlp.experts.81.gate_proj", "model.layers.64.mlp.experts.82.gate_proj", "model.layers.64.mlp.experts.83.gate_proj", "model.layers.64.mlp.experts.84.gate_proj", "model.layers.64.mlp.experts.85.gate_proj", "model.layers.64.mlp.experts.86.gate_proj", "model.layers.64.mlp.experts.87.gate_proj", "model.layers.64.mlp.experts.88.gate_proj", "model.layers.64.mlp.experts.89.gate_proj", "model.layers.64.mlp.experts.90.gate_proj", "model.layers.64.mlp.experts.91.gate_proj", "model.layers.64.mlp.experts.92.gate_proj", "model.layers.64.mlp.experts.93.gate_proj", "model.layers.64.mlp.experts.94.gate_proj", "model.layers.64.mlp.experts.95.gate_proj", "model.layers.64.mlp.experts.96.gate_proj", "model.layers.64.mlp.experts.97.gate_proj", "model.layers.64.mlp.experts.98.gate_proj", "model.layers.64.mlp.experts.99.gate_proj", "model.layers.64.mlp.experts.100.gate_proj", "model.layers.64.mlp.experts.101.gate_proj", "model.layers.64.mlp.experts.102.gate_proj", "model.layers.64.mlp.experts.103.gate_proj", "model.layers.64.mlp.experts.104.gate_proj", "model.layers.64.mlp.experts.105.gate_proj", "model.layers.64.mlp.experts.106.gate_proj", "model.layers.64.mlp.experts.107.gate_proj", "model.layers.64.mlp.experts.108.gate_proj", "model.layers.64.mlp.experts.109.gate_proj", "model.layers.64.mlp.experts.110.gate_proj", "model.layers.64.mlp.experts.111.gate_proj", "model.layers.64.mlp.experts.112.gate_proj", "model.layers.64.mlp.experts.113.gate_proj", "model.layers.64.mlp.experts.114.gate_proj", "model.layers.64.mlp.experts.115.gate_proj", "model.layers.64.mlp.experts.116.gate_proj", "model.layers.64.mlp.experts.117.gate_proj", "model.layers.64.mlp.experts.118.gate_proj", "model.layers.64.mlp.experts.119.gate_proj", "model.layers.64.mlp.experts.120.gate_proj", "model.layers.64.mlp.experts.121.gate_proj", "model.layers.64.mlp.experts.122.gate_proj", "model.layers.64.mlp.experts.123.gate_proj", "model.layers.64.mlp.experts.124.gate_proj", "model.layers.64.mlp.experts.125.gate_proj", "model.layers.64.mlp.experts.126.gate_proj", "model.layers.64.mlp.experts.127.gate_proj", "model.layers.64.mlp.experts.128.gate_proj", "model.layers.64.mlp.experts.129.gate_proj", "model.layers.64.mlp.experts.130.gate_proj", "model.layers.64.mlp.experts.131.gate_proj", "model.layers.64.mlp.experts.132.gate_proj", "model.layers.64.mlp.experts.133.gate_proj", "model.layers.64.mlp.experts.134.gate_proj", "model.layers.64.mlp.experts.135.gate_proj", "model.layers.64.mlp.experts.136.gate_proj", "model.layers.64.mlp.experts.137.gate_proj", "model.layers.64.mlp.experts.138.gate_proj", "model.layers.64.mlp.experts.139.gate_proj", "model.layers.64.mlp.experts.140.gate_proj", "model.layers.64.mlp.experts.141.gate_proj", "model.layers.64.mlp.experts.142.gate_proj", "model.layers.64.mlp.experts.143.gate_proj", "model.layers.64.mlp.experts.144.gate_proj", "model.layers.64.mlp.experts.145.gate_proj", "model.layers.64.mlp.experts.146.gate_proj", "model.layers.64.mlp.experts.147.gate_proj", "model.layers.64.mlp.experts.148.gate_proj", "model.layers.64.mlp.experts.149.gate_proj", "model.layers.64.mlp.experts.150.gate_proj", "model.layers.64.mlp.experts.151.gate_proj", "model.layers.64.mlp.experts.152.gate_proj", "model.layers.64.mlp.experts.153.gate_proj", "model.layers.64.mlp.experts.154.gate_proj", "model.layers.64.mlp.experts.155.gate_proj", "model.layers.64.mlp.experts.156.gate_proj", "model.layers.64.mlp.experts.157.gate_proj", "model.layers.64.mlp.experts.158.gate_proj", "model.layers.64.mlp.experts.159.gate_proj", "model.layers.64.mlp.experts.0.up_proj", "model.layers.64.mlp.experts.1.up_proj", "model.layers.64.mlp.experts.2.up_proj", "model.layers.64.mlp.experts.3.up_proj", "model.layers.64.mlp.experts.4.up_proj", "model.layers.64.mlp.experts.5.up_proj", "model.layers.64.mlp.experts.6.up_proj", "model.layers.64.mlp.experts.7.up_proj", "model.layers.64.mlp.experts.8.up_proj", "model.layers.64.mlp.experts.9.up_proj", "model.layers.64.mlp.experts.10.up_proj", "model.layers.64.mlp.experts.11.up_proj", "model.layers.64.mlp.experts.12.up_proj", "model.layers.64.mlp.experts.13.up_proj", "model.layers.64.mlp.experts.14.up_proj", "model.layers.64.mlp.experts.15.up_proj", "model.layers.64.mlp.experts.16.up_proj", "model.layers.64.mlp.experts.17.up_proj", "model.layers.64.mlp.experts.18.up_proj", "model.layers.64.mlp.experts.19.up_proj", "model.layers.64.mlp.experts.20.up_proj", "model.layers.64.mlp.experts.21.up_proj", "model.layers.64.mlp.experts.22.up_proj", "model.layers.64.mlp.experts.23.up_proj", "model.layers.64.mlp.experts.24.up_proj", "model.layers.64.mlp.experts.25.up_proj", "model.layers.64.mlp.experts.26.up_proj", "model.layers.64.mlp.experts.27.up_proj", "model.layers.64.mlp.experts.28.up_proj", "model.layers.64.mlp.experts.29.up_proj", "model.layers.64.mlp.experts.30.up_proj", "model.layers.64.mlp.experts.31.up_proj", "model.layers.64.mlp.experts.32.up_proj", "model.layers.64.mlp.experts.33.up_proj", "model.layers.64.mlp.experts.34.up_proj", "model.layers.64.mlp.experts.35.up_proj", "model.layers.64.mlp.experts.36.up_proj", "model.layers.64.mlp.experts.37.up_proj", "model.layers.64.mlp.experts.38.up_proj", "model.layers.64.mlp.experts.39.up_proj", "model.layers.64.mlp.experts.40.up_proj", "model.layers.64.mlp.experts.41.up_proj", "model.layers.64.mlp.experts.42.up_proj", "model.layers.64.mlp.experts.43.up_proj", "model.layers.64.mlp.experts.44.up_proj", "model.layers.64.mlp.experts.45.up_proj", "model.layers.64.mlp.experts.46.up_proj", "model.layers.64.mlp.experts.47.up_proj", "model.layers.64.mlp.experts.48.up_proj", "model.layers.64.mlp.experts.49.up_proj", "model.layers.64.mlp.experts.50.up_proj", "model.layers.64.mlp.experts.51.up_proj", "model.layers.64.mlp.experts.52.up_proj", "model.layers.64.mlp.experts.53.up_proj", "model.layers.64.mlp.experts.54.up_proj", "model.layers.64.mlp.experts.55.up_proj", "model.layers.64.mlp.experts.56.up_proj", "model.layers.64.mlp.experts.57.up_proj", "model.layers.64.mlp.experts.58.up_proj", "model.layers.64.mlp.experts.59.up_proj", "model.layers.64.mlp.experts.60.up_proj", "model.layers.64.mlp.experts.61.up_proj", "model.layers.64.mlp.experts.62.up_proj", "model.layers.64.mlp.experts.63.up_proj", "model.layers.64.mlp.experts.64.up_proj", "model.layers.64.mlp.experts.65.up_proj", "model.layers.64.mlp.experts.66.up_proj", "model.layers.64.mlp.experts.67.up_proj", "model.layers.64.mlp.experts.68.up_proj", "model.layers.64.mlp.experts.69.up_proj", "model.layers.64.mlp.experts.70.up_proj", "model.layers.64.mlp.experts.71.up_proj", "model.layers.64.mlp.experts.72.up_proj", "model.layers.64.mlp.experts.73.up_proj", "model.layers.64.mlp.experts.74.up_proj", "model.layers.64.mlp.experts.75.up_proj", "model.layers.64.mlp.experts.76.up_proj", "model.layers.64.mlp.experts.77.up_proj", "model.layers.64.mlp.experts.78.up_proj", "model.layers.64.mlp.experts.79.up_proj", "model.layers.64.mlp.experts.80.up_proj", "model.layers.64.mlp.experts.81.up_proj", "model.layers.64.mlp.experts.82.up_proj", "model.layers.64.mlp.experts.83.up_proj", "model.layers.64.mlp.experts.84.up_proj", "model.layers.64.mlp.experts.85.up_proj", "model.layers.64.mlp.experts.86.up_proj", "model.layers.64.mlp.experts.87.up_proj", "model.layers.64.mlp.experts.88.up_proj", "model.layers.64.mlp.experts.89.up_proj", "model.layers.64.mlp.experts.90.up_proj", "model.layers.64.mlp.experts.91.up_proj", "model.layers.64.mlp.experts.92.up_proj", "model.layers.64.mlp.experts.93.up_proj", "model.layers.64.mlp.experts.94.up_proj", "model.layers.64.mlp.experts.95.up_proj", "model.layers.64.mlp.experts.96.up_proj", "model.layers.64.mlp.experts.97.up_proj", "model.layers.64.mlp.experts.98.up_proj", "model.layers.64.mlp.experts.99.up_proj", "model.layers.64.mlp.experts.100.up_proj", "model.layers.64.mlp.experts.101.up_proj", "model.layers.64.mlp.experts.102.up_proj", "model.layers.64.mlp.experts.103.up_proj", "model.layers.64.mlp.experts.104.up_proj", "model.layers.64.mlp.experts.105.up_proj", "model.layers.64.mlp.experts.106.up_proj", "model.layers.64.mlp.experts.107.up_proj", "model.layers.64.mlp.experts.108.up_proj", "model.layers.64.mlp.experts.109.up_proj", "model.layers.64.mlp.experts.110.up_proj", "model.layers.64.mlp.experts.111.up_proj", "model.layers.64.mlp.experts.112.up_proj", "model.layers.64.mlp.experts.113.up_proj", "model.layers.64.mlp.experts.114.up_proj", "model.layers.64.mlp.experts.115.up_proj", "model.layers.64.mlp.experts.116.up_proj", "model.layers.64.mlp.experts.117.up_proj", "model.layers.64.mlp.experts.118.up_proj", "model.layers.64.mlp.experts.119.up_proj", "model.layers.64.mlp.experts.120.up_proj", "model.layers.64.mlp.experts.121.up_proj", "model.layers.64.mlp.experts.122.up_proj", "model.layers.64.mlp.experts.123.up_proj", "model.layers.64.mlp.experts.124.up_proj", "model.layers.64.mlp.experts.125.up_proj", "model.layers.64.mlp.experts.126.up_proj", "model.layers.64.mlp.experts.127.up_proj", "model.layers.64.mlp.experts.128.up_proj", "model.layers.64.mlp.experts.129.up_proj", "model.layers.64.mlp.experts.130.up_proj", "model.layers.64.mlp.experts.131.up_proj", "model.layers.64.mlp.experts.132.up_proj", "model.layers.64.mlp.experts.133.up_proj", "model.layers.64.mlp.experts.134.up_proj", "model.layers.64.mlp.experts.135.up_proj", "model.layers.64.mlp.experts.136.up_proj", "model.layers.64.mlp.experts.137.up_proj", "model.layers.64.mlp.experts.138.up_proj", "model.layers.64.mlp.experts.139.up_proj", "model.layers.64.mlp.experts.140.up_proj", "model.layers.64.mlp.experts.141.up_proj", "model.layers.64.mlp.experts.142.up_proj", "model.layers.64.mlp.experts.143.up_proj", "model.layers.64.mlp.experts.144.up_proj", "model.layers.64.mlp.experts.145.up_proj", "model.layers.64.mlp.experts.146.up_proj", "model.layers.64.mlp.experts.147.up_proj", "model.layers.64.mlp.experts.148.up_proj", "model.layers.64.mlp.experts.149.up_proj", "model.layers.64.mlp.experts.150.up_proj", "model.layers.64.mlp.experts.151.up_proj", "model.layers.64.mlp.experts.152.up_proj", "model.layers.64.mlp.experts.153.up_proj", "model.layers.64.mlp.experts.154.up_proj", "model.layers.64.mlp.experts.155.up_proj", "model.layers.64.mlp.experts.156.up_proj", "model.layers.64.mlp.experts.157.up_proj", "model.layers.64.mlp.experts.158.up_proj", "model.layers.64.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -9.550377726555426e-05, "dbits": 2516582400 }, { "dkld": -0.00014476887881756384, "dbits": 5033164800 }, { "dkld": -0.00014689099043607712, "dbits": 7549747200 }, { "dkld": -0.0001484552398324096, "dbits": 12582912000 } ] }, { "idx": 386, "layers": [ "model.layers.64.mlp.experts.0.down_proj", "model.layers.64.mlp.experts.1.down_proj", "model.layers.64.mlp.experts.2.down_proj", "model.layers.64.mlp.experts.3.down_proj", "model.layers.64.mlp.experts.4.down_proj", "model.layers.64.mlp.experts.5.down_proj", "model.layers.64.mlp.experts.6.down_proj", "model.layers.64.mlp.experts.7.down_proj", "model.layers.64.mlp.experts.8.down_proj", "model.layers.64.mlp.experts.9.down_proj", "model.layers.64.mlp.experts.10.down_proj", "model.layers.64.mlp.experts.11.down_proj", "model.layers.64.mlp.experts.12.down_proj", "model.layers.64.mlp.experts.13.down_proj", "model.layers.64.mlp.experts.14.down_proj", "model.layers.64.mlp.experts.15.down_proj", "model.layers.64.mlp.experts.16.down_proj", "model.layers.64.mlp.experts.17.down_proj", "model.layers.64.mlp.experts.18.down_proj", "model.layers.64.mlp.experts.19.down_proj", "model.layers.64.mlp.experts.20.down_proj", "model.layers.64.mlp.experts.21.down_proj", "model.layers.64.mlp.experts.22.down_proj", "model.layers.64.mlp.experts.23.down_proj", "model.layers.64.mlp.experts.24.down_proj", "model.layers.64.mlp.experts.25.down_proj", "model.layers.64.mlp.experts.26.down_proj", "model.layers.64.mlp.experts.27.down_proj", "model.layers.64.mlp.experts.28.down_proj", "model.layers.64.mlp.experts.29.down_proj", "model.layers.64.mlp.experts.30.down_proj", "model.layers.64.mlp.experts.31.down_proj", "model.layers.64.mlp.experts.32.down_proj", "model.layers.64.mlp.experts.33.down_proj", "model.layers.64.mlp.experts.34.down_proj", "model.layers.64.mlp.experts.35.down_proj", "model.layers.64.mlp.experts.36.down_proj", "model.layers.64.mlp.experts.37.down_proj", "model.layers.64.mlp.experts.38.down_proj", "model.layers.64.mlp.experts.39.down_proj", "model.layers.64.mlp.experts.40.down_proj", "model.layers.64.mlp.experts.41.down_proj", "model.layers.64.mlp.experts.42.down_proj", "model.layers.64.mlp.experts.43.down_proj", "model.layers.64.mlp.experts.44.down_proj", "model.layers.64.mlp.experts.45.down_proj", "model.layers.64.mlp.experts.46.down_proj", "model.layers.64.mlp.experts.47.down_proj", "model.layers.64.mlp.experts.48.down_proj", "model.layers.64.mlp.experts.49.down_proj", "model.layers.64.mlp.experts.50.down_proj", "model.layers.64.mlp.experts.51.down_proj", "model.layers.64.mlp.experts.52.down_proj", "model.layers.64.mlp.experts.53.down_proj", "model.layers.64.mlp.experts.54.down_proj", "model.layers.64.mlp.experts.55.down_proj", "model.layers.64.mlp.experts.56.down_proj", "model.layers.64.mlp.experts.57.down_proj", "model.layers.64.mlp.experts.58.down_proj", "model.layers.64.mlp.experts.59.down_proj", "model.layers.64.mlp.experts.60.down_proj", "model.layers.64.mlp.experts.61.down_proj", "model.layers.64.mlp.experts.62.down_proj", "model.layers.64.mlp.experts.63.down_proj", "model.layers.64.mlp.experts.64.down_proj", "model.layers.64.mlp.experts.65.down_proj", "model.layers.64.mlp.experts.66.down_proj", "model.layers.64.mlp.experts.67.down_proj", "model.layers.64.mlp.experts.68.down_proj", "model.layers.64.mlp.experts.69.down_proj", "model.layers.64.mlp.experts.70.down_proj", "model.layers.64.mlp.experts.71.down_proj", "model.layers.64.mlp.experts.72.down_proj", "model.layers.64.mlp.experts.73.down_proj", "model.layers.64.mlp.experts.74.down_proj", "model.layers.64.mlp.experts.75.down_proj", "model.layers.64.mlp.experts.76.down_proj", "model.layers.64.mlp.experts.77.down_proj", "model.layers.64.mlp.experts.78.down_proj", "model.layers.64.mlp.experts.79.down_proj", "model.layers.64.mlp.experts.80.down_proj", "model.layers.64.mlp.experts.81.down_proj", "model.layers.64.mlp.experts.82.down_proj", "model.layers.64.mlp.experts.83.down_proj", "model.layers.64.mlp.experts.84.down_proj", "model.layers.64.mlp.experts.85.down_proj", "model.layers.64.mlp.experts.86.down_proj", "model.layers.64.mlp.experts.87.down_proj", "model.layers.64.mlp.experts.88.down_proj", "model.layers.64.mlp.experts.89.down_proj", "model.layers.64.mlp.experts.90.down_proj", "model.layers.64.mlp.experts.91.down_proj", "model.layers.64.mlp.experts.92.down_proj", "model.layers.64.mlp.experts.93.down_proj", "model.layers.64.mlp.experts.94.down_proj", "model.layers.64.mlp.experts.95.down_proj", "model.layers.64.mlp.experts.96.down_proj", "model.layers.64.mlp.experts.97.down_proj", "model.layers.64.mlp.experts.98.down_proj", "model.layers.64.mlp.experts.99.down_proj", "model.layers.64.mlp.experts.100.down_proj", "model.layers.64.mlp.experts.101.down_proj", "model.layers.64.mlp.experts.102.down_proj", "model.layers.64.mlp.experts.103.down_proj", "model.layers.64.mlp.experts.104.down_proj", "model.layers.64.mlp.experts.105.down_proj", "model.layers.64.mlp.experts.106.down_proj", "model.layers.64.mlp.experts.107.down_proj", "model.layers.64.mlp.experts.108.down_proj", "model.layers.64.mlp.experts.109.down_proj", "model.layers.64.mlp.experts.110.down_proj", "model.layers.64.mlp.experts.111.down_proj", "model.layers.64.mlp.experts.112.down_proj", "model.layers.64.mlp.experts.113.down_proj", "model.layers.64.mlp.experts.114.down_proj", "model.layers.64.mlp.experts.115.down_proj", "model.layers.64.mlp.experts.116.down_proj", "model.layers.64.mlp.experts.117.down_proj", "model.layers.64.mlp.experts.118.down_proj", "model.layers.64.mlp.experts.119.down_proj", "model.layers.64.mlp.experts.120.down_proj", "model.layers.64.mlp.experts.121.down_proj", "model.layers.64.mlp.experts.122.down_proj", "model.layers.64.mlp.experts.123.down_proj", "model.layers.64.mlp.experts.124.down_proj", "model.layers.64.mlp.experts.125.down_proj", "model.layers.64.mlp.experts.126.down_proj", "model.layers.64.mlp.experts.127.down_proj", "model.layers.64.mlp.experts.128.down_proj", "model.layers.64.mlp.experts.129.down_proj", "model.layers.64.mlp.experts.130.down_proj", "model.layers.64.mlp.experts.131.down_proj", "model.layers.64.mlp.experts.132.down_proj", "model.layers.64.mlp.experts.133.down_proj", "model.layers.64.mlp.experts.134.down_proj", "model.layers.64.mlp.experts.135.down_proj", "model.layers.64.mlp.experts.136.down_proj", "model.layers.64.mlp.experts.137.down_proj", "model.layers.64.mlp.experts.138.down_proj", "model.layers.64.mlp.experts.139.down_proj", "model.layers.64.mlp.experts.140.down_proj", "model.layers.64.mlp.experts.141.down_proj", "model.layers.64.mlp.experts.142.down_proj", "model.layers.64.mlp.experts.143.down_proj", "model.layers.64.mlp.experts.144.down_proj", "model.layers.64.mlp.experts.145.down_proj", "model.layers.64.mlp.experts.146.down_proj", "model.layers.64.mlp.experts.147.down_proj", "model.layers.64.mlp.experts.148.down_proj", "model.layers.64.mlp.experts.149.down_proj", "model.layers.64.mlp.experts.150.down_proj", "model.layers.64.mlp.experts.151.down_proj", "model.layers.64.mlp.experts.152.down_proj", "model.layers.64.mlp.experts.153.down_proj", "model.layers.64.mlp.experts.154.down_proj", "model.layers.64.mlp.experts.155.down_proj", "model.layers.64.mlp.experts.156.down_proj", "model.layers.64.mlp.experts.157.down_proj", "model.layers.64.mlp.experts.158.down_proj", "model.layers.64.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": 8.84272158145627e-06, "dbits": 1258291200 }, { "dkld": -0.00010475069284439364, "dbits": 2516582400 }, { "dkld": -8.093686774374165e-05, "dbits": 3774873600 }, { "dkld": -5.861278623342514e-05, "dbits": 6291456000 } ] }, { "idx": 387, "layers": [ "model.layers.65.self_attn.q_proj" ], "candidates": [ { "dkld": -6.0871243476867676e-05, "dbits": 62914560 }, { "dkld": 9.64608043431403e-06, "dbits": 125829120 }, { "dkld": -5.8827921748161316e-05, "dbits": 188743680 }, { "dkld": -9.949132800113336e-06, "dbits": 314572800 } ] }, { "idx": 388, "layers": [ "model.layers.65.self_attn.k_proj", "model.layers.65.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0002603513188660228, "dbits": 10485760 }, { "dkld": -0.0003869227133691394, "dbits": 20971520 }, { "dkld": -0.0003024564124643886, "dbits": 31457280 }, { "dkld": -0.00031655160710215846, "dbits": 52428800 } ] }, { "idx": 389, "layers": [ "model.layers.65.self_attn.o_proj" ], "candidates": [ { "dkld": -8.084159344434738e-05, "dbits": 62914560 }, { "dkld": -8.97064805030906e-05, "dbits": 125829120 }, { "dkld": -9.884256869555075e-05, "dbits": 188743680 }, { "dkld": -7.70461745560197e-05, "dbits": 314572800 } ] }, { "idx": 390, "layers": [ "model.layers.65.mlp.shared_experts.gate_proj", "model.layers.65.mlp.shared_experts.up_proj", "model.layers.65.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00022509563714266656, "dbits": 23592960 }, { "dkld": -0.00022751092910766602, "dbits": 47185920 }, { "dkld": -0.00013096779584885476, "dbits": 70778880 }, { "dkld": -0.00015886072069407065, "dbits": 117964800 } ] }, { "idx": 391, "layers": [ "model.layers.65.mlp.experts.0.gate_proj", "model.layers.65.mlp.experts.1.gate_proj", "model.layers.65.mlp.experts.2.gate_proj", "model.layers.65.mlp.experts.3.gate_proj", "model.layers.65.mlp.experts.4.gate_proj", "model.layers.65.mlp.experts.5.gate_proj", "model.layers.65.mlp.experts.6.gate_proj", "model.layers.65.mlp.experts.7.gate_proj", "model.layers.65.mlp.experts.8.gate_proj", "model.layers.65.mlp.experts.9.gate_proj", "model.layers.65.mlp.experts.10.gate_proj", "model.layers.65.mlp.experts.11.gate_proj", "model.layers.65.mlp.experts.12.gate_proj", "model.layers.65.mlp.experts.13.gate_proj", "model.layers.65.mlp.experts.14.gate_proj", "model.layers.65.mlp.experts.15.gate_proj", "model.layers.65.mlp.experts.16.gate_proj", "model.layers.65.mlp.experts.17.gate_proj", "model.layers.65.mlp.experts.18.gate_proj", "model.layers.65.mlp.experts.19.gate_proj", "model.layers.65.mlp.experts.20.gate_proj", "model.layers.65.mlp.experts.21.gate_proj", "model.layers.65.mlp.experts.22.gate_proj", "model.layers.65.mlp.experts.23.gate_proj", "model.layers.65.mlp.experts.24.gate_proj", "model.layers.65.mlp.experts.25.gate_proj", "model.layers.65.mlp.experts.26.gate_proj", "model.layers.65.mlp.experts.27.gate_proj", "model.layers.65.mlp.experts.28.gate_proj", "model.layers.65.mlp.experts.29.gate_proj", "model.layers.65.mlp.experts.30.gate_proj", "model.layers.65.mlp.experts.31.gate_proj", "model.layers.65.mlp.experts.32.gate_proj", "model.layers.65.mlp.experts.33.gate_proj", "model.layers.65.mlp.experts.34.gate_proj", "model.layers.65.mlp.experts.35.gate_proj", "model.layers.65.mlp.experts.36.gate_proj", "model.layers.65.mlp.experts.37.gate_proj", "model.layers.65.mlp.experts.38.gate_proj", "model.layers.65.mlp.experts.39.gate_proj", "model.layers.65.mlp.experts.40.gate_proj", "model.layers.65.mlp.experts.41.gate_proj", "model.layers.65.mlp.experts.42.gate_proj", "model.layers.65.mlp.experts.43.gate_proj", "model.layers.65.mlp.experts.44.gate_proj", "model.layers.65.mlp.experts.45.gate_proj", "model.layers.65.mlp.experts.46.gate_proj", "model.layers.65.mlp.experts.47.gate_proj", "model.layers.65.mlp.experts.48.gate_proj", "model.layers.65.mlp.experts.49.gate_proj", "model.layers.65.mlp.experts.50.gate_proj", "model.layers.65.mlp.experts.51.gate_proj", "model.layers.65.mlp.experts.52.gate_proj", "model.layers.65.mlp.experts.53.gate_proj", "model.layers.65.mlp.experts.54.gate_proj", "model.layers.65.mlp.experts.55.gate_proj", "model.layers.65.mlp.experts.56.gate_proj", "model.layers.65.mlp.experts.57.gate_proj", "model.layers.65.mlp.experts.58.gate_proj", "model.layers.65.mlp.experts.59.gate_proj", "model.layers.65.mlp.experts.60.gate_proj", "model.layers.65.mlp.experts.61.gate_proj", "model.layers.65.mlp.experts.62.gate_proj", "model.layers.65.mlp.experts.63.gate_proj", "model.layers.65.mlp.experts.64.gate_proj", "model.layers.65.mlp.experts.65.gate_proj", "model.layers.65.mlp.experts.66.gate_proj", "model.layers.65.mlp.experts.67.gate_proj", "model.layers.65.mlp.experts.68.gate_proj", "model.layers.65.mlp.experts.69.gate_proj", "model.layers.65.mlp.experts.70.gate_proj", "model.layers.65.mlp.experts.71.gate_proj", "model.layers.65.mlp.experts.72.gate_proj", "model.layers.65.mlp.experts.73.gate_proj", "model.layers.65.mlp.experts.74.gate_proj", "model.layers.65.mlp.experts.75.gate_proj", "model.layers.65.mlp.experts.76.gate_proj", "model.layers.65.mlp.experts.77.gate_proj", "model.layers.65.mlp.experts.78.gate_proj", "model.layers.65.mlp.experts.79.gate_proj", "model.layers.65.mlp.experts.80.gate_proj", "model.layers.65.mlp.experts.81.gate_proj", "model.layers.65.mlp.experts.82.gate_proj", "model.layers.65.mlp.experts.83.gate_proj", "model.layers.65.mlp.experts.84.gate_proj", "model.layers.65.mlp.experts.85.gate_proj", "model.layers.65.mlp.experts.86.gate_proj", "model.layers.65.mlp.experts.87.gate_proj", "model.layers.65.mlp.experts.88.gate_proj", "model.layers.65.mlp.experts.89.gate_proj", "model.layers.65.mlp.experts.90.gate_proj", "model.layers.65.mlp.experts.91.gate_proj", "model.layers.65.mlp.experts.92.gate_proj", "model.layers.65.mlp.experts.93.gate_proj", "model.layers.65.mlp.experts.94.gate_proj", "model.layers.65.mlp.experts.95.gate_proj", "model.layers.65.mlp.experts.96.gate_proj", "model.layers.65.mlp.experts.97.gate_proj", "model.layers.65.mlp.experts.98.gate_proj", "model.layers.65.mlp.experts.99.gate_proj", "model.layers.65.mlp.experts.100.gate_proj", "model.layers.65.mlp.experts.101.gate_proj", "model.layers.65.mlp.experts.102.gate_proj", "model.layers.65.mlp.experts.103.gate_proj", "model.layers.65.mlp.experts.104.gate_proj", "model.layers.65.mlp.experts.105.gate_proj", "model.layers.65.mlp.experts.106.gate_proj", "model.layers.65.mlp.experts.107.gate_proj", "model.layers.65.mlp.experts.108.gate_proj", "model.layers.65.mlp.experts.109.gate_proj", "model.layers.65.mlp.experts.110.gate_proj", "model.layers.65.mlp.experts.111.gate_proj", "model.layers.65.mlp.experts.112.gate_proj", "model.layers.65.mlp.experts.113.gate_proj", "model.layers.65.mlp.experts.114.gate_proj", "model.layers.65.mlp.experts.115.gate_proj", "model.layers.65.mlp.experts.116.gate_proj", "model.layers.65.mlp.experts.117.gate_proj", "model.layers.65.mlp.experts.118.gate_proj", "model.layers.65.mlp.experts.119.gate_proj", "model.layers.65.mlp.experts.120.gate_proj", "model.layers.65.mlp.experts.121.gate_proj", "model.layers.65.mlp.experts.122.gate_proj", "model.layers.65.mlp.experts.123.gate_proj", "model.layers.65.mlp.experts.124.gate_proj", "model.layers.65.mlp.experts.125.gate_proj", "model.layers.65.mlp.experts.126.gate_proj", "model.layers.65.mlp.experts.127.gate_proj", "model.layers.65.mlp.experts.128.gate_proj", "model.layers.65.mlp.experts.129.gate_proj", "model.layers.65.mlp.experts.130.gate_proj", "model.layers.65.mlp.experts.131.gate_proj", "model.layers.65.mlp.experts.132.gate_proj", "model.layers.65.mlp.experts.133.gate_proj", "model.layers.65.mlp.experts.134.gate_proj", "model.layers.65.mlp.experts.135.gate_proj", "model.layers.65.mlp.experts.136.gate_proj", "model.layers.65.mlp.experts.137.gate_proj", "model.layers.65.mlp.experts.138.gate_proj", "model.layers.65.mlp.experts.139.gate_proj", "model.layers.65.mlp.experts.140.gate_proj", "model.layers.65.mlp.experts.141.gate_proj", "model.layers.65.mlp.experts.142.gate_proj", "model.layers.65.mlp.experts.143.gate_proj", "model.layers.65.mlp.experts.144.gate_proj", "model.layers.65.mlp.experts.145.gate_proj", "model.layers.65.mlp.experts.146.gate_proj", "model.layers.65.mlp.experts.147.gate_proj", "model.layers.65.mlp.experts.148.gate_proj", "model.layers.65.mlp.experts.149.gate_proj", "model.layers.65.mlp.experts.150.gate_proj", "model.layers.65.mlp.experts.151.gate_proj", "model.layers.65.mlp.experts.152.gate_proj", "model.layers.65.mlp.experts.153.gate_proj", "model.layers.65.mlp.experts.154.gate_proj", "model.layers.65.mlp.experts.155.gate_proj", "model.layers.65.mlp.experts.156.gate_proj", "model.layers.65.mlp.experts.157.gate_proj", "model.layers.65.mlp.experts.158.gate_proj", "model.layers.65.mlp.experts.159.gate_proj", "model.layers.65.mlp.experts.0.up_proj", "model.layers.65.mlp.experts.1.up_proj", "model.layers.65.mlp.experts.2.up_proj", "model.layers.65.mlp.experts.3.up_proj", "model.layers.65.mlp.experts.4.up_proj", "model.layers.65.mlp.experts.5.up_proj", "model.layers.65.mlp.experts.6.up_proj", "model.layers.65.mlp.experts.7.up_proj", "model.layers.65.mlp.experts.8.up_proj", "model.layers.65.mlp.experts.9.up_proj", "model.layers.65.mlp.experts.10.up_proj", "model.layers.65.mlp.experts.11.up_proj", "model.layers.65.mlp.experts.12.up_proj", "model.layers.65.mlp.experts.13.up_proj", "model.layers.65.mlp.experts.14.up_proj", "model.layers.65.mlp.experts.15.up_proj", "model.layers.65.mlp.experts.16.up_proj", "model.layers.65.mlp.experts.17.up_proj", "model.layers.65.mlp.experts.18.up_proj", "model.layers.65.mlp.experts.19.up_proj", "model.layers.65.mlp.experts.20.up_proj", "model.layers.65.mlp.experts.21.up_proj", "model.layers.65.mlp.experts.22.up_proj", "model.layers.65.mlp.experts.23.up_proj", "model.layers.65.mlp.experts.24.up_proj", "model.layers.65.mlp.experts.25.up_proj", "model.layers.65.mlp.experts.26.up_proj", "model.layers.65.mlp.experts.27.up_proj", "model.layers.65.mlp.experts.28.up_proj", "model.layers.65.mlp.experts.29.up_proj", "model.layers.65.mlp.experts.30.up_proj", "model.layers.65.mlp.experts.31.up_proj", "model.layers.65.mlp.experts.32.up_proj", "model.layers.65.mlp.experts.33.up_proj", "model.layers.65.mlp.experts.34.up_proj", "model.layers.65.mlp.experts.35.up_proj", "model.layers.65.mlp.experts.36.up_proj", "model.layers.65.mlp.experts.37.up_proj", "model.layers.65.mlp.experts.38.up_proj", "model.layers.65.mlp.experts.39.up_proj", "model.layers.65.mlp.experts.40.up_proj", "model.layers.65.mlp.experts.41.up_proj", "model.layers.65.mlp.experts.42.up_proj", "model.layers.65.mlp.experts.43.up_proj", "model.layers.65.mlp.experts.44.up_proj", "model.layers.65.mlp.experts.45.up_proj", "model.layers.65.mlp.experts.46.up_proj", "model.layers.65.mlp.experts.47.up_proj", "model.layers.65.mlp.experts.48.up_proj", "model.layers.65.mlp.experts.49.up_proj", "model.layers.65.mlp.experts.50.up_proj", "model.layers.65.mlp.experts.51.up_proj", "model.layers.65.mlp.experts.52.up_proj", "model.layers.65.mlp.experts.53.up_proj", "model.layers.65.mlp.experts.54.up_proj", "model.layers.65.mlp.experts.55.up_proj", "model.layers.65.mlp.experts.56.up_proj", "model.layers.65.mlp.experts.57.up_proj", "model.layers.65.mlp.experts.58.up_proj", "model.layers.65.mlp.experts.59.up_proj", "model.layers.65.mlp.experts.60.up_proj", "model.layers.65.mlp.experts.61.up_proj", "model.layers.65.mlp.experts.62.up_proj", "model.layers.65.mlp.experts.63.up_proj", "model.layers.65.mlp.experts.64.up_proj", "model.layers.65.mlp.experts.65.up_proj", "model.layers.65.mlp.experts.66.up_proj", "model.layers.65.mlp.experts.67.up_proj", "model.layers.65.mlp.experts.68.up_proj", "model.layers.65.mlp.experts.69.up_proj", "model.layers.65.mlp.experts.70.up_proj", "model.layers.65.mlp.experts.71.up_proj", "model.layers.65.mlp.experts.72.up_proj", "model.layers.65.mlp.experts.73.up_proj", "model.layers.65.mlp.experts.74.up_proj", "model.layers.65.mlp.experts.75.up_proj", "model.layers.65.mlp.experts.76.up_proj", "model.layers.65.mlp.experts.77.up_proj", "model.layers.65.mlp.experts.78.up_proj", "model.layers.65.mlp.experts.79.up_proj", "model.layers.65.mlp.experts.80.up_proj", "model.layers.65.mlp.experts.81.up_proj", "model.layers.65.mlp.experts.82.up_proj", "model.layers.65.mlp.experts.83.up_proj", "model.layers.65.mlp.experts.84.up_proj", "model.layers.65.mlp.experts.85.up_proj", "model.layers.65.mlp.experts.86.up_proj", "model.layers.65.mlp.experts.87.up_proj", "model.layers.65.mlp.experts.88.up_proj", "model.layers.65.mlp.experts.89.up_proj", "model.layers.65.mlp.experts.90.up_proj", "model.layers.65.mlp.experts.91.up_proj", "model.layers.65.mlp.experts.92.up_proj", "model.layers.65.mlp.experts.93.up_proj", "model.layers.65.mlp.experts.94.up_proj", "model.layers.65.mlp.experts.95.up_proj", "model.layers.65.mlp.experts.96.up_proj", "model.layers.65.mlp.experts.97.up_proj", "model.layers.65.mlp.experts.98.up_proj", "model.layers.65.mlp.experts.99.up_proj", "model.layers.65.mlp.experts.100.up_proj", "model.layers.65.mlp.experts.101.up_proj", "model.layers.65.mlp.experts.102.up_proj", "model.layers.65.mlp.experts.103.up_proj", "model.layers.65.mlp.experts.104.up_proj", "model.layers.65.mlp.experts.105.up_proj", "model.layers.65.mlp.experts.106.up_proj", "model.layers.65.mlp.experts.107.up_proj", "model.layers.65.mlp.experts.108.up_proj", "model.layers.65.mlp.experts.109.up_proj", "model.layers.65.mlp.experts.110.up_proj", "model.layers.65.mlp.experts.111.up_proj", "model.layers.65.mlp.experts.112.up_proj", "model.layers.65.mlp.experts.113.up_proj", "model.layers.65.mlp.experts.114.up_proj", "model.layers.65.mlp.experts.115.up_proj", "model.layers.65.mlp.experts.116.up_proj", "model.layers.65.mlp.experts.117.up_proj", "model.layers.65.mlp.experts.118.up_proj", "model.layers.65.mlp.experts.119.up_proj", "model.layers.65.mlp.experts.120.up_proj", "model.layers.65.mlp.experts.121.up_proj", "model.layers.65.mlp.experts.122.up_proj", "model.layers.65.mlp.experts.123.up_proj", "model.layers.65.mlp.experts.124.up_proj", "model.layers.65.mlp.experts.125.up_proj", "model.layers.65.mlp.experts.126.up_proj", "model.layers.65.mlp.experts.127.up_proj", "model.layers.65.mlp.experts.128.up_proj", "model.layers.65.mlp.experts.129.up_proj", "model.layers.65.mlp.experts.130.up_proj", "model.layers.65.mlp.experts.131.up_proj", "model.layers.65.mlp.experts.132.up_proj", "model.layers.65.mlp.experts.133.up_proj", "model.layers.65.mlp.experts.134.up_proj", "model.layers.65.mlp.experts.135.up_proj", "model.layers.65.mlp.experts.136.up_proj", "model.layers.65.mlp.experts.137.up_proj", "model.layers.65.mlp.experts.138.up_proj", "model.layers.65.mlp.experts.139.up_proj", "model.layers.65.mlp.experts.140.up_proj", "model.layers.65.mlp.experts.141.up_proj", "model.layers.65.mlp.experts.142.up_proj", "model.layers.65.mlp.experts.143.up_proj", "model.layers.65.mlp.experts.144.up_proj", "model.layers.65.mlp.experts.145.up_proj", "model.layers.65.mlp.experts.146.up_proj", "model.layers.65.mlp.experts.147.up_proj", "model.layers.65.mlp.experts.148.up_proj", "model.layers.65.mlp.experts.149.up_proj", "model.layers.65.mlp.experts.150.up_proj", "model.layers.65.mlp.experts.151.up_proj", "model.layers.65.mlp.experts.152.up_proj", "model.layers.65.mlp.experts.153.up_proj", "model.layers.65.mlp.experts.154.up_proj", "model.layers.65.mlp.experts.155.up_proj", "model.layers.65.mlp.experts.156.up_proj", "model.layers.65.mlp.experts.157.up_proj", "model.layers.65.mlp.experts.158.up_proj", "model.layers.65.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -1.348704099655984e-05, "dbits": 2516582400 }, { "dkld": -6.862804293633618e-05, "dbits": 5033164800 }, { "dkld": -6.80992379784584e-05, "dbits": 7549747200 }, { "dkld": -7.112650200724602e-05, "dbits": 12582912000 } ] }, { "idx": 392, "layers": [ "model.layers.65.mlp.experts.0.down_proj", "model.layers.65.mlp.experts.1.down_proj", "model.layers.65.mlp.experts.2.down_proj", "model.layers.65.mlp.experts.3.down_proj", "model.layers.65.mlp.experts.4.down_proj", "model.layers.65.mlp.experts.5.down_proj", "model.layers.65.mlp.experts.6.down_proj", "model.layers.65.mlp.experts.7.down_proj", "model.layers.65.mlp.experts.8.down_proj", "model.layers.65.mlp.experts.9.down_proj", "model.layers.65.mlp.experts.10.down_proj", "model.layers.65.mlp.experts.11.down_proj", "model.layers.65.mlp.experts.12.down_proj", "model.layers.65.mlp.experts.13.down_proj", "model.layers.65.mlp.experts.14.down_proj", "model.layers.65.mlp.experts.15.down_proj", "model.layers.65.mlp.experts.16.down_proj", "model.layers.65.mlp.experts.17.down_proj", "model.layers.65.mlp.experts.18.down_proj", "model.layers.65.mlp.experts.19.down_proj", "model.layers.65.mlp.experts.20.down_proj", "model.layers.65.mlp.experts.21.down_proj", "model.layers.65.mlp.experts.22.down_proj", "model.layers.65.mlp.experts.23.down_proj", "model.layers.65.mlp.experts.24.down_proj", "model.layers.65.mlp.experts.25.down_proj", "model.layers.65.mlp.experts.26.down_proj", "model.layers.65.mlp.experts.27.down_proj", "model.layers.65.mlp.experts.28.down_proj", "model.layers.65.mlp.experts.29.down_proj", "model.layers.65.mlp.experts.30.down_proj", "model.layers.65.mlp.experts.31.down_proj", "model.layers.65.mlp.experts.32.down_proj", "model.layers.65.mlp.experts.33.down_proj", "model.layers.65.mlp.experts.34.down_proj", "model.layers.65.mlp.experts.35.down_proj", "model.layers.65.mlp.experts.36.down_proj", "model.layers.65.mlp.experts.37.down_proj", "model.layers.65.mlp.experts.38.down_proj", "model.layers.65.mlp.experts.39.down_proj", "model.layers.65.mlp.experts.40.down_proj", "model.layers.65.mlp.experts.41.down_proj", "model.layers.65.mlp.experts.42.down_proj", "model.layers.65.mlp.experts.43.down_proj", "model.layers.65.mlp.experts.44.down_proj", "model.layers.65.mlp.experts.45.down_proj", "model.layers.65.mlp.experts.46.down_proj", "model.layers.65.mlp.experts.47.down_proj", "model.layers.65.mlp.experts.48.down_proj", "model.layers.65.mlp.experts.49.down_proj", "model.layers.65.mlp.experts.50.down_proj", "model.layers.65.mlp.experts.51.down_proj", "model.layers.65.mlp.experts.52.down_proj", "model.layers.65.mlp.experts.53.down_proj", "model.layers.65.mlp.experts.54.down_proj", "model.layers.65.mlp.experts.55.down_proj", "model.layers.65.mlp.experts.56.down_proj", "model.layers.65.mlp.experts.57.down_proj", "model.layers.65.mlp.experts.58.down_proj", "model.layers.65.mlp.experts.59.down_proj", "model.layers.65.mlp.experts.60.down_proj", "model.layers.65.mlp.experts.61.down_proj", "model.layers.65.mlp.experts.62.down_proj", "model.layers.65.mlp.experts.63.down_proj", "model.layers.65.mlp.experts.64.down_proj", "model.layers.65.mlp.experts.65.down_proj", "model.layers.65.mlp.experts.66.down_proj", "model.layers.65.mlp.experts.67.down_proj", "model.layers.65.mlp.experts.68.down_proj", "model.layers.65.mlp.experts.69.down_proj", "model.layers.65.mlp.experts.70.down_proj", "model.layers.65.mlp.experts.71.down_proj", "model.layers.65.mlp.experts.72.down_proj", "model.layers.65.mlp.experts.73.down_proj", "model.layers.65.mlp.experts.74.down_proj", "model.layers.65.mlp.experts.75.down_proj", "model.layers.65.mlp.experts.76.down_proj", "model.layers.65.mlp.experts.77.down_proj", "model.layers.65.mlp.experts.78.down_proj", "model.layers.65.mlp.experts.79.down_proj", "model.layers.65.mlp.experts.80.down_proj", "model.layers.65.mlp.experts.81.down_proj", "model.layers.65.mlp.experts.82.down_proj", "model.layers.65.mlp.experts.83.down_proj", "model.layers.65.mlp.experts.84.down_proj", "model.layers.65.mlp.experts.85.down_proj", "model.layers.65.mlp.experts.86.down_proj", "model.layers.65.mlp.experts.87.down_proj", "model.layers.65.mlp.experts.88.down_proj", "model.layers.65.mlp.experts.89.down_proj", "model.layers.65.mlp.experts.90.down_proj", "model.layers.65.mlp.experts.91.down_proj", "model.layers.65.mlp.experts.92.down_proj", "model.layers.65.mlp.experts.93.down_proj", "model.layers.65.mlp.experts.94.down_proj", "model.layers.65.mlp.experts.95.down_proj", "model.layers.65.mlp.experts.96.down_proj", "model.layers.65.mlp.experts.97.down_proj", "model.layers.65.mlp.experts.98.down_proj", "model.layers.65.mlp.experts.99.down_proj", "model.layers.65.mlp.experts.100.down_proj", "model.layers.65.mlp.experts.101.down_proj", "model.layers.65.mlp.experts.102.down_proj", "model.layers.65.mlp.experts.103.down_proj", "model.layers.65.mlp.experts.104.down_proj", "model.layers.65.mlp.experts.105.down_proj", "model.layers.65.mlp.experts.106.down_proj", "model.layers.65.mlp.experts.107.down_proj", "model.layers.65.mlp.experts.108.down_proj", "model.layers.65.mlp.experts.109.down_proj", "model.layers.65.mlp.experts.110.down_proj", "model.layers.65.mlp.experts.111.down_proj", "model.layers.65.mlp.experts.112.down_proj", "model.layers.65.mlp.experts.113.down_proj", "model.layers.65.mlp.experts.114.down_proj", "model.layers.65.mlp.experts.115.down_proj", "model.layers.65.mlp.experts.116.down_proj", "model.layers.65.mlp.experts.117.down_proj", "model.layers.65.mlp.experts.118.down_proj", "model.layers.65.mlp.experts.119.down_proj", "model.layers.65.mlp.experts.120.down_proj", "model.layers.65.mlp.experts.121.down_proj", "model.layers.65.mlp.experts.122.down_proj", "model.layers.65.mlp.experts.123.down_proj", "model.layers.65.mlp.experts.124.down_proj", "model.layers.65.mlp.experts.125.down_proj", "model.layers.65.mlp.experts.126.down_proj", "model.layers.65.mlp.experts.127.down_proj", "model.layers.65.mlp.experts.128.down_proj", "model.layers.65.mlp.experts.129.down_proj", "model.layers.65.mlp.experts.130.down_proj", "model.layers.65.mlp.experts.131.down_proj", "model.layers.65.mlp.experts.132.down_proj", "model.layers.65.mlp.experts.133.down_proj", "model.layers.65.mlp.experts.134.down_proj", "model.layers.65.mlp.experts.135.down_proj", "model.layers.65.mlp.experts.136.down_proj", "model.layers.65.mlp.experts.137.down_proj", "model.layers.65.mlp.experts.138.down_proj", "model.layers.65.mlp.experts.139.down_proj", "model.layers.65.mlp.experts.140.down_proj", "model.layers.65.mlp.experts.141.down_proj", "model.layers.65.mlp.experts.142.down_proj", "model.layers.65.mlp.experts.143.down_proj", "model.layers.65.mlp.experts.144.down_proj", "model.layers.65.mlp.experts.145.down_proj", "model.layers.65.mlp.experts.146.down_proj", "model.layers.65.mlp.experts.147.down_proj", "model.layers.65.mlp.experts.148.down_proj", "model.layers.65.mlp.experts.149.down_proj", "model.layers.65.mlp.experts.150.down_proj", "model.layers.65.mlp.experts.151.down_proj", "model.layers.65.mlp.experts.152.down_proj", "model.layers.65.mlp.experts.153.down_proj", "model.layers.65.mlp.experts.154.down_proj", "model.layers.65.mlp.experts.155.down_proj", "model.layers.65.mlp.experts.156.down_proj", "model.layers.65.mlp.experts.157.down_proj", "model.layers.65.mlp.experts.158.down_proj", "model.layers.65.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00010626101866365034, "dbits": 1258291200 }, { "dkld": -8.94891098141698e-05, "dbits": 2516582400 }, { "dkld": -0.00011762324720621109, "dbits": 3774873600 }, { "dkld": -0.00012416373938323177, "dbits": 6291456000 } ] }, { "idx": 393, "layers": [ "model.layers.66.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0005996043793857209, "dbits": 62914560 }, { "dkld": -0.0006185835227370262, "dbits": 125829120 }, { "dkld": -0.000612364150583744, "dbits": 188743680 }, { "dkld": -0.0006304371170699624, "dbits": 314572800 } ] }, { "idx": 394, "layers": [ "model.layers.66.self_attn.k_proj", "model.layers.66.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0002784778364002677, "dbits": 10485760 }, { "dkld": 0.0003503489308059188, "dbits": 20971520 }, { "dkld": 0.00029968991875647943, "dbits": 31457280 }, { "dkld": 0.0003326368518173639, "dbits": 52428800 } ] }, { "idx": 395, "layers": [ "model.layers.66.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0001565466634929208, "dbits": 62914560 }, { "dkld": -0.00019450103864074308, "dbits": 125829120 }, { "dkld": -0.00023939907550812878, "dbits": 188743680 }, { "dkld": -0.00026130517944694676, "dbits": 314572800 } ] }, { "idx": 396, "layers": [ "model.layers.66.mlp.shared_experts.gate_proj", "model.layers.66.mlp.shared_experts.up_proj", "model.layers.66.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00011599911376834748, "dbits": 23592960 }, { "dkld": -0.0001941711641848115, "dbits": 47185920 }, { "dkld": -0.00017445888370276608, "dbits": 70778880 }, { "dkld": -0.00013511339202523787, "dbits": 117964800 } ] }, { "idx": 397, "layers": [ "model.layers.66.mlp.experts.0.gate_proj", "model.layers.66.mlp.experts.1.gate_proj", "model.layers.66.mlp.experts.2.gate_proj", "model.layers.66.mlp.experts.3.gate_proj", "model.layers.66.mlp.experts.4.gate_proj", "model.layers.66.mlp.experts.5.gate_proj", "model.layers.66.mlp.experts.6.gate_proj", "model.layers.66.mlp.experts.7.gate_proj", "model.layers.66.mlp.experts.8.gate_proj", "model.layers.66.mlp.experts.9.gate_proj", "model.layers.66.mlp.experts.10.gate_proj", "model.layers.66.mlp.experts.11.gate_proj", "model.layers.66.mlp.experts.12.gate_proj", "model.layers.66.mlp.experts.13.gate_proj", "model.layers.66.mlp.experts.14.gate_proj", "model.layers.66.mlp.experts.15.gate_proj", "model.layers.66.mlp.experts.16.gate_proj", "model.layers.66.mlp.experts.17.gate_proj", "model.layers.66.mlp.experts.18.gate_proj", "model.layers.66.mlp.experts.19.gate_proj", "model.layers.66.mlp.experts.20.gate_proj", "model.layers.66.mlp.experts.21.gate_proj", "model.layers.66.mlp.experts.22.gate_proj", "model.layers.66.mlp.experts.23.gate_proj", "model.layers.66.mlp.experts.24.gate_proj", "model.layers.66.mlp.experts.25.gate_proj", "model.layers.66.mlp.experts.26.gate_proj", "model.layers.66.mlp.experts.27.gate_proj", "model.layers.66.mlp.experts.28.gate_proj", "model.layers.66.mlp.experts.29.gate_proj", "model.layers.66.mlp.experts.30.gate_proj", "model.layers.66.mlp.experts.31.gate_proj", "model.layers.66.mlp.experts.32.gate_proj", "model.layers.66.mlp.experts.33.gate_proj", "model.layers.66.mlp.experts.34.gate_proj", "model.layers.66.mlp.experts.35.gate_proj", "model.layers.66.mlp.experts.36.gate_proj", "model.layers.66.mlp.experts.37.gate_proj", "model.layers.66.mlp.experts.38.gate_proj", "model.layers.66.mlp.experts.39.gate_proj", "model.layers.66.mlp.experts.40.gate_proj", "model.layers.66.mlp.experts.41.gate_proj", "model.layers.66.mlp.experts.42.gate_proj", "model.layers.66.mlp.experts.43.gate_proj", "model.layers.66.mlp.experts.44.gate_proj", "model.layers.66.mlp.experts.45.gate_proj", "model.layers.66.mlp.experts.46.gate_proj", "model.layers.66.mlp.experts.47.gate_proj", "model.layers.66.mlp.experts.48.gate_proj", "model.layers.66.mlp.experts.49.gate_proj", "model.layers.66.mlp.experts.50.gate_proj", "model.layers.66.mlp.experts.51.gate_proj", "model.layers.66.mlp.experts.52.gate_proj", "model.layers.66.mlp.experts.53.gate_proj", "model.layers.66.mlp.experts.54.gate_proj", "model.layers.66.mlp.experts.55.gate_proj", "model.layers.66.mlp.experts.56.gate_proj", "model.layers.66.mlp.experts.57.gate_proj", "model.layers.66.mlp.experts.58.gate_proj", "model.layers.66.mlp.experts.59.gate_proj", "model.layers.66.mlp.experts.60.gate_proj", "model.layers.66.mlp.experts.61.gate_proj", "model.layers.66.mlp.experts.62.gate_proj", "model.layers.66.mlp.experts.63.gate_proj", "model.layers.66.mlp.experts.64.gate_proj", "model.layers.66.mlp.experts.65.gate_proj", "model.layers.66.mlp.experts.66.gate_proj", "model.layers.66.mlp.experts.67.gate_proj", "model.layers.66.mlp.experts.68.gate_proj", "model.layers.66.mlp.experts.69.gate_proj", "model.layers.66.mlp.experts.70.gate_proj", "model.layers.66.mlp.experts.71.gate_proj", "model.layers.66.mlp.experts.72.gate_proj", "model.layers.66.mlp.experts.73.gate_proj", "model.layers.66.mlp.experts.74.gate_proj", "model.layers.66.mlp.experts.75.gate_proj", "model.layers.66.mlp.experts.76.gate_proj", "model.layers.66.mlp.experts.77.gate_proj", "model.layers.66.mlp.experts.78.gate_proj", "model.layers.66.mlp.experts.79.gate_proj", "model.layers.66.mlp.experts.80.gate_proj", "model.layers.66.mlp.experts.81.gate_proj", "model.layers.66.mlp.experts.82.gate_proj", "model.layers.66.mlp.experts.83.gate_proj", "model.layers.66.mlp.experts.84.gate_proj", "model.layers.66.mlp.experts.85.gate_proj", "model.layers.66.mlp.experts.86.gate_proj", "model.layers.66.mlp.experts.87.gate_proj", "model.layers.66.mlp.experts.88.gate_proj", "model.layers.66.mlp.experts.89.gate_proj", "model.layers.66.mlp.experts.90.gate_proj", "model.layers.66.mlp.experts.91.gate_proj", "model.layers.66.mlp.experts.92.gate_proj", "model.layers.66.mlp.experts.93.gate_proj", "model.layers.66.mlp.experts.94.gate_proj", "model.layers.66.mlp.experts.95.gate_proj", "model.layers.66.mlp.experts.96.gate_proj", "model.layers.66.mlp.experts.97.gate_proj", "model.layers.66.mlp.experts.98.gate_proj", "model.layers.66.mlp.experts.99.gate_proj", "model.layers.66.mlp.experts.100.gate_proj", "model.layers.66.mlp.experts.101.gate_proj", "model.layers.66.mlp.experts.102.gate_proj", "model.layers.66.mlp.experts.103.gate_proj", "model.layers.66.mlp.experts.104.gate_proj", "model.layers.66.mlp.experts.105.gate_proj", "model.layers.66.mlp.experts.106.gate_proj", "model.layers.66.mlp.experts.107.gate_proj", "model.layers.66.mlp.experts.108.gate_proj", "model.layers.66.mlp.experts.109.gate_proj", "model.layers.66.mlp.experts.110.gate_proj", "model.layers.66.mlp.experts.111.gate_proj", "model.layers.66.mlp.experts.112.gate_proj", "model.layers.66.mlp.experts.113.gate_proj", "model.layers.66.mlp.experts.114.gate_proj", "model.layers.66.mlp.experts.115.gate_proj", "model.layers.66.mlp.experts.116.gate_proj", "model.layers.66.mlp.experts.117.gate_proj", "model.layers.66.mlp.experts.118.gate_proj", "model.layers.66.mlp.experts.119.gate_proj", "model.layers.66.mlp.experts.120.gate_proj", "model.layers.66.mlp.experts.121.gate_proj", "model.layers.66.mlp.experts.122.gate_proj", "model.layers.66.mlp.experts.123.gate_proj", "model.layers.66.mlp.experts.124.gate_proj", "model.layers.66.mlp.experts.125.gate_proj", "model.layers.66.mlp.experts.126.gate_proj", "model.layers.66.mlp.experts.127.gate_proj", "model.layers.66.mlp.experts.128.gate_proj", "model.layers.66.mlp.experts.129.gate_proj", "model.layers.66.mlp.experts.130.gate_proj", "model.layers.66.mlp.experts.131.gate_proj", "model.layers.66.mlp.experts.132.gate_proj", "model.layers.66.mlp.experts.133.gate_proj", "model.layers.66.mlp.experts.134.gate_proj", "model.layers.66.mlp.experts.135.gate_proj", "model.layers.66.mlp.experts.136.gate_proj", "model.layers.66.mlp.experts.137.gate_proj", "model.layers.66.mlp.experts.138.gate_proj", "model.layers.66.mlp.experts.139.gate_proj", "model.layers.66.mlp.experts.140.gate_proj", "model.layers.66.mlp.experts.141.gate_proj", "model.layers.66.mlp.experts.142.gate_proj", "model.layers.66.mlp.experts.143.gate_proj", "model.layers.66.mlp.experts.144.gate_proj", "model.layers.66.mlp.experts.145.gate_proj", "model.layers.66.mlp.experts.146.gate_proj", "model.layers.66.mlp.experts.147.gate_proj", "model.layers.66.mlp.experts.148.gate_proj", "model.layers.66.mlp.experts.149.gate_proj", "model.layers.66.mlp.experts.150.gate_proj", "model.layers.66.mlp.experts.151.gate_proj", "model.layers.66.mlp.experts.152.gate_proj", "model.layers.66.mlp.experts.153.gate_proj", "model.layers.66.mlp.experts.154.gate_proj", "model.layers.66.mlp.experts.155.gate_proj", "model.layers.66.mlp.experts.156.gate_proj", "model.layers.66.mlp.experts.157.gate_proj", "model.layers.66.mlp.experts.158.gate_proj", "model.layers.66.mlp.experts.159.gate_proj", "model.layers.66.mlp.experts.0.up_proj", "model.layers.66.mlp.experts.1.up_proj", "model.layers.66.mlp.experts.2.up_proj", "model.layers.66.mlp.experts.3.up_proj", "model.layers.66.mlp.experts.4.up_proj", "model.layers.66.mlp.experts.5.up_proj", "model.layers.66.mlp.experts.6.up_proj", "model.layers.66.mlp.experts.7.up_proj", "model.layers.66.mlp.experts.8.up_proj", "model.layers.66.mlp.experts.9.up_proj", "model.layers.66.mlp.experts.10.up_proj", "model.layers.66.mlp.experts.11.up_proj", "model.layers.66.mlp.experts.12.up_proj", "model.layers.66.mlp.experts.13.up_proj", "model.layers.66.mlp.experts.14.up_proj", "model.layers.66.mlp.experts.15.up_proj", "model.layers.66.mlp.experts.16.up_proj", "model.layers.66.mlp.experts.17.up_proj", "model.layers.66.mlp.experts.18.up_proj", "model.layers.66.mlp.experts.19.up_proj", "model.layers.66.mlp.experts.20.up_proj", "model.layers.66.mlp.experts.21.up_proj", "model.layers.66.mlp.experts.22.up_proj", "model.layers.66.mlp.experts.23.up_proj", "model.layers.66.mlp.experts.24.up_proj", "model.layers.66.mlp.experts.25.up_proj", "model.layers.66.mlp.experts.26.up_proj", "model.layers.66.mlp.experts.27.up_proj", "model.layers.66.mlp.experts.28.up_proj", "model.layers.66.mlp.experts.29.up_proj", "model.layers.66.mlp.experts.30.up_proj", "model.layers.66.mlp.experts.31.up_proj", "model.layers.66.mlp.experts.32.up_proj", "model.layers.66.mlp.experts.33.up_proj", "model.layers.66.mlp.experts.34.up_proj", "model.layers.66.mlp.experts.35.up_proj", "model.layers.66.mlp.experts.36.up_proj", "model.layers.66.mlp.experts.37.up_proj", "model.layers.66.mlp.experts.38.up_proj", "model.layers.66.mlp.experts.39.up_proj", "model.layers.66.mlp.experts.40.up_proj", "model.layers.66.mlp.experts.41.up_proj", "model.layers.66.mlp.experts.42.up_proj", "model.layers.66.mlp.experts.43.up_proj", "model.layers.66.mlp.experts.44.up_proj", "model.layers.66.mlp.experts.45.up_proj", "model.layers.66.mlp.experts.46.up_proj", "model.layers.66.mlp.experts.47.up_proj", "model.layers.66.mlp.experts.48.up_proj", "model.layers.66.mlp.experts.49.up_proj", "model.layers.66.mlp.experts.50.up_proj", "model.layers.66.mlp.experts.51.up_proj", "model.layers.66.mlp.experts.52.up_proj", "model.layers.66.mlp.experts.53.up_proj", "model.layers.66.mlp.experts.54.up_proj", "model.layers.66.mlp.experts.55.up_proj", "model.layers.66.mlp.experts.56.up_proj", "model.layers.66.mlp.experts.57.up_proj", "model.layers.66.mlp.experts.58.up_proj", "model.layers.66.mlp.experts.59.up_proj", "model.layers.66.mlp.experts.60.up_proj", "model.layers.66.mlp.experts.61.up_proj", "model.layers.66.mlp.experts.62.up_proj", "model.layers.66.mlp.experts.63.up_proj", "model.layers.66.mlp.experts.64.up_proj", "model.layers.66.mlp.experts.65.up_proj", "model.layers.66.mlp.experts.66.up_proj", "model.layers.66.mlp.experts.67.up_proj", "model.layers.66.mlp.experts.68.up_proj", "model.layers.66.mlp.experts.69.up_proj", "model.layers.66.mlp.experts.70.up_proj", "model.layers.66.mlp.experts.71.up_proj", "model.layers.66.mlp.experts.72.up_proj", "model.layers.66.mlp.experts.73.up_proj", "model.layers.66.mlp.experts.74.up_proj", "model.layers.66.mlp.experts.75.up_proj", "model.layers.66.mlp.experts.76.up_proj", "model.layers.66.mlp.experts.77.up_proj", "model.layers.66.mlp.experts.78.up_proj", "model.layers.66.mlp.experts.79.up_proj", "model.layers.66.mlp.experts.80.up_proj", "model.layers.66.mlp.experts.81.up_proj", "model.layers.66.mlp.experts.82.up_proj", "model.layers.66.mlp.experts.83.up_proj", "model.layers.66.mlp.experts.84.up_proj", "model.layers.66.mlp.experts.85.up_proj", "model.layers.66.mlp.experts.86.up_proj", "model.layers.66.mlp.experts.87.up_proj", "model.layers.66.mlp.experts.88.up_proj", "model.layers.66.mlp.experts.89.up_proj", "model.layers.66.mlp.experts.90.up_proj", "model.layers.66.mlp.experts.91.up_proj", "model.layers.66.mlp.experts.92.up_proj", "model.layers.66.mlp.experts.93.up_proj", "model.layers.66.mlp.experts.94.up_proj", "model.layers.66.mlp.experts.95.up_proj", "model.layers.66.mlp.experts.96.up_proj", "model.layers.66.mlp.experts.97.up_proj", "model.layers.66.mlp.experts.98.up_proj", "model.layers.66.mlp.experts.99.up_proj", "model.layers.66.mlp.experts.100.up_proj", "model.layers.66.mlp.experts.101.up_proj", "model.layers.66.mlp.experts.102.up_proj", "model.layers.66.mlp.experts.103.up_proj", "model.layers.66.mlp.experts.104.up_proj", "model.layers.66.mlp.experts.105.up_proj", "model.layers.66.mlp.experts.106.up_proj", "model.layers.66.mlp.experts.107.up_proj", "model.layers.66.mlp.experts.108.up_proj", "model.layers.66.mlp.experts.109.up_proj", "model.layers.66.mlp.experts.110.up_proj", "model.layers.66.mlp.experts.111.up_proj", "model.layers.66.mlp.experts.112.up_proj", "model.layers.66.mlp.experts.113.up_proj", "model.layers.66.mlp.experts.114.up_proj", "model.layers.66.mlp.experts.115.up_proj", "model.layers.66.mlp.experts.116.up_proj", "model.layers.66.mlp.experts.117.up_proj", "model.layers.66.mlp.experts.118.up_proj", "model.layers.66.mlp.experts.119.up_proj", "model.layers.66.mlp.experts.120.up_proj", "model.layers.66.mlp.experts.121.up_proj", "model.layers.66.mlp.experts.122.up_proj", "model.layers.66.mlp.experts.123.up_proj", "model.layers.66.mlp.experts.124.up_proj", "model.layers.66.mlp.experts.125.up_proj", "model.layers.66.mlp.experts.126.up_proj", "model.layers.66.mlp.experts.127.up_proj", "model.layers.66.mlp.experts.128.up_proj", "model.layers.66.mlp.experts.129.up_proj", "model.layers.66.mlp.experts.130.up_proj", "model.layers.66.mlp.experts.131.up_proj", "model.layers.66.mlp.experts.132.up_proj", "model.layers.66.mlp.experts.133.up_proj", "model.layers.66.mlp.experts.134.up_proj", "model.layers.66.mlp.experts.135.up_proj", "model.layers.66.mlp.experts.136.up_proj", "model.layers.66.mlp.experts.137.up_proj", "model.layers.66.mlp.experts.138.up_proj", "model.layers.66.mlp.experts.139.up_proj", "model.layers.66.mlp.experts.140.up_proj", "model.layers.66.mlp.experts.141.up_proj", "model.layers.66.mlp.experts.142.up_proj", "model.layers.66.mlp.experts.143.up_proj", "model.layers.66.mlp.experts.144.up_proj", "model.layers.66.mlp.experts.145.up_proj", "model.layers.66.mlp.experts.146.up_proj", "model.layers.66.mlp.experts.147.up_proj", "model.layers.66.mlp.experts.148.up_proj", "model.layers.66.mlp.experts.149.up_proj", "model.layers.66.mlp.experts.150.up_proj", "model.layers.66.mlp.experts.151.up_proj", "model.layers.66.mlp.experts.152.up_proj", "model.layers.66.mlp.experts.153.up_proj", "model.layers.66.mlp.experts.154.up_proj", "model.layers.66.mlp.experts.155.up_proj", "model.layers.66.mlp.experts.156.up_proj", "model.layers.66.mlp.experts.157.up_proj", "model.layers.66.mlp.experts.158.up_proj", "model.layers.66.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00021883528679610131, "dbits": 2516582400 }, { "dkld": -0.0002022096887230873, "dbits": 5033164800 }, { "dkld": -0.00021421080455184816, "dbits": 7549747200 }, { "dkld": -0.00019587744027377285, "dbits": 12582912000 } ] }, { "idx": 398, "layers": [ "model.layers.66.mlp.experts.0.down_proj", "model.layers.66.mlp.experts.1.down_proj", "model.layers.66.mlp.experts.2.down_proj", "model.layers.66.mlp.experts.3.down_proj", "model.layers.66.mlp.experts.4.down_proj", "model.layers.66.mlp.experts.5.down_proj", "model.layers.66.mlp.experts.6.down_proj", "model.layers.66.mlp.experts.7.down_proj", "model.layers.66.mlp.experts.8.down_proj", "model.layers.66.mlp.experts.9.down_proj", "model.layers.66.mlp.experts.10.down_proj", "model.layers.66.mlp.experts.11.down_proj", "model.layers.66.mlp.experts.12.down_proj", "model.layers.66.mlp.experts.13.down_proj", "model.layers.66.mlp.experts.14.down_proj", "model.layers.66.mlp.experts.15.down_proj", "model.layers.66.mlp.experts.16.down_proj", "model.layers.66.mlp.experts.17.down_proj", "model.layers.66.mlp.experts.18.down_proj", "model.layers.66.mlp.experts.19.down_proj", "model.layers.66.mlp.experts.20.down_proj", "model.layers.66.mlp.experts.21.down_proj", "model.layers.66.mlp.experts.22.down_proj", "model.layers.66.mlp.experts.23.down_proj", "model.layers.66.mlp.experts.24.down_proj", "model.layers.66.mlp.experts.25.down_proj", "model.layers.66.mlp.experts.26.down_proj", "model.layers.66.mlp.experts.27.down_proj", "model.layers.66.mlp.experts.28.down_proj", "model.layers.66.mlp.experts.29.down_proj", "model.layers.66.mlp.experts.30.down_proj", "model.layers.66.mlp.experts.31.down_proj", "model.layers.66.mlp.experts.32.down_proj", "model.layers.66.mlp.experts.33.down_proj", "model.layers.66.mlp.experts.34.down_proj", "model.layers.66.mlp.experts.35.down_proj", "model.layers.66.mlp.experts.36.down_proj", "model.layers.66.mlp.experts.37.down_proj", "model.layers.66.mlp.experts.38.down_proj", "model.layers.66.mlp.experts.39.down_proj", "model.layers.66.mlp.experts.40.down_proj", "model.layers.66.mlp.experts.41.down_proj", "model.layers.66.mlp.experts.42.down_proj", "model.layers.66.mlp.experts.43.down_proj", "model.layers.66.mlp.experts.44.down_proj", "model.layers.66.mlp.experts.45.down_proj", "model.layers.66.mlp.experts.46.down_proj", "model.layers.66.mlp.experts.47.down_proj", "model.layers.66.mlp.experts.48.down_proj", "model.layers.66.mlp.experts.49.down_proj", "model.layers.66.mlp.experts.50.down_proj", "model.layers.66.mlp.experts.51.down_proj", "model.layers.66.mlp.experts.52.down_proj", "model.layers.66.mlp.experts.53.down_proj", "model.layers.66.mlp.experts.54.down_proj", "model.layers.66.mlp.experts.55.down_proj", "model.layers.66.mlp.experts.56.down_proj", "model.layers.66.mlp.experts.57.down_proj", "model.layers.66.mlp.experts.58.down_proj", "model.layers.66.mlp.experts.59.down_proj", "model.layers.66.mlp.experts.60.down_proj", "model.layers.66.mlp.experts.61.down_proj", "model.layers.66.mlp.experts.62.down_proj", "model.layers.66.mlp.experts.63.down_proj", "model.layers.66.mlp.experts.64.down_proj", "model.layers.66.mlp.experts.65.down_proj", "model.layers.66.mlp.experts.66.down_proj", "model.layers.66.mlp.experts.67.down_proj", "model.layers.66.mlp.experts.68.down_proj", "model.layers.66.mlp.experts.69.down_proj", "model.layers.66.mlp.experts.70.down_proj", "model.layers.66.mlp.experts.71.down_proj", "model.layers.66.mlp.experts.72.down_proj", "model.layers.66.mlp.experts.73.down_proj", "model.layers.66.mlp.experts.74.down_proj", "model.layers.66.mlp.experts.75.down_proj", "model.layers.66.mlp.experts.76.down_proj", "model.layers.66.mlp.experts.77.down_proj", "model.layers.66.mlp.experts.78.down_proj", "model.layers.66.mlp.experts.79.down_proj", "model.layers.66.mlp.experts.80.down_proj", "model.layers.66.mlp.experts.81.down_proj", "model.layers.66.mlp.experts.82.down_proj", "model.layers.66.mlp.experts.83.down_proj", "model.layers.66.mlp.experts.84.down_proj", "model.layers.66.mlp.experts.85.down_proj", "model.layers.66.mlp.experts.86.down_proj", "model.layers.66.mlp.experts.87.down_proj", "model.layers.66.mlp.experts.88.down_proj", "model.layers.66.mlp.experts.89.down_proj", "model.layers.66.mlp.experts.90.down_proj", "model.layers.66.mlp.experts.91.down_proj", "model.layers.66.mlp.experts.92.down_proj", "model.layers.66.mlp.experts.93.down_proj", "model.layers.66.mlp.experts.94.down_proj", "model.layers.66.mlp.experts.95.down_proj", "model.layers.66.mlp.experts.96.down_proj", "model.layers.66.mlp.experts.97.down_proj", "model.layers.66.mlp.experts.98.down_proj", "model.layers.66.mlp.experts.99.down_proj", "model.layers.66.mlp.experts.100.down_proj", "model.layers.66.mlp.experts.101.down_proj", "model.layers.66.mlp.experts.102.down_proj", "model.layers.66.mlp.experts.103.down_proj", "model.layers.66.mlp.experts.104.down_proj", "model.layers.66.mlp.experts.105.down_proj", "model.layers.66.mlp.experts.106.down_proj", "model.layers.66.mlp.experts.107.down_proj", "model.layers.66.mlp.experts.108.down_proj", "model.layers.66.mlp.experts.109.down_proj", "model.layers.66.mlp.experts.110.down_proj", "model.layers.66.mlp.experts.111.down_proj", "model.layers.66.mlp.experts.112.down_proj", "model.layers.66.mlp.experts.113.down_proj", "model.layers.66.mlp.experts.114.down_proj", "model.layers.66.mlp.experts.115.down_proj", "model.layers.66.mlp.experts.116.down_proj", "model.layers.66.mlp.experts.117.down_proj", "model.layers.66.mlp.experts.118.down_proj", "model.layers.66.mlp.experts.119.down_proj", "model.layers.66.mlp.experts.120.down_proj", "model.layers.66.mlp.experts.121.down_proj", "model.layers.66.mlp.experts.122.down_proj", "model.layers.66.mlp.experts.123.down_proj", "model.layers.66.mlp.experts.124.down_proj", "model.layers.66.mlp.experts.125.down_proj", "model.layers.66.mlp.experts.126.down_proj", "model.layers.66.mlp.experts.127.down_proj", "model.layers.66.mlp.experts.128.down_proj", "model.layers.66.mlp.experts.129.down_proj", "model.layers.66.mlp.experts.130.down_proj", "model.layers.66.mlp.experts.131.down_proj", "model.layers.66.mlp.experts.132.down_proj", "model.layers.66.mlp.experts.133.down_proj", "model.layers.66.mlp.experts.134.down_proj", "model.layers.66.mlp.experts.135.down_proj", "model.layers.66.mlp.experts.136.down_proj", "model.layers.66.mlp.experts.137.down_proj", "model.layers.66.mlp.experts.138.down_proj", "model.layers.66.mlp.experts.139.down_proj", "model.layers.66.mlp.experts.140.down_proj", "model.layers.66.mlp.experts.141.down_proj", "model.layers.66.mlp.experts.142.down_proj", "model.layers.66.mlp.experts.143.down_proj", "model.layers.66.mlp.experts.144.down_proj", "model.layers.66.mlp.experts.145.down_proj", "model.layers.66.mlp.experts.146.down_proj", "model.layers.66.mlp.experts.147.down_proj", "model.layers.66.mlp.experts.148.down_proj", "model.layers.66.mlp.experts.149.down_proj", "model.layers.66.mlp.experts.150.down_proj", "model.layers.66.mlp.experts.151.down_proj", "model.layers.66.mlp.experts.152.down_proj", "model.layers.66.mlp.experts.153.down_proj", "model.layers.66.mlp.experts.154.down_proj", "model.layers.66.mlp.experts.155.down_proj", "model.layers.66.mlp.experts.156.down_proj", "model.layers.66.mlp.experts.157.down_proj", "model.layers.66.mlp.experts.158.down_proj", "model.layers.66.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0002089120447635734, "dbits": 1258291200 }, { "dkld": -0.00023786928504705707, "dbits": 2516582400 }, { "dkld": -0.0002443930134177208, "dbits": 3774873600 }, { "dkld": -0.00026262896135449687, "dbits": 6291456000 } ] }, { "idx": 399, "layers": [ "model.layers.67.self_attn.q_proj" ], "candidates": [ { "dkld": -2.1053943783044815e-05, "dbits": 62914560 }, { "dkld": -5.9612654149540645e-05, "dbits": 125829120 }, { "dkld": -9.084800258279957e-05, "dbits": 188743680 }, { "dkld": -3.761900588870881e-05, "dbits": 314572800 } ] }, { "idx": 400, "layers": [ "model.layers.67.self_attn.k_proj", "model.layers.67.self_attn.v_proj" ], "candidates": [ { "dkld": -0.000281264446675783, "dbits": 10485760 }, { "dkld": -0.0001909993588924408, "dbits": 20971520 }, { "dkld": -0.00021434668451547623, "dbits": 31457280 }, { "dkld": -0.00020231604576111117, "dbits": 52428800 } ] }, { "idx": 401, "layers": [ "model.layers.67.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00010914234444499293, "dbits": 62914560 }, { "dkld": -9.056245908141414e-05, "dbits": 125829120 }, { "dkld": -0.00017785867676139988, "dbits": 188743680 }, { "dkld": -0.00015001222491264898, "dbits": 314572800 } ] }, { "idx": 402, "layers": [ "model.layers.67.mlp.shared_experts.gate_proj", "model.layers.67.mlp.shared_experts.up_proj", "model.layers.67.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -1.3995170593264494e-05, "dbits": 23592960 }, { "dkld": -7.55310058593861e-05, "dbits": 47185920 }, { "dkld": -4.8873201012614165e-05, "dbits": 70778880 }, { "dkld": -2.064686268568594e-05, "dbits": 117964800 } ] }, { "idx": 403, "layers": [ "model.layers.67.mlp.experts.0.gate_proj", "model.layers.67.mlp.experts.1.gate_proj", "model.layers.67.mlp.experts.2.gate_proj", "model.layers.67.mlp.experts.3.gate_proj", "model.layers.67.mlp.experts.4.gate_proj", "model.layers.67.mlp.experts.5.gate_proj", "model.layers.67.mlp.experts.6.gate_proj", "model.layers.67.mlp.experts.7.gate_proj", "model.layers.67.mlp.experts.8.gate_proj", "model.layers.67.mlp.experts.9.gate_proj", "model.layers.67.mlp.experts.10.gate_proj", "model.layers.67.mlp.experts.11.gate_proj", "model.layers.67.mlp.experts.12.gate_proj", "model.layers.67.mlp.experts.13.gate_proj", "model.layers.67.mlp.experts.14.gate_proj", "model.layers.67.mlp.experts.15.gate_proj", "model.layers.67.mlp.experts.16.gate_proj", "model.layers.67.mlp.experts.17.gate_proj", "model.layers.67.mlp.experts.18.gate_proj", "model.layers.67.mlp.experts.19.gate_proj", "model.layers.67.mlp.experts.20.gate_proj", "model.layers.67.mlp.experts.21.gate_proj", "model.layers.67.mlp.experts.22.gate_proj", "model.layers.67.mlp.experts.23.gate_proj", "model.layers.67.mlp.experts.24.gate_proj", "model.layers.67.mlp.experts.25.gate_proj", "model.layers.67.mlp.experts.26.gate_proj", "model.layers.67.mlp.experts.27.gate_proj", "model.layers.67.mlp.experts.28.gate_proj", "model.layers.67.mlp.experts.29.gate_proj", "model.layers.67.mlp.experts.30.gate_proj", "model.layers.67.mlp.experts.31.gate_proj", "model.layers.67.mlp.experts.32.gate_proj", "model.layers.67.mlp.experts.33.gate_proj", "model.layers.67.mlp.experts.34.gate_proj", "model.layers.67.mlp.experts.35.gate_proj", "model.layers.67.mlp.experts.36.gate_proj", "model.layers.67.mlp.experts.37.gate_proj", "model.layers.67.mlp.experts.38.gate_proj", "model.layers.67.mlp.experts.39.gate_proj", "model.layers.67.mlp.experts.40.gate_proj", "model.layers.67.mlp.experts.41.gate_proj", "model.layers.67.mlp.experts.42.gate_proj", "model.layers.67.mlp.experts.43.gate_proj", "model.layers.67.mlp.experts.44.gate_proj", "model.layers.67.mlp.experts.45.gate_proj", "model.layers.67.mlp.experts.46.gate_proj", "model.layers.67.mlp.experts.47.gate_proj", "model.layers.67.mlp.experts.48.gate_proj", "model.layers.67.mlp.experts.49.gate_proj", "model.layers.67.mlp.experts.50.gate_proj", "model.layers.67.mlp.experts.51.gate_proj", "model.layers.67.mlp.experts.52.gate_proj", "model.layers.67.mlp.experts.53.gate_proj", "model.layers.67.mlp.experts.54.gate_proj", "model.layers.67.mlp.experts.55.gate_proj", "model.layers.67.mlp.experts.56.gate_proj", "model.layers.67.mlp.experts.57.gate_proj", "model.layers.67.mlp.experts.58.gate_proj", "model.layers.67.mlp.experts.59.gate_proj", "model.layers.67.mlp.experts.60.gate_proj", "model.layers.67.mlp.experts.61.gate_proj", "model.layers.67.mlp.experts.62.gate_proj", "model.layers.67.mlp.experts.63.gate_proj", "model.layers.67.mlp.experts.64.gate_proj", "model.layers.67.mlp.experts.65.gate_proj", "model.layers.67.mlp.experts.66.gate_proj", "model.layers.67.mlp.experts.67.gate_proj", "model.layers.67.mlp.experts.68.gate_proj", "model.layers.67.mlp.experts.69.gate_proj", "model.layers.67.mlp.experts.70.gate_proj", "model.layers.67.mlp.experts.71.gate_proj", "model.layers.67.mlp.experts.72.gate_proj", "model.layers.67.mlp.experts.73.gate_proj", "model.layers.67.mlp.experts.74.gate_proj", "model.layers.67.mlp.experts.75.gate_proj", "model.layers.67.mlp.experts.76.gate_proj", "model.layers.67.mlp.experts.77.gate_proj", "model.layers.67.mlp.experts.78.gate_proj", "model.layers.67.mlp.experts.79.gate_proj", "model.layers.67.mlp.experts.80.gate_proj", "model.layers.67.mlp.experts.81.gate_proj", "model.layers.67.mlp.experts.82.gate_proj", "model.layers.67.mlp.experts.83.gate_proj", "model.layers.67.mlp.experts.84.gate_proj", "model.layers.67.mlp.experts.85.gate_proj", "model.layers.67.mlp.experts.86.gate_proj", "model.layers.67.mlp.experts.87.gate_proj", "model.layers.67.mlp.experts.88.gate_proj", "model.layers.67.mlp.experts.89.gate_proj", "model.layers.67.mlp.experts.90.gate_proj", "model.layers.67.mlp.experts.91.gate_proj", "model.layers.67.mlp.experts.92.gate_proj", "model.layers.67.mlp.experts.93.gate_proj", "model.layers.67.mlp.experts.94.gate_proj", "model.layers.67.mlp.experts.95.gate_proj", "model.layers.67.mlp.experts.96.gate_proj", "model.layers.67.mlp.experts.97.gate_proj", "model.layers.67.mlp.experts.98.gate_proj", "model.layers.67.mlp.experts.99.gate_proj", "model.layers.67.mlp.experts.100.gate_proj", "model.layers.67.mlp.experts.101.gate_proj", "model.layers.67.mlp.experts.102.gate_proj", "model.layers.67.mlp.experts.103.gate_proj", "model.layers.67.mlp.experts.104.gate_proj", "model.layers.67.mlp.experts.105.gate_proj", "model.layers.67.mlp.experts.106.gate_proj", "model.layers.67.mlp.experts.107.gate_proj", "model.layers.67.mlp.experts.108.gate_proj", "model.layers.67.mlp.experts.109.gate_proj", "model.layers.67.mlp.experts.110.gate_proj", "model.layers.67.mlp.experts.111.gate_proj", "model.layers.67.mlp.experts.112.gate_proj", "model.layers.67.mlp.experts.113.gate_proj", "model.layers.67.mlp.experts.114.gate_proj", "model.layers.67.mlp.experts.115.gate_proj", "model.layers.67.mlp.experts.116.gate_proj", "model.layers.67.mlp.experts.117.gate_proj", "model.layers.67.mlp.experts.118.gate_proj", "model.layers.67.mlp.experts.119.gate_proj", "model.layers.67.mlp.experts.120.gate_proj", "model.layers.67.mlp.experts.121.gate_proj", "model.layers.67.mlp.experts.122.gate_proj", "model.layers.67.mlp.experts.123.gate_proj", "model.layers.67.mlp.experts.124.gate_proj", "model.layers.67.mlp.experts.125.gate_proj", "model.layers.67.mlp.experts.126.gate_proj", "model.layers.67.mlp.experts.127.gate_proj", "model.layers.67.mlp.experts.128.gate_proj", "model.layers.67.mlp.experts.129.gate_proj", "model.layers.67.mlp.experts.130.gate_proj", "model.layers.67.mlp.experts.131.gate_proj", "model.layers.67.mlp.experts.132.gate_proj", "model.layers.67.mlp.experts.133.gate_proj", "model.layers.67.mlp.experts.134.gate_proj", "model.layers.67.mlp.experts.135.gate_proj", "model.layers.67.mlp.experts.136.gate_proj", "model.layers.67.mlp.experts.137.gate_proj", "model.layers.67.mlp.experts.138.gate_proj", "model.layers.67.mlp.experts.139.gate_proj", "model.layers.67.mlp.experts.140.gate_proj", "model.layers.67.mlp.experts.141.gate_proj", "model.layers.67.mlp.experts.142.gate_proj", "model.layers.67.mlp.experts.143.gate_proj", "model.layers.67.mlp.experts.144.gate_proj", "model.layers.67.mlp.experts.145.gate_proj", "model.layers.67.mlp.experts.146.gate_proj", "model.layers.67.mlp.experts.147.gate_proj", "model.layers.67.mlp.experts.148.gate_proj", "model.layers.67.mlp.experts.149.gate_proj", "model.layers.67.mlp.experts.150.gate_proj", "model.layers.67.mlp.experts.151.gate_proj", "model.layers.67.mlp.experts.152.gate_proj", "model.layers.67.mlp.experts.153.gate_proj", "model.layers.67.mlp.experts.154.gate_proj", "model.layers.67.mlp.experts.155.gate_proj", "model.layers.67.mlp.experts.156.gate_proj", "model.layers.67.mlp.experts.157.gate_proj", "model.layers.67.mlp.experts.158.gate_proj", "model.layers.67.mlp.experts.159.gate_proj", "model.layers.67.mlp.experts.0.up_proj", "model.layers.67.mlp.experts.1.up_proj", "model.layers.67.mlp.experts.2.up_proj", "model.layers.67.mlp.experts.3.up_proj", "model.layers.67.mlp.experts.4.up_proj", "model.layers.67.mlp.experts.5.up_proj", "model.layers.67.mlp.experts.6.up_proj", "model.layers.67.mlp.experts.7.up_proj", "model.layers.67.mlp.experts.8.up_proj", "model.layers.67.mlp.experts.9.up_proj", "model.layers.67.mlp.experts.10.up_proj", "model.layers.67.mlp.experts.11.up_proj", "model.layers.67.mlp.experts.12.up_proj", "model.layers.67.mlp.experts.13.up_proj", "model.layers.67.mlp.experts.14.up_proj", "model.layers.67.mlp.experts.15.up_proj", "model.layers.67.mlp.experts.16.up_proj", "model.layers.67.mlp.experts.17.up_proj", "model.layers.67.mlp.experts.18.up_proj", "model.layers.67.mlp.experts.19.up_proj", "model.layers.67.mlp.experts.20.up_proj", "model.layers.67.mlp.experts.21.up_proj", "model.layers.67.mlp.experts.22.up_proj", "model.layers.67.mlp.experts.23.up_proj", "model.layers.67.mlp.experts.24.up_proj", "model.layers.67.mlp.experts.25.up_proj", "model.layers.67.mlp.experts.26.up_proj", "model.layers.67.mlp.experts.27.up_proj", "model.layers.67.mlp.experts.28.up_proj", "model.layers.67.mlp.experts.29.up_proj", "model.layers.67.mlp.experts.30.up_proj", "model.layers.67.mlp.experts.31.up_proj", "model.layers.67.mlp.experts.32.up_proj", "model.layers.67.mlp.experts.33.up_proj", "model.layers.67.mlp.experts.34.up_proj", "model.layers.67.mlp.experts.35.up_proj", "model.layers.67.mlp.experts.36.up_proj", "model.layers.67.mlp.experts.37.up_proj", "model.layers.67.mlp.experts.38.up_proj", "model.layers.67.mlp.experts.39.up_proj", "model.layers.67.mlp.experts.40.up_proj", "model.layers.67.mlp.experts.41.up_proj", "model.layers.67.mlp.experts.42.up_proj", "model.layers.67.mlp.experts.43.up_proj", "model.layers.67.mlp.experts.44.up_proj", "model.layers.67.mlp.experts.45.up_proj", "model.layers.67.mlp.experts.46.up_proj", "model.layers.67.mlp.experts.47.up_proj", "model.layers.67.mlp.experts.48.up_proj", "model.layers.67.mlp.experts.49.up_proj", "model.layers.67.mlp.experts.50.up_proj", "model.layers.67.mlp.experts.51.up_proj", "model.layers.67.mlp.experts.52.up_proj", "model.layers.67.mlp.experts.53.up_proj", "model.layers.67.mlp.experts.54.up_proj", "model.layers.67.mlp.experts.55.up_proj", "model.layers.67.mlp.experts.56.up_proj", "model.layers.67.mlp.experts.57.up_proj", "model.layers.67.mlp.experts.58.up_proj", "model.layers.67.mlp.experts.59.up_proj", "model.layers.67.mlp.experts.60.up_proj", "model.layers.67.mlp.experts.61.up_proj", "model.layers.67.mlp.experts.62.up_proj", "model.layers.67.mlp.experts.63.up_proj", "model.layers.67.mlp.experts.64.up_proj", "model.layers.67.mlp.experts.65.up_proj", "model.layers.67.mlp.experts.66.up_proj", "model.layers.67.mlp.experts.67.up_proj", "model.layers.67.mlp.experts.68.up_proj", "model.layers.67.mlp.experts.69.up_proj", "model.layers.67.mlp.experts.70.up_proj", "model.layers.67.mlp.experts.71.up_proj", "model.layers.67.mlp.experts.72.up_proj", "model.layers.67.mlp.experts.73.up_proj", "model.layers.67.mlp.experts.74.up_proj", "model.layers.67.mlp.experts.75.up_proj", "model.layers.67.mlp.experts.76.up_proj", "model.layers.67.mlp.experts.77.up_proj", "model.layers.67.mlp.experts.78.up_proj", "model.layers.67.mlp.experts.79.up_proj", "model.layers.67.mlp.experts.80.up_proj", "model.layers.67.mlp.experts.81.up_proj", "model.layers.67.mlp.experts.82.up_proj", "model.layers.67.mlp.experts.83.up_proj", "model.layers.67.mlp.experts.84.up_proj", "model.layers.67.mlp.experts.85.up_proj", "model.layers.67.mlp.experts.86.up_proj", "model.layers.67.mlp.experts.87.up_proj", "model.layers.67.mlp.experts.88.up_proj", "model.layers.67.mlp.experts.89.up_proj", "model.layers.67.mlp.experts.90.up_proj", "model.layers.67.mlp.experts.91.up_proj", "model.layers.67.mlp.experts.92.up_proj", "model.layers.67.mlp.experts.93.up_proj", "model.layers.67.mlp.experts.94.up_proj", "model.layers.67.mlp.experts.95.up_proj", "model.layers.67.mlp.experts.96.up_proj", "model.layers.67.mlp.experts.97.up_proj", "model.layers.67.mlp.experts.98.up_proj", "model.layers.67.mlp.experts.99.up_proj", "model.layers.67.mlp.experts.100.up_proj", "model.layers.67.mlp.experts.101.up_proj", "model.layers.67.mlp.experts.102.up_proj", "model.layers.67.mlp.experts.103.up_proj", "model.layers.67.mlp.experts.104.up_proj", "model.layers.67.mlp.experts.105.up_proj", "model.layers.67.mlp.experts.106.up_proj", "model.layers.67.mlp.experts.107.up_proj", "model.layers.67.mlp.experts.108.up_proj", "model.layers.67.mlp.experts.109.up_proj", "model.layers.67.mlp.experts.110.up_proj", "model.layers.67.mlp.experts.111.up_proj", "model.layers.67.mlp.experts.112.up_proj", "model.layers.67.mlp.experts.113.up_proj", "model.layers.67.mlp.experts.114.up_proj", "model.layers.67.mlp.experts.115.up_proj", "model.layers.67.mlp.experts.116.up_proj", "model.layers.67.mlp.experts.117.up_proj", "model.layers.67.mlp.experts.118.up_proj", "model.layers.67.mlp.experts.119.up_proj", "model.layers.67.mlp.experts.120.up_proj", "model.layers.67.mlp.experts.121.up_proj", "model.layers.67.mlp.experts.122.up_proj", "model.layers.67.mlp.experts.123.up_proj", "model.layers.67.mlp.experts.124.up_proj", "model.layers.67.mlp.experts.125.up_proj", "model.layers.67.mlp.experts.126.up_proj", "model.layers.67.mlp.experts.127.up_proj", "model.layers.67.mlp.experts.128.up_proj", "model.layers.67.mlp.experts.129.up_proj", "model.layers.67.mlp.experts.130.up_proj", "model.layers.67.mlp.experts.131.up_proj", "model.layers.67.mlp.experts.132.up_proj", "model.layers.67.mlp.experts.133.up_proj", "model.layers.67.mlp.experts.134.up_proj", "model.layers.67.mlp.experts.135.up_proj", "model.layers.67.mlp.experts.136.up_proj", "model.layers.67.mlp.experts.137.up_proj", "model.layers.67.mlp.experts.138.up_proj", "model.layers.67.mlp.experts.139.up_proj", "model.layers.67.mlp.experts.140.up_proj", "model.layers.67.mlp.experts.141.up_proj", "model.layers.67.mlp.experts.142.up_proj", "model.layers.67.mlp.experts.143.up_proj", "model.layers.67.mlp.experts.144.up_proj", "model.layers.67.mlp.experts.145.up_proj", "model.layers.67.mlp.experts.146.up_proj", "model.layers.67.mlp.experts.147.up_proj", "model.layers.67.mlp.experts.148.up_proj", "model.layers.67.mlp.experts.149.up_proj", "model.layers.67.mlp.experts.150.up_proj", "model.layers.67.mlp.experts.151.up_proj", "model.layers.67.mlp.experts.152.up_proj", "model.layers.67.mlp.experts.153.up_proj", "model.layers.67.mlp.experts.154.up_proj", "model.layers.67.mlp.experts.155.up_proj", "model.layers.67.mlp.experts.156.up_proj", "model.layers.67.mlp.experts.157.up_proj", "model.layers.67.mlp.experts.158.up_proj", "model.layers.67.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 4.036743193863712e-05, "dbits": 2516582400 }, { "dkld": -5.247266963125108e-05, "dbits": 5033164800 }, { "dkld": 7.946742698550224e-05, "dbits": 7549747200 }, { "dkld": 7.374417036770664e-05, "dbits": 12582912000 } ] }, { "idx": 404, "layers": [ "model.layers.67.mlp.experts.0.down_proj", "model.layers.67.mlp.experts.1.down_proj", "model.layers.67.mlp.experts.2.down_proj", "model.layers.67.mlp.experts.3.down_proj", "model.layers.67.mlp.experts.4.down_proj", "model.layers.67.mlp.experts.5.down_proj", "model.layers.67.mlp.experts.6.down_proj", "model.layers.67.mlp.experts.7.down_proj", "model.layers.67.mlp.experts.8.down_proj", "model.layers.67.mlp.experts.9.down_proj", "model.layers.67.mlp.experts.10.down_proj", "model.layers.67.mlp.experts.11.down_proj", "model.layers.67.mlp.experts.12.down_proj", "model.layers.67.mlp.experts.13.down_proj", "model.layers.67.mlp.experts.14.down_proj", "model.layers.67.mlp.experts.15.down_proj", "model.layers.67.mlp.experts.16.down_proj", "model.layers.67.mlp.experts.17.down_proj", "model.layers.67.mlp.experts.18.down_proj", "model.layers.67.mlp.experts.19.down_proj", "model.layers.67.mlp.experts.20.down_proj", "model.layers.67.mlp.experts.21.down_proj", "model.layers.67.mlp.experts.22.down_proj", "model.layers.67.mlp.experts.23.down_proj", "model.layers.67.mlp.experts.24.down_proj", "model.layers.67.mlp.experts.25.down_proj", "model.layers.67.mlp.experts.26.down_proj", "model.layers.67.mlp.experts.27.down_proj", "model.layers.67.mlp.experts.28.down_proj", "model.layers.67.mlp.experts.29.down_proj", "model.layers.67.mlp.experts.30.down_proj", "model.layers.67.mlp.experts.31.down_proj", "model.layers.67.mlp.experts.32.down_proj", "model.layers.67.mlp.experts.33.down_proj", "model.layers.67.mlp.experts.34.down_proj", "model.layers.67.mlp.experts.35.down_proj", "model.layers.67.mlp.experts.36.down_proj", "model.layers.67.mlp.experts.37.down_proj", "model.layers.67.mlp.experts.38.down_proj", "model.layers.67.mlp.experts.39.down_proj", "model.layers.67.mlp.experts.40.down_proj", "model.layers.67.mlp.experts.41.down_proj", "model.layers.67.mlp.experts.42.down_proj", "model.layers.67.mlp.experts.43.down_proj", "model.layers.67.mlp.experts.44.down_proj", "model.layers.67.mlp.experts.45.down_proj", "model.layers.67.mlp.experts.46.down_proj", "model.layers.67.mlp.experts.47.down_proj", "model.layers.67.mlp.experts.48.down_proj", "model.layers.67.mlp.experts.49.down_proj", "model.layers.67.mlp.experts.50.down_proj", "model.layers.67.mlp.experts.51.down_proj", "model.layers.67.mlp.experts.52.down_proj", "model.layers.67.mlp.experts.53.down_proj", "model.layers.67.mlp.experts.54.down_proj", "model.layers.67.mlp.experts.55.down_proj", "model.layers.67.mlp.experts.56.down_proj", "model.layers.67.mlp.experts.57.down_proj", "model.layers.67.mlp.experts.58.down_proj", "model.layers.67.mlp.experts.59.down_proj", "model.layers.67.mlp.experts.60.down_proj", "model.layers.67.mlp.experts.61.down_proj", "model.layers.67.mlp.experts.62.down_proj", "model.layers.67.mlp.experts.63.down_proj", "model.layers.67.mlp.experts.64.down_proj", "model.layers.67.mlp.experts.65.down_proj", "model.layers.67.mlp.experts.66.down_proj", "model.layers.67.mlp.experts.67.down_proj", "model.layers.67.mlp.experts.68.down_proj", "model.layers.67.mlp.experts.69.down_proj", "model.layers.67.mlp.experts.70.down_proj", "model.layers.67.mlp.experts.71.down_proj", "model.layers.67.mlp.experts.72.down_proj", "model.layers.67.mlp.experts.73.down_proj", "model.layers.67.mlp.experts.74.down_proj", "model.layers.67.mlp.experts.75.down_proj", "model.layers.67.mlp.experts.76.down_proj", "model.layers.67.mlp.experts.77.down_proj", "model.layers.67.mlp.experts.78.down_proj", "model.layers.67.mlp.experts.79.down_proj", "model.layers.67.mlp.experts.80.down_proj", "model.layers.67.mlp.experts.81.down_proj", "model.layers.67.mlp.experts.82.down_proj", "model.layers.67.mlp.experts.83.down_proj", "model.layers.67.mlp.experts.84.down_proj", "model.layers.67.mlp.experts.85.down_proj", "model.layers.67.mlp.experts.86.down_proj", "model.layers.67.mlp.experts.87.down_proj", "model.layers.67.mlp.experts.88.down_proj", "model.layers.67.mlp.experts.89.down_proj", "model.layers.67.mlp.experts.90.down_proj", "model.layers.67.mlp.experts.91.down_proj", "model.layers.67.mlp.experts.92.down_proj", "model.layers.67.mlp.experts.93.down_proj", "model.layers.67.mlp.experts.94.down_proj", "model.layers.67.mlp.experts.95.down_proj", "model.layers.67.mlp.experts.96.down_proj", "model.layers.67.mlp.experts.97.down_proj", "model.layers.67.mlp.experts.98.down_proj", "model.layers.67.mlp.experts.99.down_proj", "model.layers.67.mlp.experts.100.down_proj", "model.layers.67.mlp.experts.101.down_proj", "model.layers.67.mlp.experts.102.down_proj", "model.layers.67.mlp.experts.103.down_proj", "model.layers.67.mlp.experts.104.down_proj", "model.layers.67.mlp.experts.105.down_proj", "model.layers.67.mlp.experts.106.down_proj", "model.layers.67.mlp.experts.107.down_proj", "model.layers.67.mlp.experts.108.down_proj", "model.layers.67.mlp.experts.109.down_proj", "model.layers.67.mlp.experts.110.down_proj", "model.layers.67.mlp.experts.111.down_proj", "model.layers.67.mlp.experts.112.down_proj", "model.layers.67.mlp.experts.113.down_proj", "model.layers.67.mlp.experts.114.down_proj", "model.layers.67.mlp.experts.115.down_proj", "model.layers.67.mlp.experts.116.down_proj", "model.layers.67.mlp.experts.117.down_proj", "model.layers.67.mlp.experts.118.down_proj", "model.layers.67.mlp.experts.119.down_proj", "model.layers.67.mlp.experts.120.down_proj", "model.layers.67.mlp.experts.121.down_proj", "model.layers.67.mlp.experts.122.down_proj", "model.layers.67.mlp.experts.123.down_proj", "model.layers.67.mlp.experts.124.down_proj", "model.layers.67.mlp.experts.125.down_proj", "model.layers.67.mlp.experts.126.down_proj", "model.layers.67.mlp.experts.127.down_proj", "model.layers.67.mlp.experts.128.down_proj", "model.layers.67.mlp.experts.129.down_proj", "model.layers.67.mlp.experts.130.down_proj", "model.layers.67.mlp.experts.131.down_proj", "model.layers.67.mlp.experts.132.down_proj", "model.layers.67.mlp.experts.133.down_proj", "model.layers.67.mlp.experts.134.down_proj", "model.layers.67.mlp.experts.135.down_proj", "model.layers.67.mlp.experts.136.down_proj", "model.layers.67.mlp.experts.137.down_proj", "model.layers.67.mlp.experts.138.down_proj", "model.layers.67.mlp.experts.139.down_proj", "model.layers.67.mlp.experts.140.down_proj", "model.layers.67.mlp.experts.141.down_proj", "model.layers.67.mlp.experts.142.down_proj", "model.layers.67.mlp.experts.143.down_proj", "model.layers.67.mlp.experts.144.down_proj", "model.layers.67.mlp.experts.145.down_proj", "model.layers.67.mlp.experts.146.down_proj", "model.layers.67.mlp.experts.147.down_proj", "model.layers.67.mlp.experts.148.down_proj", "model.layers.67.mlp.experts.149.down_proj", "model.layers.67.mlp.experts.150.down_proj", "model.layers.67.mlp.experts.151.down_proj", "model.layers.67.mlp.experts.152.down_proj", "model.layers.67.mlp.experts.153.down_proj", "model.layers.67.mlp.experts.154.down_proj", "model.layers.67.mlp.experts.155.down_proj", "model.layers.67.mlp.experts.156.down_proj", "model.layers.67.mlp.experts.157.down_proj", "model.layers.67.mlp.experts.158.down_proj", "model.layers.67.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00025773588567973293, "dbits": 1258291200 }, { "dkld": -0.00022119088098407347, "dbits": 2516582400 }, { "dkld": -0.0001980166882276646, "dbits": 3774873600 }, { "dkld": -0.00019019767642021457, "dbits": 6291456000 } ] }, { "idx": 405, "layers": [ "model.layers.68.self_attn.q_proj" ], "candidates": [ { "dkld": -8.805645629764158e-05, "dbits": 62914560 }, { "dkld": -0.00021300856024027148, "dbits": 125829120 }, { "dkld": -0.0001812529750168379, "dbits": 188743680 }, { "dkld": -0.00020125862210990109, "dbits": 314572800 } ] }, { "idx": 406, "layers": [ "model.layers.68.self_attn.k_proj", "model.layers.68.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0003111756406724564, "dbits": 10485760 }, { "dkld": -0.00011232830584050058, "dbits": 20971520 }, { "dkld": -0.00011614570394158363, "dbits": 31457280 }, { "dkld": -0.0001009100116789452, "dbits": 52428800 } ] }, { "idx": 407, "layers": [ "model.layers.68.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00022122431546450771, "dbits": 62914560 }, { "dkld": -0.00022586956620217202, "dbits": 125829120 }, { "dkld": -0.00029792990535498776, "dbits": 188743680 }, { "dkld": -0.0002729000523686437, "dbits": 314572800 } ] }, { "idx": 408, "layers": [ "model.layers.68.mlp.shared_experts.gate_proj", "model.layers.68.mlp.shared_experts.up_proj", "model.layers.68.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 6.902143359183155e-05, "dbits": 23592960 }, { "dkld": 1.845583319663724e-05, "dbits": 47185920 }, { "dkld": -3.065215423703194e-05, "dbits": 70778880 }, { "dkld": 1.5136785805197617e-06, "dbits": 117964800 } ] }, { "idx": 409, "layers": [ "model.layers.68.mlp.experts.0.gate_proj", "model.layers.68.mlp.experts.1.gate_proj", "model.layers.68.mlp.experts.2.gate_proj", "model.layers.68.mlp.experts.3.gate_proj", "model.layers.68.mlp.experts.4.gate_proj", "model.layers.68.mlp.experts.5.gate_proj", "model.layers.68.mlp.experts.6.gate_proj", "model.layers.68.mlp.experts.7.gate_proj", "model.layers.68.mlp.experts.8.gate_proj", "model.layers.68.mlp.experts.9.gate_proj", "model.layers.68.mlp.experts.10.gate_proj", "model.layers.68.mlp.experts.11.gate_proj", "model.layers.68.mlp.experts.12.gate_proj", "model.layers.68.mlp.experts.13.gate_proj", "model.layers.68.mlp.experts.14.gate_proj", "model.layers.68.mlp.experts.15.gate_proj", "model.layers.68.mlp.experts.16.gate_proj", "model.layers.68.mlp.experts.17.gate_proj", "model.layers.68.mlp.experts.18.gate_proj", "model.layers.68.mlp.experts.19.gate_proj", "model.layers.68.mlp.experts.20.gate_proj", "model.layers.68.mlp.experts.21.gate_proj", "model.layers.68.mlp.experts.22.gate_proj", "model.layers.68.mlp.experts.23.gate_proj", "model.layers.68.mlp.experts.24.gate_proj", "model.layers.68.mlp.experts.25.gate_proj", "model.layers.68.mlp.experts.26.gate_proj", "model.layers.68.mlp.experts.27.gate_proj", "model.layers.68.mlp.experts.28.gate_proj", "model.layers.68.mlp.experts.29.gate_proj", "model.layers.68.mlp.experts.30.gate_proj", "model.layers.68.mlp.experts.31.gate_proj", "model.layers.68.mlp.experts.32.gate_proj", "model.layers.68.mlp.experts.33.gate_proj", "model.layers.68.mlp.experts.34.gate_proj", "model.layers.68.mlp.experts.35.gate_proj", "model.layers.68.mlp.experts.36.gate_proj", "model.layers.68.mlp.experts.37.gate_proj", "model.layers.68.mlp.experts.38.gate_proj", "model.layers.68.mlp.experts.39.gate_proj", "model.layers.68.mlp.experts.40.gate_proj", "model.layers.68.mlp.experts.41.gate_proj", "model.layers.68.mlp.experts.42.gate_proj", "model.layers.68.mlp.experts.43.gate_proj", "model.layers.68.mlp.experts.44.gate_proj", "model.layers.68.mlp.experts.45.gate_proj", "model.layers.68.mlp.experts.46.gate_proj", "model.layers.68.mlp.experts.47.gate_proj", "model.layers.68.mlp.experts.48.gate_proj", "model.layers.68.mlp.experts.49.gate_proj", "model.layers.68.mlp.experts.50.gate_proj", "model.layers.68.mlp.experts.51.gate_proj", "model.layers.68.mlp.experts.52.gate_proj", "model.layers.68.mlp.experts.53.gate_proj", "model.layers.68.mlp.experts.54.gate_proj", "model.layers.68.mlp.experts.55.gate_proj", "model.layers.68.mlp.experts.56.gate_proj", "model.layers.68.mlp.experts.57.gate_proj", "model.layers.68.mlp.experts.58.gate_proj", "model.layers.68.mlp.experts.59.gate_proj", "model.layers.68.mlp.experts.60.gate_proj", "model.layers.68.mlp.experts.61.gate_proj", "model.layers.68.mlp.experts.62.gate_proj", "model.layers.68.mlp.experts.63.gate_proj", "model.layers.68.mlp.experts.64.gate_proj", "model.layers.68.mlp.experts.65.gate_proj", "model.layers.68.mlp.experts.66.gate_proj", "model.layers.68.mlp.experts.67.gate_proj", "model.layers.68.mlp.experts.68.gate_proj", "model.layers.68.mlp.experts.69.gate_proj", "model.layers.68.mlp.experts.70.gate_proj", "model.layers.68.mlp.experts.71.gate_proj", "model.layers.68.mlp.experts.72.gate_proj", "model.layers.68.mlp.experts.73.gate_proj", "model.layers.68.mlp.experts.74.gate_proj", "model.layers.68.mlp.experts.75.gate_proj", "model.layers.68.mlp.experts.76.gate_proj", "model.layers.68.mlp.experts.77.gate_proj", "model.layers.68.mlp.experts.78.gate_proj", "model.layers.68.mlp.experts.79.gate_proj", "model.layers.68.mlp.experts.80.gate_proj", "model.layers.68.mlp.experts.81.gate_proj", "model.layers.68.mlp.experts.82.gate_proj", "model.layers.68.mlp.experts.83.gate_proj", "model.layers.68.mlp.experts.84.gate_proj", "model.layers.68.mlp.experts.85.gate_proj", "model.layers.68.mlp.experts.86.gate_proj", "model.layers.68.mlp.experts.87.gate_proj", "model.layers.68.mlp.experts.88.gate_proj", "model.layers.68.mlp.experts.89.gate_proj", "model.layers.68.mlp.experts.90.gate_proj", "model.layers.68.mlp.experts.91.gate_proj", "model.layers.68.mlp.experts.92.gate_proj", "model.layers.68.mlp.experts.93.gate_proj", "model.layers.68.mlp.experts.94.gate_proj", "model.layers.68.mlp.experts.95.gate_proj", "model.layers.68.mlp.experts.96.gate_proj", "model.layers.68.mlp.experts.97.gate_proj", "model.layers.68.mlp.experts.98.gate_proj", "model.layers.68.mlp.experts.99.gate_proj", "model.layers.68.mlp.experts.100.gate_proj", "model.layers.68.mlp.experts.101.gate_proj", "model.layers.68.mlp.experts.102.gate_proj", "model.layers.68.mlp.experts.103.gate_proj", "model.layers.68.mlp.experts.104.gate_proj", "model.layers.68.mlp.experts.105.gate_proj", "model.layers.68.mlp.experts.106.gate_proj", "model.layers.68.mlp.experts.107.gate_proj", "model.layers.68.mlp.experts.108.gate_proj", "model.layers.68.mlp.experts.109.gate_proj", "model.layers.68.mlp.experts.110.gate_proj", "model.layers.68.mlp.experts.111.gate_proj", "model.layers.68.mlp.experts.112.gate_proj", "model.layers.68.mlp.experts.113.gate_proj", "model.layers.68.mlp.experts.114.gate_proj", "model.layers.68.mlp.experts.115.gate_proj", "model.layers.68.mlp.experts.116.gate_proj", "model.layers.68.mlp.experts.117.gate_proj", "model.layers.68.mlp.experts.118.gate_proj", "model.layers.68.mlp.experts.119.gate_proj", "model.layers.68.mlp.experts.120.gate_proj", "model.layers.68.mlp.experts.121.gate_proj", "model.layers.68.mlp.experts.122.gate_proj", "model.layers.68.mlp.experts.123.gate_proj", "model.layers.68.mlp.experts.124.gate_proj", "model.layers.68.mlp.experts.125.gate_proj", "model.layers.68.mlp.experts.126.gate_proj", "model.layers.68.mlp.experts.127.gate_proj", "model.layers.68.mlp.experts.128.gate_proj", "model.layers.68.mlp.experts.129.gate_proj", "model.layers.68.mlp.experts.130.gate_proj", "model.layers.68.mlp.experts.131.gate_proj", "model.layers.68.mlp.experts.132.gate_proj", "model.layers.68.mlp.experts.133.gate_proj", "model.layers.68.mlp.experts.134.gate_proj", "model.layers.68.mlp.experts.135.gate_proj", "model.layers.68.mlp.experts.136.gate_proj", "model.layers.68.mlp.experts.137.gate_proj", "model.layers.68.mlp.experts.138.gate_proj", "model.layers.68.mlp.experts.139.gate_proj", "model.layers.68.mlp.experts.140.gate_proj", "model.layers.68.mlp.experts.141.gate_proj", "model.layers.68.mlp.experts.142.gate_proj", "model.layers.68.mlp.experts.143.gate_proj", "model.layers.68.mlp.experts.144.gate_proj", "model.layers.68.mlp.experts.145.gate_proj", "model.layers.68.mlp.experts.146.gate_proj", "model.layers.68.mlp.experts.147.gate_proj", "model.layers.68.mlp.experts.148.gate_proj", "model.layers.68.mlp.experts.149.gate_proj", "model.layers.68.mlp.experts.150.gate_proj", "model.layers.68.mlp.experts.151.gate_proj", "model.layers.68.mlp.experts.152.gate_proj", "model.layers.68.mlp.experts.153.gate_proj", "model.layers.68.mlp.experts.154.gate_proj", "model.layers.68.mlp.experts.155.gate_proj", "model.layers.68.mlp.experts.156.gate_proj", "model.layers.68.mlp.experts.157.gate_proj", "model.layers.68.mlp.experts.158.gate_proj", "model.layers.68.mlp.experts.159.gate_proj", "model.layers.68.mlp.experts.0.up_proj", "model.layers.68.mlp.experts.1.up_proj", "model.layers.68.mlp.experts.2.up_proj", "model.layers.68.mlp.experts.3.up_proj", "model.layers.68.mlp.experts.4.up_proj", "model.layers.68.mlp.experts.5.up_proj", "model.layers.68.mlp.experts.6.up_proj", "model.layers.68.mlp.experts.7.up_proj", "model.layers.68.mlp.experts.8.up_proj", "model.layers.68.mlp.experts.9.up_proj", "model.layers.68.mlp.experts.10.up_proj", "model.layers.68.mlp.experts.11.up_proj", "model.layers.68.mlp.experts.12.up_proj", "model.layers.68.mlp.experts.13.up_proj", "model.layers.68.mlp.experts.14.up_proj", "model.layers.68.mlp.experts.15.up_proj", "model.layers.68.mlp.experts.16.up_proj", "model.layers.68.mlp.experts.17.up_proj", "model.layers.68.mlp.experts.18.up_proj", "model.layers.68.mlp.experts.19.up_proj", "model.layers.68.mlp.experts.20.up_proj", "model.layers.68.mlp.experts.21.up_proj", "model.layers.68.mlp.experts.22.up_proj", "model.layers.68.mlp.experts.23.up_proj", "model.layers.68.mlp.experts.24.up_proj", "model.layers.68.mlp.experts.25.up_proj", "model.layers.68.mlp.experts.26.up_proj", "model.layers.68.mlp.experts.27.up_proj", "model.layers.68.mlp.experts.28.up_proj", "model.layers.68.mlp.experts.29.up_proj", "model.layers.68.mlp.experts.30.up_proj", "model.layers.68.mlp.experts.31.up_proj", "model.layers.68.mlp.experts.32.up_proj", "model.layers.68.mlp.experts.33.up_proj", "model.layers.68.mlp.experts.34.up_proj", "model.layers.68.mlp.experts.35.up_proj", "model.layers.68.mlp.experts.36.up_proj", "model.layers.68.mlp.experts.37.up_proj", "model.layers.68.mlp.experts.38.up_proj", "model.layers.68.mlp.experts.39.up_proj", "model.layers.68.mlp.experts.40.up_proj", "model.layers.68.mlp.experts.41.up_proj", "model.layers.68.mlp.experts.42.up_proj", "model.layers.68.mlp.experts.43.up_proj", "model.layers.68.mlp.experts.44.up_proj", "model.layers.68.mlp.experts.45.up_proj", "model.layers.68.mlp.experts.46.up_proj", "model.layers.68.mlp.experts.47.up_proj", "model.layers.68.mlp.experts.48.up_proj", "model.layers.68.mlp.experts.49.up_proj", "model.layers.68.mlp.experts.50.up_proj", "model.layers.68.mlp.experts.51.up_proj", "model.layers.68.mlp.experts.52.up_proj", "model.layers.68.mlp.experts.53.up_proj", "model.layers.68.mlp.experts.54.up_proj", "model.layers.68.mlp.experts.55.up_proj", "model.layers.68.mlp.experts.56.up_proj", "model.layers.68.mlp.experts.57.up_proj", "model.layers.68.mlp.experts.58.up_proj", "model.layers.68.mlp.experts.59.up_proj", "model.layers.68.mlp.experts.60.up_proj", "model.layers.68.mlp.experts.61.up_proj", "model.layers.68.mlp.experts.62.up_proj", "model.layers.68.mlp.experts.63.up_proj", "model.layers.68.mlp.experts.64.up_proj", "model.layers.68.mlp.experts.65.up_proj", "model.layers.68.mlp.experts.66.up_proj", "model.layers.68.mlp.experts.67.up_proj", "model.layers.68.mlp.experts.68.up_proj", "model.layers.68.mlp.experts.69.up_proj", "model.layers.68.mlp.experts.70.up_proj", "model.layers.68.mlp.experts.71.up_proj", "model.layers.68.mlp.experts.72.up_proj", "model.layers.68.mlp.experts.73.up_proj", "model.layers.68.mlp.experts.74.up_proj", "model.layers.68.mlp.experts.75.up_proj", "model.layers.68.mlp.experts.76.up_proj", "model.layers.68.mlp.experts.77.up_proj", "model.layers.68.mlp.experts.78.up_proj", "model.layers.68.mlp.experts.79.up_proj", "model.layers.68.mlp.experts.80.up_proj", "model.layers.68.mlp.experts.81.up_proj", "model.layers.68.mlp.experts.82.up_proj", "model.layers.68.mlp.experts.83.up_proj", "model.layers.68.mlp.experts.84.up_proj", "model.layers.68.mlp.experts.85.up_proj", "model.layers.68.mlp.experts.86.up_proj", "model.layers.68.mlp.experts.87.up_proj", "model.layers.68.mlp.experts.88.up_proj", "model.layers.68.mlp.experts.89.up_proj", "model.layers.68.mlp.experts.90.up_proj", "model.layers.68.mlp.experts.91.up_proj", "model.layers.68.mlp.experts.92.up_proj", "model.layers.68.mlp.experts.93.up_proj", "model.layers.68.mlp.experts.94.up_proj", "model.layers.68.mlp.experts.95.up_proj", "model.layers.68.mlp.experts.96.up_proj", "model.layers.68.mlp.experts.97.up_proj", "model.layers.68.mlp.experts.98.up_proj", "model.layers.68.mlp.experts.99.up_proj", "model.layers.68.mlp.experts.100.up_proj", "model.layers.68.mlp.experts.101.up_proj", "model.layers.68.mlp.experts.102.up_proj", "model.layers.68.mlp.experts.103.up_proj", "model.layers.68.mlp.experts.104.up_proj", "model.layers.68.mlp.experts.105.up_proj", "model.layers.68.mlp.experts.106.up_proj", "model.layers.68.mlp.experts.107.up_proj", "model.layers.68.mlp.experts.108.up_proj", "model.layers.68.mlp.experts.109.up_proj", "model.layers.68.mlp.experts.110.up_proj", "model.layers.68.mlp.experts.111.up_proj", "model.layers.68.mlp.experts.112.up_proj", "model.layers.68.mlp.experts.113.up_proj", "model.layers.68.mlp.experts.114.up_proj", "model.layers.68.mlp.experts.115.up_proj", "model.layers.68.mlp.experts.116.up_proj", "model.layers.68.mlp.experts.117.up_proj", "model.layers.68.mlp.experts.118.up_proj", "model.layers.68.mlp.experts.119.up_proj", "model.layers.68.mlp.experts.120.up_proj", "model.layers.68.mlp.experts.121.up_proj", "model.layers.68.mlp.experts.122.up_proj", "model.layers.68.mlp.experts.123.up_proj", "model.layers.68.mlp.experts.124.up_proj", "model.layers.68.mlp.experts.125.up_proj", "model.layers.68.mlp.experts.126.up_proj", "model.layers.68.mlp.experts.127.up_proj", "model.layers.68.mlp.experts.128.up_proj", "model.layers.68.mlp.experts.129.up_proj", "model.layers.68.mlp.experts.130.up_proj", "model.layers.68.mlp.experts.131.up_proj", "model.layers.68.mlp.experts.132.up_proj", "model.layers.68.mlp.experts.133.up_proj", "model.layers.68.mlp.experts.134.up_proj", "model.layers.68.mlp.experts.135.up_proj", "model.layers.68.mlp.experts.136.up_proj", "model.layers.68.mlp.experts.137.up_proj", "model.layers.68.mlp.experts.138.up_proj", "model.layers.68.mlp.experts.139.up_proj", "model.layers.68.mlp.experts.140.up_proj", "model.layers.68.mlp.experts.141.up_proj", "model.layers.68.mlp.experts.142.up_proj", "model.layers.68.mlp.experts.143.up_proj", "model.layers.68.mlp.experts.144.up_proj", "model.layers.68.mlp.experts.145.up_proj", "model.layers.68.mlp.experts.146.up_proj", "model.layers.68.mlp.experts.147.up_proj", "model.layers.68.mlp.experts.148.up_proj", "model.layers.68.mlp.experts.149.up_proj", "model.layers.68.mlp.experts.150.up_proj", "model.layers.68.mlp.experts.151.up_proj", "model.layers.68.mlp.experts.152.up_proj", "model.layers.68.mlp.experts.153.up_proj", "model.layers.68.mlp.experts.154.up_proj", "model.layers.68.mlp.experts.155.up_proj", "model.layers.68.mlp.experts.156.up_proj", "model.layers.68.mlp.experts.157.up_proj", "model.layers.68.mlp.experts.158.up_proj", "model.layers.68.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -5.769561976194659e-05, "dbits": 2516582400 }, { "dkld": -0.00013026306405664045, "dbits": 5033164800 }, { "dkld": -7.172888144851841e-05, "dbits": 7549747200 }, { "dkld": -7.909536361694336e-05, "dbits": 12582912000 } ] }, { "idx": 410, "layers": [ "model.layers.68.mlp.experts.0.down_proj", "model.layers.68.mlp.experts.1.down_proj", "model.layers.68.mlp.experts.2.down_proj", "model.layers.68.mlp.experts.3.down_proj", "model.layers.68.mlp.experts.4.down_proj", "model.layers.68.mlp.experts.5.down_proj", "model.layers.68.mlp.experts.6.down_proj", "model.layers.68.mlp.experts.7.down_proj", "model.layers.68.mlp.experts.8.down_proj", "model.layers.68.mlp.experts.9.down_proj", "model.layers.68.mlp.experts.10.down_proj", "model.layers.68.mlp.experts.11.down_proj", "model.layers.68.mlp.experts.12.down_proj", "model.layers.68.mlp.experts.13.down_proj", "model.layers.68.mlp.experts.14.down_proj", "model.layers.68.mlp.experts.15.down_proj", "model.layers.68.mlp.experts.16.down_proj", "model.layers.68.mlp.experts.17.down_proj", "model.layers.68.mlp.experts.18.down_proj", "model.layers.68.mlp.experts.19.down_proj", "model.layers.68.mlp.experts.20.down_proj", "model.layers.68.mlp.experts.21.down_proj", "model.layers.68.mlp.experts.22.down_proj", "model.layers.68.mlp.experts.23.down_proj", "model.layers.68.mlp.experts.24.down_proj", "model.layers.68.mlp.experts.25.down_proj", "model.layers.68.mlp.experts.26.down_proj", "model.layers.68.mlp.experts.27.down_proj", "model.layers.68.mlp.experts.28.down_proj", "model.layers.68.mlp.experts.29.down_proj", "model.layers.68.mlp.experts.30.down_proj", "model.layers.68.mlp.experts.31.down_proj", "model.layers.68.mlp.experts.32.down_proj", "model.layers.68.mlp.experts.33.down_proj", "model.layers.68.mlp.experts.34.down_proj", "model.layers.68.mlp.experts.35.down_proj", "model.layers.68.mlp.experts.36.down_proj", "model.layers.68.mlp.experts.37.down_proj", "model.layers.68.mlp.experts.38.down_proj", "model.layers.68.mlp.experts.39.down_proj", "model.layers.68.mlp.experts.40.down_proj", "model.layers.68.mlp.experts.41.down_proj", "model.layers.68.mlp.experts.42.down_proj", "model.layers.68.mlp.experts.43.down_proj", "model.layers.68.mlp.experts.44.down_proj", "model.layers.68.mlp.experts.45.down_proj", "model.layers.68.mlp.experts.46.down_proj", "model.layers.68.mlp.experts.47.down_proj", "model.layers.68.mlp.experts.48.down_proj", "model.layers.68.mlp.experts.49.down_proj", "model.layers.68.mlp.experts.50.down_proj", "model.layers.68.mlp.experts.51.down_proj", "model.layers.68.mlp.experts.52.down_proj", "model.layers.68.mlp.experts.53.down_proj", "model.layers.68.mlp.experts.54.down_proj", "model.layers.68.mlp.experts.55.down_proj", "model.layers.68.mlp.experts.56.down_proj", "model.layers.68.mlp.experts.57.down_proj", "model.layers.68.mlp.experts.58.down_proj", "model.layers.68.mlp.experts.59.down_proj", "model.layers.68.mlp.experts.60.down_proj", "model.layers.68.mlp.experts.61.down_proj", "model.layers.68.mlp.experts.62.down_proj", "model.layers.68.mlp.experts.63.down_proj", "model.layers.68.mlp.experts.64.down_proj", "model.layers.68.mlp.experts.65.down_proj", "model.layers.68.mlp.experts.66.down_proj", "model.layers.68.mlp.experts.67.down_proj", "model.layers.68.mlp.experts.68.down_proj", "model.layers.68.mlp.experts.69.down_proj", "model.layers.68.mlp.experts.70.down_proj", "model.layers.68.mlp.experts.71.down_proj", "model.layers.68.mlp.experts.72.down_proj", "model.layers.68.mlp.experts.73.down_proj", "model.layers.68.mlp.experts.74.down_proj", "model.layers.68.mlp.experts.75.down_proj", "model.layers.68.mlp.experts.76.down_proj", "model.layers.68.mlp.experts.77.down_proj", "model.layers.68.mlp.experts.78.down_proj", "model.layers.68.mlp.experts.79.down_proj", "model.layers.68.mlp.experts.80.down_proj", "model.layers.68.mlp.experts.81.down_proj", "model.layers.68.mlp.experts.82.down_proj", "model.layers.68.mlp.experts.83.down_proj", "model.layers.68.mlp.experts.84.down_proj", "model.layers.68.mlp.experts.85.down_proj", "model.layers.68.mlp.experts.86.down_proj", "model.layers.68.mlp.experts.87.down_proj", "model.layers.68.mlp.experts.88.down_proj", "model.layers.68.mlp.experts.89.down_proj", "model.layers.68.mlp.experts.90.down_proj", "model.layers.68.mlp.experts.91.down_proj", "model.layers.68.mlp.experts.92.down_proj", "model.layers.68.mlp.experts.93.down_proj", "model.layers.68.mlp.experts.94.down_proj", "model.layers.68.mlp.experts.95.down_proj", "model.layers.68.mlp.experts.96.down_proj", "model.layers.68.mlp.experts.97.down_proj", "model.layers.68.mlp.experts.98.down_proj", "model.layers.68.mlp.experts.99.down_proj", "model.layers.68.mlp.experts.100.down_proj", "model.layers.68.mlp.experts.101.down_proj", "model.layers.68.mlp.experts.102.down_proj", "model.layers.68.mlp.experts.103.down_proj", "model.layers.68.mlp.experts.104.down_proj", "model.layers.68.mlp.experts.105.down_proj", "model.layers.68.mlp.experts.106.down_proj", "model.layers.68.mlp.experts.107.down_proj", "model.layers.68.mlp.experts.108.down_proj", "model.layers.68.mlp.experts.109.down_proj", "model.layers.68.mlp.experts.110.down_proj", "model.layers.68.mlp.experts.111.down_proj", "model.layers.68.mlp.experts.112.down_proj", "model.layers.68.mlp.experts.113.down_proj", "model.layers.68.mlp.experts.114.down_proj", "model.layers.68.mlp.experts.115.down_proj", "model.layers.68.mlp.experts.116.down_proj", "model.layers.68.mlp.experts.117.down_proj", "model.layers.68.mlp.experts.118.down_proj", "model.layers.68.mlp.experts.119.down_proj", "model.layers.68.mlp.experts.120.down_proj", "model.layers.68.mlp.experts.121.down_proj", "model.layers.68.mlp.experts.122.down_proj", "model.layers.68.mlp.experts.123.down_proj", "model.layers.68.mlp.experts.124.down_proj", "model.layers.68.mlp.experts.125.down_proj", "model.layers.68.mlp.experts.126.down_proj", "model.layers.68.mlp.experts.127.down_proj", "model.layers.68.mlp.experts.128.down_proj", "model.layers.68.mlp.experts.129.down_proj", "model.layers.68.mlp.experts.130.down_proj", "model.layers.68.mlp.experts.131.down_proj", "model.layers.68.mlp.experts.132.down_proj", "model.layers.68.mlp.experts.133.down_proj", "model.layers.68.mlp.experts.134.down_proj", "model.layers.68.mlp.experts.135.down_proj", "model.layers.68.mlp.experts.136.down_proj", "model.layers.68.mlp.experts.137.down_proj", "model.layers.68.mlp.experts.138.down_proj", "model.layers.68.mlp.experts.139.down_proj", "model.layers.68.mlp.experts.140.down_proj", "model.layers.68.mlp.experts.141.down_proj", "model.layers.68.mlp.experts.142.down_proj", "model.layers.68.mlp.experts.143.down_proj", "model.layers.68.mlp.experts.144.down_proj", "model.layers.68.mlp.experts.145.down_proj", "model.layers.68.mlp.experts.146.down_proj", "model.layers.68.mlp.experts.147.down_proj", "model.layers.68.mlp.experts.148.down_proj", "model.layers.68.mlp.experts.149.down_proj", "model.layers.68.mlp.experts.150.down_proj", "model.layers.68.mlp.experts.151.down_proj", "model.layers.68.mlp.experts.152.down_proj", "model.layers.68.mlp.experts.153.down_proj", "model.layers.68.mlp.experts.154.down_proj", "model.layers.68.mlp.experts.155.down_proj", "model.layers.68.mlp.experts.156.down_proj", "model.layers.68.mlp.experts.157.down_proj", "model.layers.68.mlp.experts.158.down_proj", "model.layers.68.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0001759103499353043, "dbits": 1258291200 }, { "dkld": -0.0003074872307479437, "dbits": 2516582400 }, { "dkld": -0.0002849592827260522, "dbits": 3774873600 }, { "dkld": -0.00032230373471975327, "dbits": 6291456000 } ] }, { "idx": 411, "layers": [ "model.layers.69.self_attn.q_proj" ], "candidates": [ { "dkld": -7.452890276909985e-05, "dbits": 62914560 }, { "dkld": -0.00010588578879833221, "dbits": 125829120 }, { "dkld": -0.00010245274752379574, "dbits": 188743680 }, { "dkld": -0.0001354400068521583, "dbits": 314572800 } ] }, { "idx": 412, "layers": [ "model.layers.69.self_attn.k_proj", "model.layers.69.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00018239375203848995, "dbits": 10485760 }, { "dkld": -0.00032797139137984155, "dbits": 20971520 }, { "dkld": -0.00028563644737006344, "dbits": 31457280 }, { "dkld": -0.0003016667440533666, "dbits": 52428800 } ] }, { "idx": 413, "layers": [ "model.layers.69.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00012303516268730164, "dbits": 62914560 }, { "dkld": -0.00010783150792122165, "dbits": 125829120 }, { "dkld": -0.00011175330728292743, "dbits": 188743680 }, { "dkld": -0.00013362206518650888, "dbits": 314572800 } ] }, { "idx": 414, "layers": [ "model.layers.69.mlp.shared_experts.gate_proj", "model.layers.69.mlp.shared_experts.up_proj", "model.layers.69.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -3.701653331518451e-05, "dbits": 23592960 }, { "dkld": -7.82866030931556e-05, "dbits": 47185920 }, { "dkld": -6.607742980123399e-05, "dbits": 70778880 }, { "dkld": -3.2950937747966424e-05, "dbits": 117964800 } ] }, { "idx": 415, "layers": [ "model.layers.69.mlp.experts.0.gate_proj", "model.layers.69.mlp.experts.1.gate_proj", "model.layers.69.mlp.experts.2.gate_proj", "model.layers.69.mlp.experts.3.gate_proj", "model.layers.69.mlp.experts.4.gate_proj", "model.layers.69.mlp.experts.5.gate_proj", "model.layers.69.mlp.experts.6.gate_proj", "model.layers.69.mlp.experts.7.gate_proj", "model.layers.69.mlp.experts.8.gate_proj", "model.layers.69.mlp.experts.9.gate_proj", "model.layers.69.mlp.experts.10.gate_proj", "model.layers.69.mlp.experts.11.gate_proj", "model.layers.69.mlp.experts.12.gate_proj", "model.layers.69.mlp.experts.13.gate_proj", "model.layers.69.mlp.experts.14.gate_proj", "model.layers.69.mlp.experts.15.gate_proj", "model.layers.69.mlp.experts.16.gate_proj", "model.layers.69.mlp.experts.17.gate_proj", "model.layers.69.mlp.experts.18.gate_proj", "model.layers.69.mlp.experts.19.gate_proj", "model.layers.69.mlp.experts.20.gate_proj", "model.layers.69.mlp.experts.21.gate_proj", "model.layers.69.mlp.experts.22.gate_proj", "model.layers.69.mlp.experts.23.gate_proj", "model.layers.69.mlp.experts.24.gate_proj", "model.layers.69.mlp.experts.25.gate_proj", "model.layers.69.mlp.experts.26.gate_proj", "model.layers.69.mlp.experts.27.gate_proj", "model.layers.69.mlp.experts.28.gate_proj", "model.layers.69.mlp.experts.29.gate_proj", "model.layers.69.mlp.experts.30.gate_proj", "model.layers.69.mlp.experts.31.gate_proj", "model.layers.69.mlp.experts.32.gate_proj", "model.layers.69.mlp.experts.33.gate_proj", "model.layers.69.mlp.experts.34.gate_proj", "model.layers.69.mlp.experts.35.gate_proj", "model.layers.69.mlp.experts.36.gate_proj", "model.layers.69.mlp.experts.37.gate_proj", "model.layers.69.mlp.experts.38.gate_proj", "model.layers.69.mlp.experts.39.gate_proj", "model.layers.69.mlp.experts.40.gate_proj", "model.layers.69.mlp.experts.41.gate_proj", "model.layers.69.mlp.experts.42.gate_proj", "model.layers.69.mlp.experts.43.gate_proj", "model.layers.69.mlp.experts.44.gate_proj", "model.layers.69.mlp.experts.45.gate_proj", "model.layers.69.mlp.experts.46.gate_proj", "model.layers.69.mlp.experts.47.gate_proj", "model.layers.69.mlp.experts.48.gate_proj", "model.layers.69.mlp.experts.49.gate_proj", "model.layers.69.mlp.experts.50.gate_proj", "model.layers.69.mlp.experts.51.gate_proj", "model.layers.69.mlp.experts.52.gate_proj", "model.layers.69.mlp.experts.53.gate_proj", "model.layers.69.mlp.experts.54.gate_proj", "model.layers.69.mlp.experts.55.gate_proj", "model.layers.69.mlp.experts.56.gate_proj", "model.layers.69.mlp.experts.57.gate_proj", "model.layers.69.mlp.experts.58.gate_proj", "model.layers.69.mlp.experts.59.gate_proj", "model.layers.69.mlp.experts.60.gate_proj", "model.layers.69.mlp.experts.61.gate_proj", "model.layers.69.mlp.experts.62.gate_proj", "model.layers.69.mlp.experts.63.gate_proj", "model.layers.69.mlp.experts.64.gate_proj", "model.layers.69.mlp.experts.65.gate_proj", "model.layers.69.mlp.experts.66.gate_proj", "model.layers.69.mlp.experts.67.gate_proj", "model.layers.69.mlp.experts.68.gate_proj", "model.layers.69.mlp.experts.69.gate_proj", "model.layers.69.mlp.experts.70.gate_proj", "model.layers.69.mlp.experts.71.gate_proj", "model.layers.69.mlp.experts.72.gate_proj", "model.layers.69.mlp.experts.73.gate_proj", "model.layers.69.mlp.experts.74.gate_proj", "model.layers.69.mlp.experts.75.gate_proj", "model.layers.69.mlp.experts.76.gate_proj", "model.layers.69.mlp.experts.77.gate_proj", "model.layers.69.mlp.experts.78.gate_proj", "model.layers.69.mlp.experts.79.gate_proj", "model.layers.69.mlp.experts.80.gate_proj", "model.layers.69.mlp.experts.81.gate_proj", "model.layers.69.mlp.experts.82.gate_proj", "model.layers.69.mlp.experts.83.gate_proj", "model.layers.69.mlp.experts.84.gate_proj", "model.layers.69.mlp.experts.85.gate_proj", "model.layers.69.mlp.experts.86.gate_proj", "model.layers.69.mlp.experts.87.gate_proj", "model.layers.69.mlp.experts.88.gate_proj", "model.layers.69.mlp.experts.89.gate_proj", "model.layers.69.mlp.experts.90.gate_proj", "model.layers.69.mlp.experts.91.gate_proj", "model.layers.69.mlp.experts.92.gate_proj", "model.layers.69.mlp.experts.93.gate_proj", "model.layers.69.mlp.experts.94.gate_proj", "model.layers.69.mlp.experts.95.gate_proj", "model.layers.69.mlp.experts.96.gate_proj", "model.layers.69.mlp.experts.97.gate_proj", "model.layers.69.mlp.experts.98.gate_proj", "model.layers.69.mlp.experts.99.gate_proj", "model.layers.69.mlp.experts.100.gate_proj", "model.layers.69.mlp.experts.101.gate_proj", "model.layers.69.mlp.experts.102.gate_proj", "model.layers.69.mlp.experts.103.gate_proj", "model.layers.69.mlp.experts.104.gate_proj", "model.layers.69.mlp.experts.105.gate_proj", "model.layers.69.mlp.experts.106.gate_proj", "model.layers.69.mlp.experts.107.gate_proj", "model.layers.69.mlp.experts.108.gate_proj", "model.layers.69.mlp.experts.109.gate_proj", "model.layers.69.mlp.experts.110.gate_proj", "model.layers.69.mlp.experts.111.gate_proj", "model.layers.69.mlp.experts.112.gate_proj", "model.layers.69.mlp.experts.113.gate_proj", "model.layers.69.mlp.experts.114.gate_proj", "model.layers.69.mlp.experts.115.gate_proj", "model.layers.69.mlp.experts.116.gate_proj", "model.layers.69.mlp.experts.117.gate_proj", "model.layers.69.mlp.experts.118.gate_proj", "model.layers.69.mlp.experts.119.gate_proj", "model.layers.69.mlp.experts.120.gate_proj", "model.layers.69.mlp.experts.121.gate_proj", "model.layers.69.mlp.experts.122.gate_proj", "model.layers.69.mlp.experts.123.gate_proj", "model.layers.69.mlp.experts.124.gate_proj", "model.layers.69.mlp.experts.125.gate_proj", "model.layers.69.mlp.experts.126.gate_proj", "model.layers.69.mlp.experts.127.gate_proj", "model.layers.69.mlp.experts.128.gate_proj", "model.layers.69.mlp.experts.129.gate_proj", "model.layers.69.mlp.experts.130.gate_proj", "model.layers.69.mlp.experts.131.gate_proj", "model.layers.69.mlp.experts.132.gate_proj", "model.layers.69.mlp.experts.133.gate_proj", "model.layers.69.mlp.experts.134.gate_proj", "model.layers.69.mlp.experts.135.gate_proj", "model.layers.69.mlp.experts.136.gate_proj", "model.layers.69.mlp.experts.137.gate_proj", "model.layers.69.mlp.experts.138.gate_proj", "model.layers.69.mlp.experts.139.gate_proj", "model.layers.69.mlp.experts.140.gate_proj", "model.layers.69.mlp.experts.141.gate_proj", "model.layers.69.mlp.experts.142.gate_proj", "model.layers.69.mlp.experts.143.gate_proj", "model.layers.69.mlp.experts.144.gate_proj", "model.layers.69.mlp.experts.145.gate_proj", "model.layers.69.mlp.experts.146.gate_proj", "model.layers.69.mlp.experts.147.gate_proj", "model.layers.69.mlp.experts.148.gate_proj", "model.layers.69.mlp.experts.149.gate_proj", "model.layers.69.mlp.experts.150.gate_proj", "model.layers.69.mlp.experts.151.gate_proj", "model.layers.69.mlp.experts.152.gate_proj", "model.layers.69.mlp.experts.153.gate_proj", "model.layers.69.mlp.experts.154.gate_proj", "model.layers.69.mlp.experts.155.gate_proj", "model.layers.69.mlp.experts.156.gate_proj", "model.layers.69.mlp.experts.157.gate_proj", "model.layers.69.mlp.experts.158.gate_proj", "model.layers.69.mlp.experts.159.gate_proj", "model.layers.69.mlp.experts.0.up_proj", "model.layers.69.mlp.experts.1.up_proj", "model.layers.69.mlp.experts.2.up_proj", "model.layers.69.mlp.experts.3.up_proj", "model.layers.69.mlp.experts.4.up_proj", "model.layers.69.mlp.experts.5.up_proj", "model.layers.69.mlp.experts.6.up_proj", "model.layers.69.mlp.experts.7.up_proj", "model.layers.69.mlp.experts.8.up_proj", "model.layers.69.mlp.experts.9.up_proj", "model.layers.69.mlp.experts.10.up_proj", "model.layers.69.mlp.experts.11.up_proj", "model.layers.69.mlp.experts.12.up_proj", "model.layers.69.mlp.experts.13.up_proj", "model.layers.69.mlp.experts.14.up_proj", "model.layers.69.mlp.experts.15.up_proj", "model.layers.69.mlp.experts.16.up_proj", "model.layers.69.mlp.experts.17.up_proj", "model.layers.69.mlp.experts.18.up_proj", "model.layers.69.mlp.experts.19.up_proj", "model.layers.69.mlp.experts.20.up_proj", "model.layers.69.mlp.experts.21.up_proj", "model.layers.69.mlp.experts.22.up_proj", "model.layers.69.mlp.experts.23.up_proj", "model.layers.69.mlp.experts.24.up_proj", "model.layers.69.mlp.experts.25.up_proj", "model.layers.69.mlp.experts.26.up_proj", "model.layers.69.mlp.experts.27.up_proj", "model.layers.69.mlp.experts.28.up_proj", "model.layers.69.mlp.experts.29.up_proj", "model.layers.69.mlp.experts.30.up_proj", "model.layers.69.mlp.experts.31.up_proj", "model.layers.69.mlp.experts.32.up_proj", "model.layers.69.mlp.experts.33.up_proj", "model.layers.69.mlp.experts.34.up_proj", "model.layers.69.mlp.experts.35.up_proj", "model.layers.69.mlp.experts.36.up_proj", "model.layers.69.mlp.experts.37.up_proj", "model.layers.69.mlp.experts.38.up_proj", "model.layers.69.mlp.experts.39.up_proj", "model.layers.69.mlp.experts.40.up_proj", "model.layers.69.mlp.experts.41.up_proj", "model.layers.69.mlp.experts.42.up_proj", "model.layers.69.mlp.experts.43.up_proj", "model.layers.69.mlp.experts.44.up_proj", "model.layers.69.mlp.experts.45.up_proj", "model.layers.69.mlp.experts.46.up_proj", "model.layers.69.mlp.experts.47.up_proj", "model.layers.69.mlp.experts.48.up_proj", "model.layers.69.mlp.experts.49.up_proj", "model.layers.69.mlp.experts.50.up_proj", "model.layers.69.mlp.experts.51.up_proj", "model.layers.69.mlp.experts.52.up_proj", "model.layers.69.mlp.experts.53.up_proj", "model.layers.69.mlp.experts.54.up_proj", "model.layers.69.mlp.experts.55.up_proj", "model.layers.69.mlp.experts.56.up_proj", "model.layers.69.mlp.experts.57.up_proj", "model.layers.69.mlp.experts.58.up_proj", "model.layers.69.mlp.experts.59.up_proj", "model.layers.69.mlp.experts.60.up_proj", "model.layers.69.mlp.experts.61.up_proj", "model.layers.69.mlp.experts.62.up_proj", "model.layers.69.mlp.experts.63.up_proj", "model.layers.69.mlp.experts.64.up_proj", "model.layers.69.mlp.experts.65.up_proj", "model.layers.69.mlp.experts.66.up_proj", "model.layers.69.mlp.experts.67.up_proj", "model.layers.69.mlp.experts.68.up_proj", "model.layers.69.mlp.experts.69.up_proj", "model.layers.69.mlp.experts.70.up_proj", "model.layers.69.mlp.experts.71.up_proj", "model.layers.69.mlp.experts.72.up_proj", "model.layers.69.mlp.experts.73.up_proj", "model.layers.69.mlp.experts.74.up_proj", "model.layers.69.mlp.experts.75.up_proj", "model.layers.69.mlp.experts.76.up_proj", "model.layers.69.mlp.experts.77.up_proj", "model.layers.69.mlp.experts.78.up_proj", "model.layers.69.mlp.experts.79.up_proj", "model.layers.69.mlp.experts.80.up_proj", "model.layers.69.mlp.experts.81.up_proj", "model.layers.69.mlp.experts.82.up_proj", "model.layers.69.mlp.experts.83.up_proj", "model.layers.69.mlp.experts.84.up_proj", "model.layers.69.mlp.experts.85.up_proj", "model.layers.69.mlp.experts.86.up_proj", "model.layers.69.mlp.experts.87.up_proj", "model.layers.69.mlp.experts.88.up_proj", "model.layers.69.mlp.experts.89.up_proj", "model.layers.69.mlp.experts.90.up_proj", "model.layers.69.mlp.experts.91.up_proj", "model.layers.69.mlp.experts.92.up_proj", "model.layers.69.mlp.experts.93.up_proj", "model.layers.69.mlp.experts.94.up_proj", "model.layers.69.mlp.experts.95.up_proj", "model.layers.69.mlp.experts.96.up_proj", "model.layers.69.mlp.experts.97.up_proj", "model.layers.69.mlp.experts.98.up_proj", "model.layers.69.mlp.experts.99.up_proj", "model.layers.69.mlp.experts.100.up_proj", "model.layers.69.mlp.experts.101.up_proj", "model.layers.69.mlp.experts.102.up_proj", "model.layers.69.mlp.experts.103.up_proj", "model.layers.69.mlp.experts.104.up_proj", "model.layers.69.mlp.experts.105.up_proj", "model.layers.69.mlp.experts.106.up_proj", "model.layers.69.mlp.experts.107.up_proj", "model.layers.69.mlp.experts.108.up_proj", "model.layers.69.mlp.experts.109.up_proj", "model.layers.69.mlp.experts.110.up_proj", "model.layers.69.mlp.experts.111.up_proj", "model.layers.69.mlp.experts.112.up_proj", "model.layers.69.mlp.experts.113.up_proj", "model.layers.69.mlp.experts.114.up_proj", "model.layers.69.mlp.experts.115.up_proj", "model.layers.69.mlp.experts.116.up_proj", "model.layers.69.mlp.experts.117.up_proj", "model.layers.69.mlp.experts.118.up_proj", "model.layers.69.mlp.experts.119.up_proj", "model.layers.69.mlp.experts.120.up_proj", "model.layers.69.mlp.experts.121.up_proj", "model.layers.69.mlp.experts.122.up_proj", "model.layers.69.mlp.experts.123.up_proj", "model.layers.69.mlp.experts.124.up_proj", "model.layers.69.mlp.experts.125.up_proj", "model.layers.69.mlp.experts.126.up_proj", "model.layers.69.mlp.experts.127.up_proj", "model.layers.69.mlp.experts.128.up_proj", "model.layers.69.mlp.experts.129.up_proj", "model.layers.69.mlp.experts.130.up_proj", "model.layers.69.mlp.experts.131.up_proj", "model.layers.69.mlp.experts.132.up_proj", "model.layers.69.mlp.experts.133.up_proj", "model.layers.69.mlp.experts.134.up_proj", "model.layers.69.mlp.experts.135.up_proj", "model.layers.69.mlp.experts.136.up_proj", "model.layers.69.mlp.experts.137.up_proj", "model.layers.69.mlp.experts.138.up_proj", "model.layers.69.mlp.experts.139.up_proj", "model.layers.69.mlp.experts.140.up_proj", "model.layers.69.mlp.experts.141.up_proj", "model.layers.69.mlp.experts.142.up_proj", "model.layers.69.mlp.experts.143.up_proj", "model.layers.69.mlp.experts.144.up_proj", "model.layers.69.mlp.experts.145.up_proj", "model.layers.69.mlp.experts.146.up_proj", "model.layers.69.mlp.experts.147.up_proj", "model.layers.69.mlp.experts.148.up_proj", "model.layers.69.mlp.experts.149.up_proj", "model.layers.69.mlp.experts.150.up_proj", "model.layers.69.mlp.experts.151.up_proj", "model.layers.69.mlp.experts.152.up_proj", "model.layers.69.mlp.experts.153.up_proj", "model.layers.69.mlp.experts.154.up_proj", "model.layers.69.mlp.experts.155.up_proj", "model.layers.69.mlp.experts.156.up_proj", "model.layers.69.mlp.experts.157.up_proj", "model.layers.69.mlp.experts.158.up_proj", "model.layers.69.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00024588126689195633, "dbits": 2516582400 }, { "dkld": -0.00029598549008369446, "dbits": 5033164800 }, { "dkld": -0.00031067859381438134, "dbits": 7549747200 }, { "dkld": -0.00030775927007198334, "dbits": 12582912000 } ] }, { "idx": 416, "layers": [ "model.layers.69.mlp.experts.0.down_proj", "model.layers.69.mlp.experts.1.down_proj", "model.layers.69.mlp.experts.2.down_proj", "model.layers.69.mlp.experts.3.down_proj", "model.layers.69.mlp.experts.4.down_proj", "model.layers.69.mlp.experts.5.down_proj", "model.layers.69.mlp.experts.6.down_proj", "model.layers.69.mlp.experts.7.down_proj", "model.layers.69.mlp.experts.8.down_proj", "model.layers.69.mlp.experts.9.down_proj", "model.layers.69.mlp.experts.10.down_proj", "model.layers.69.mlp.experts.11.down_proj", "model.layers.69.mlp.experts.12.down_proj", "model.layers.69.mlp.experts.13.down_proj", "model.layers.69.mlp.experts.14.down_proj", "model.layers.69.mlp.experts.15.down_proj", "model.layers.69.mlp.experts.16.down_proj", "model.layers.69.mlp.experts.17.down_proj", "model.layers.69.mlp.experts.18.down_proj", "model.layers.69.mlp.experts.19.down_proj", "model.layers.69.mlp.experts.20.down_proj", "model.layers.69.mlp.experts.21.down_proj", "model.layers.69.mlp.experts.22.down_proj", "model.layers.69.mlp.experts.23.down_proj", "model.layers.69.mlp.experts.24.down_proj", "model.layers.69.mlp.experts.25.down_proj", "model.layers.69.mlp.experts.26.down_proj", "model.layers.69.mlp.experts.27.down_proj", "model.layers.69.mlp.experts.28.down_proj", "model.layers.69.mlp.experts.29.down_proj", "model.layers.69.mlp.experts.30.down_proj", "model.layers.69.mlp.experts.31.down_proj", "model.layers.69.mlp.experts.32.down_proj", "model.layers.69.mlp.experts.33.down_proj", "model.layers.69.mlp.experts.34.down_proj", "model.layers.69.mlp.experts.35.down_proj", "model.layers.69.mlp.experts.36.down_proj", "model.layers.69.mlp.experts.37.down_proj", "model.layers.69.mlp.experts.38.down_proj", "model.layers.69.mlp.experts.39.down_proj", "model.layers.69.mlp.experts.40.down_proj", "model.layers.69.mlp.experts.41.down_proj", "model.layers.69.mlp.experts.42.down_proj", "model.layers.69.mlp.experts.43.down_proj", "model.layers.69.mlp.experts.44.down_proj", "model.layers.69.mlp.experts.45.down_proj", "model.layers.69.mlp.experts.46.down_proj", "model.layers.69.mlp.experts.47.down_proj", "model.layers.69.mlp.experts.48.down_proj", "model.layers.69.mlp.experts.49.down_proj", "model.layers.69.mlp.experts.50.down_proj", "model.layers.69.mlp.experts.51.down_proj", "model.layers.69.mlp.experts.52.down_proj", "model.layers.69.mlp.experts.53.down_proj", "model.layers.69.mlp.experts.54.down_proj", "model.layers.69.mlp.experts.55.down_proj", "model.layers.69.mlp.experts.56.down_proj", "model.layers.69.mlp.experts.57.down_proj", "model.layers.69.mlp.experts.58.down_proj", "model.layers.69.mlp.experts.59.down_proj", "model.layers.69.mlp.experts.60.down_proj", "model.layers.69.mlp.experts.61.down_proj", "model.layers.69.mlp.experts.62.down_proj", "model.layers.69.mlp.experts.63.down_proj", "model.layers.69.mlp.experts.64.down_proj", "model.layers.69.mlp.experts.65.down_proj", "model.layers.69.mlp.experts.66.down_proj", "model.layers.69.mlp.experts.67.down_proj", "model.layers.69.mlp.experts.68.down_proj", "model.layers.69.mlp.experts.69.down_proj", "model.layers.69.mlp.experts.70.down_proj", "model.layers.69.mlp.experts.71.down_proj", "model.layers.69.mlp.experts.72.down_proj", "model.layers.69.mlp.experts.73.down_proj", "model.layers.69.mlp.experts.74.down_proj", "model.layers.69.mlp.experts.75.down_proj", "model.layers.69.mlp.experts.76.down_proj", "model.layers.69.mlp.experts.77.down_proj", "model.layers.69.mlp.experts.78.down_proj", "model.layers.69.mlp.experts.79.down_proj", "model.layers.69.mlp.experts.80.down_proj", "model.layers.69.mlp.experts.81.down_proj", "model.layers.69.mlp.experts.82.down_proj", "model.layers.69.mlp.experts.83.down_proj", "model.layers.69.mlp.experts.84.down_proj", "model.layers.69.mlp.experts.85.down_proj", "model.layers.69.mlp.experts.86.down_proj", "model.layers.69.mlp.experts.87.down_proj", "model.layers.69.mlp.experts.88.down_proj", "model.layers.69.mlp.experts.89.down_proj", "model.layers.69.mlp.experts.90.down_proj", "model.layers.69.mlp.experts.91.down_proj", "model.layers.69.mlp.experts.92.down_proj", "model.layers.69.mlp.experts.93.down_proj", "model.layers.69.mlp.experts.94.down_proj", "model.layers.69.mlp.experts.95.down_proj", "model.layers.69.mlp.experts.96.down_proj", "model.layers.69.mlp.experts.97.down_proj", "model.layers.69.mlp.experts.98.down_proj", "model.layers.69.mlp.experts.99.down_proj", "model.layers.69.mlp.experts.100.down_proj", "model.layers.69.mlp.experts.101.down_proj", "model.layers.69.mlp.experts.102.down_proj", "model.layers.69.mlp.experts.103.down_proj", "model.layers.69.mlp.experts.104.down_proj", "model.layers.69.mlp.experts.105.down_proj", "model.layers.69.mlp.experts.106.down_proj", "model.layers.69.mlp.experts.107.down_proj", "model.layers.69.mlp.experts.108.down_proj", "model.layers.69.mlp.experts.109.down_proj", "model.layers.69.mlp.experts.110.down_proj", "model.layers.69.mlp.experts.111.down_proj", "model.layers.69.mlp.experts.112.down_proj", "model.layers.69.mlp.experts.113.down_proj", "model.layers.69.mlp.experts.114.down_proj", "model.layers.69.mlp.experts.115.down_proj", "model.layers.69.mlp.experts.116.down_proj", "model.layers.69.mlp.experts.117.down_proj", "model.layers.69.mlp.experts.118.down_proj", "model.layers.69.mlp.experts.119.down_proj", "model.layers.69.mlp.experts.120.down_proj", "model.layers.69.mlp.experts.121.down_proj", "model.layers.69.mlp.experts.122.down_proj", "model.layers.69.mlp.experts.123.down_proj", "model.layers.69.mlp.experts.124.down_proj", "model.layers.69.mlp.experts.125.down_proj", "model.layers.69.mlp.experts.126.down_proj", "model.layers.69.mlp.experts.127.down_proj", "model.layers.69.mlp.experts.128.down_proj", "model.layers.69.mlp.experts.129.down_proj", "model.layers.69.mlp.experts.130.down_proj", "model.layers.69.mlp.experts.131.down_proj", "model.layers.69.mlp.experts.132.down_proj", "model.layers.69.mlp.experts.133.down_proj", "model.layers.69.mlp.experts.134.down_proj", "model.layers.69.mlp.experts.135.down_proj", "model.layers.69.mlp.experts.136.down_proj", "model.layers.69.mlp.experts.137.down_proj", "model.layers.69.mlp.experts.138.down_proj", "model.layers.69.mlp.experts.139.down_proj", "model.layers.69.mlp.experts.140.down_proj", "model.layers.69.mlp.experts.141.down_proj", "model.layers.69.mlp.experts.142.down_proj", "model.layers.69.mlp.experts.143.down_proj", "model.layers.69.mlp.experts.144.down_proj", "model.layers.69.mlp.experts.145.down_proj", "model.layers.69.mlp.experts.146.down_proj", "model.layers.69.mlp.experts.147.down_proj", "model.layers.69.mlp.experts.148.down_proj", "model.layers.69.mlp.experts.149.down_proj", "model.layers.69.mlp.experts.150.down_proj", "model.layers.69.mlp.experts.151.down_proj", "model.layers.69.mlp.experts.152.down_proj", "model.layers.69.mlp.experts.153.down_proj", "model.layers.69.mlp.experts.154.down_proj", "model.layers.69.mlp.experts.155.down_proj", "model.layers.69.mlp.experts.156.down_proj", "model.layers.69.mlp.experts.157.down_proj", "model.layers.69.mlp.experts.158.down_proj", "model.layers.69.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": 2.337284386157157e-05, "dbits": 1258291200 }, { "dkld": -6.443206220865805e-05, "dbits": 2516582400 }, { "dkld": -1.9518844783308897e-05, "dbits": 3774873600 }, { "dkld": -4.2877066880470105e-05, "dbits": 6291456000 } ] }, { "idx": 417, "layers": [ "model.layers.70.self_attn.q_proj" ], "candidates": [ { "dkld": 2.096444368361594e-05, "dbits": 62914560 }, { "dkld": 3.3608358353365286e-05, "dbits": 125829120 }, { "dkld": 2.6759505271908846e-05, "dbits": 188743680 }, { "dkld": -2.642460167408267e-05, "dbits": 314572800 } ] }, { "idx": 418, "layers": [ "model.layers.70.self_attn.k_proj", "model.layers.70.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00014342032372952063, "dbits": 10485760 }, { "dkld": -0.0001100318506360165, "dbits": 20971520 }, { "dkld": -0.0001387738622725121, "dbits": 31457280 }, { "dkld": -0.00016121920198203243, "dbits": 52428800 } ] }, { "idx": 419, "layers": [ "model.layers.70.self_attn.o_proj" ], "candidates": [ { "dkld": -8.212774991994687e-06, "dbits": 62914560 }, { "dkld": -3.368724137545187e-05, "dbits": 125829120 }, { "dkld": -5.064830183983127e-05, "dbits": 188743680 }, { "dkld": -7.184529677034535e-05, "dbits": 314572800 } ] }, { "idx": 420, "layers": [ "model.layers.70.mlp.shared_experts.gate_proj", "model.layers.70.mlp.shared_experts.up_proj", "model.layers.70.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0003463792614638861, "dbits": 23592960 }, { "dkld": -0.0002598459832370281, "dbits": 47185920 }, { "dkld": -0.0002510666847229115, "dbits": 70778880 }, { "dkld": -0.0002918002195656383, "dbits": 117964800 } ] }, { "idx": 421, "layers": [ "model.layers.70.mlp.experts.0.gate_proj", "model.layers.70.mlp.experts.1.gate_proj", "model.layers.70.mlp.experts.2.gate_proj", "model.layers.70.mlp.experts.3.gate_proj", "model.layers.70.mlp.experts.4.gate_proj", "model.layers.70.mlp.experts.5.gate_proj", "model.layers.70.mlp.experts.6.gate_proj", "model.layers.70.mlp.experts.7.gate_proj", "model.layers.70.mlp.experts.8.gate_proj", "model.layers.70.mlp.experts.9.gate_proj", "model.layers.70.mlp.experts.10.gate_proj", "model.layers.70.mlp.experts.11.gate_proj", "model.layers.70.mlp.experts.12.gate_proj", "model.layers.70.mlp.experts.13.gate_proj", "model.layers.70.mlp.experts.14.gate_proj", "model.layers.70.mlp.experts.15.gate_proj", "model.layers.70.mlp.experts.16.gate_proj", "model.layers.70.mlp.experts.17.gate_proj", "model.layers.70.mlp.experts.18.gate_proj", "model.layers.70.mlp.experts.19.gate_proj", "model.layers.70.mlp.experts.20.gate_proj", "model.layers.70.mlp.experts.21.gate_proj", "model.layers.70.mlp.experts.22.gate_proj", "model.layers.70.mlp.experts.23.gate_proj", "model.layers.70.mlp.experts.24.gate_proj", "model.layers.70.mlp.experts.25.gate_proj", "model.layers.70.mlp.experts.26.gate_proj", "model.layers.70.mlp.experts.27.gate_proj", "model.layers.70.mlp.experts.28.gate_proj", "model.layers.70.mlp.experts.29.gate_proj", "model.layers.70.mlp.experts.30.gate_proj", "model.layers.70.mlp.experts.31.gate_proj", "model.layers.70.mlp.experts.32.gate_proj", "model.layers.70.mlp.experts.33.gate_proj", "model.layers.70.mlp.experts.34.gate_proj", "model.layers.70.mlp.experts.35.gate_proj", "model.layers.70.mlp.experts.36.gate_proj", "model.layers.70.mlp.experts.37.gate_proj", "model.layers.70.mlp.experts.38.gate_proj", "model.layers.70.mlp.experts.39.gate_proj", "model.layers.70.mlp.experts.40.gate_proj", "model.layers.70.mlp.experts.41.gate_proj", "model.layers.70.mlp.experts.42.gate_proj", "model.layers.70.mlp.experts.43.gate_proj", "model.layers.70.mlp.experts.44.gate_proj", "model.layers.70.mlp.experts.45.gate_proj", "model.layers.70.mlp.experts.46.gate_proj", "model.layers.70.mlp.experts.47.gate_proj", "model.layers.70.mlp.experts.48.gate_proj", "model.layers.70.mlp.experts.49.gate_proj", "model.layers.70.mlp.experts.50.gate_proj", "model.layers.70.mlp.experts.51.gate_proj", "model.layers.70.mlp.experts.52.gate_proj", "model.layers.70.mlp.experts.53.gate_proj", "model.layers.70.mlp.experts.54.gate_proj", "model.layers.70.mlp.experts.55.gate_proj", "model.layers.70.mlp.experts.56.gate_proj", "model.layers.70.mlp.experts.57.gate_proj", "model.layers.70.mlp.experts.58.gate_proj", "model.layers.70.mlp.experts.59.gate_proj", "model.layers.70.mlp.experts.60.gate_proj", "model.layers.70.mlp.experts.61.gate_proj", "model.layers.70.mlp.experts.62.gate_proj", "model.layers.70.mlp.experts.63.gate_proj", "model.layers.70.mlp.experts.64.gate_proj", "model.layers.70.mlp.experts.65.gate_proj", "model.layers.70.mlp.experts.66.gate_proj", "model.layers.70.mlp.experts.67.gate_proj", "model.layers.70.mlp.experts.68.gate_proj", "model.layers.70.mlp.experts.69.gate_proj", "model.layers.70.mlp.experts.70.gate_proj", "model.layers.70.mlp.experts.71.gate_proj", "model.layers.70.mlp.experts.72.gate_proj", "model.layers.70.mlp.experts.73.gate_proj", "model.layers.70.mlp.experts.74.gate_proj", "model.layers.70.mlp.experts.75.gate_proj", "model.layers.70.mlp.experts.76.gate_proj", "model.layers.70.mlp.experts.77.gate_proj", "model.layers.70.mlp.experts.78.gate_proj", "model.layers.70.mlp.experts.79.gate_proj", "model.layers.70.mlp.experts.80.gate_proj", "model.layers.70.mlp.experts.81.gate_proj", "model.layers.70.mlp.experts.82.gate_proj", "model.layers.70.mlp.experts.83.gate_proj", "model.layers.70.mlp.experts.84.gate_proj", "model.layers.70.mlp.experts.85.gate_proj", "model.layers.70.mlp.experts.86.gate_proj", "model.layers.70.mlp.experts.87.gate_proj", "model.layers.70.mlp.experts.88.gate_proj", "model.layers.70.mlp.experts.89.gate_proj", "model.layers.70.mlp.experts.90.gate_proj", "model.layers.70.mlp.experts.91.gate_proj", "model.layers.70.mlp.experts.92.gate_proj", "model.layers.70.mlp.experts.93.gate_proj", "model.layers.70.mlp.experts.94.gate_proj", "model.layers.70.mlp.experts.95.gate_proj", "model.layers.70.mlp.experts.96.gate_proj", "model.layers.70.mlp.experts.97.gate_proj", "model.layers.70.mlp.experts.98.gate_proj", "model.layers.70.mlp.experts.99.gate_proj", "model.layers.70.mlp.experts.100.gate_proj", "model.layers.70.mlp.experts.101.gate_proj", "model.layers.70.mlp.experts.102.gate_proj", "model.layers.70.mlp.experts.103.gate_proj", "model.layers.70.mlp.experts.104.gate_proj", "model.layers.70.mlp.experts.105.gate_proj", "model.layers.70.mlp.experts.106.gate_proj", "model.layers.70.mlp.experts.107.gate_proj", "model.layers.70.mlp.experts.108.gate_proj", "model.layers.70.mlp.experts.109.gate_proj", "model.layers.70.mlp.experts.110.gate_proj", "model.layers.70.mlp.experts.111.gate_proj", "model.layers.70.mlp.experts.112.gate_proj", "model.layers.70.mlp.experts.113.gate_proj", "model.layers.70.mlp.experts.114.gate_proj", "model.layers.70.mlp.experts.115.gate_proj", "model.layers.70.mlp.experts.116.gate_proj", "model.layers.70.mlp.experts.117.gate_proj", "model.layers.70.mlp.experts.118.gate_proj", "model.layers.70.mlp.experts.119.gate_proj", "model.layers.70.mlp.experts.120.gate_proj", "model.layers.70.mlp.experts.121.gate_proj", "model.layers.70.mlp.experts.122.gate_proj", "model.layers.70.mlp.experts.123.gate_proj", "model.layers.70.mlp.experts.124.gate_proj", "model.layers.70.mlp.experts.125.gate_proj", "model.layers.70.mlp.experts.126.gate_proj", "model.layers.70.mlp.experts.127.gate_proj", "model.layers.70.mlp.experts.128.gate_proj", "model.layers.70.mlp.experts.129.gate_proj", "model.layers.70.mlp.experts.130.gate_proj", "model.layers.70.mlp.experts.131.gate_proj", "model.layers.70.mlp.experts.132.gate_proj", "model.layers.70.mlp.experts.133.gate_proj", "model.layers.70.mlp.experts.134.gate_proj", "model.layers.70.mlp.experts.135.gate_proj", "model.layers.70.mlp.experts.136.gate_proj", "model.layers.70.mlp.experts.137.gate_proj", "model.layers.70.mlp.experts.138.gate_proj", "model.layers.70.mlp.experts.139.gate_proj", "model.layers.70.mlp.experts.140.gate_proj", "model.layers.70.mlp.experts.141.gate_proj", "model.layers.70.mlp.experts.142.gate_proj", "model.layers.70.mlp.experts.143.gate_proj", "model.layers.70.mlp.experts.144.gate_proj", "model.layers.70.mlp.experts.145.gate_proj", "model.layers.70.mlp.experts.146.gate_proj", "model.layers.70.mlp.experts.147.gate_proj", "model.layers.70.mlp.experts.148.gate_proj", "model.layers.70.mlp.experts.149.gate_proj", "model.layers.70.mlp.experts.150.gate_proj", "model.layers.70.mlp.experts.151.gate_proj", "model.layers.70.mlp.experts.152.gate_proj", "model.layers.70.mlp.experts.153.gate_proj", "model.layers.70.mlp.experts.154.gate_proj", "model.layers.70.mlp.experts.155.gate_proj", "model.layers.70.mlp.experts.156.gate_proj", "model.layers.70.mlp.experts.157.gate_proj", "model.layers.70.mlp.experts.158.gate_proj", "model.layers.70.mlp.experts.159.gate_proj", "model.layers.70.mlp.experts.0.up_proj", "model.layers.70.mlp.experts.1.up_proj", "model.layers.70.mlp.experts.2.up_proj", "model.layers.70.mlp.experts.3.up_proj", "model.layers.70.mlp.experts.4.up_proj", "model.layers.70.mlp.experts.5.up_proj", "model.layers.70.mlp.experts.6.up_proj", "model.layers.70.mlp.experts.7.up_proj", "model.layers.70.mlp.experts.8.up_proj", "model.layers.70.mlp.experts.9.up_proj", "model.layers.70.mlp.experts.10.up_proj", "model.layers.70.mlp.experts.11.up_proj", "model.layers.70.mlp.experts.12.up_proj", "model.layers.70.mlp.experts.13.up_proj", "model.layers.70.mlp.experts.14.up_proj", "model.layers.70.mlp.experts.15.up_proj", "model.layers.70.mlp.experts.16.up_proj", "model.layers.70.mlp.experts.17.up_proj", "model.layers.70.mlp.experts.18.up_proj", "model.layers.70.mlp.experts.19.up_proj", "model.layers.70.mlp.experts.20.up_proj", "model.layers.70.mlp.experts.21.up_proj", "model.layers.70.mlp.experts.22.up_proj", "model.layers.70.mlp.experts.23.up_proj", "model.layers.70.mlp.experts.24.up_proj", "model.layers.70.mlp.experts.25.up_proj", "model.layers.70.mlp.experts.26.up_proj", "model.layers.70.mlp.experts.27.up_proj", "model.layers.70.mlp.experts.28.up_proj", "model.layers.70.mlp.experts.29.up_proj", "model.layers.70.mlp.experts.30.up_proj", "model.layers.70.mlp.experts.31.up_proj", "model.layers.70.mlp.experts.32.up_proj", "model.layers.70.mlp.experts.33.up_proj", "model.layers.70.mlp.experts.34.up_proj", "model.layers.70.mlp.experts.35.up_proj", "model.layers.70.mlp.experts.36.up_proj", "model.layers.70.mlp.experts.37.up_proj", "model.layers.70.mlp.experts.38.up_proj", "model.layers.70.mlp.experts.39.up_proj", "model.layers.70.mlp.experts.40.up_proj", "model.layers.70.mlp.experts.41.up_proj", "model.layers.70.mlp.experts.42.up_proj", "model.layers.70.mlp.experts.43.up_proj", "model.layers.70.mlp.experts.44.up_proj", "model.layers.70.mlp.experts.45.up_proj", "model.layers.70.mlp.experts.46.up_proj", "model.layers.70.mlp.experts.47.up_proj", "model.layers.70.mlp.experts.48.up_proj", "model.layers.70.mlp.experts.49.up_proj", "model.layers.70.mlp.experts.50.up_proj", "model.layers.70.mlp.experts.51.up_proj", "model.layers.70.mlp.experts.52.up_proj", "model.layers.70.mlp.experts.53.up_proj", "model.layers.70.mlp.experts.54.up_proj", "model.layers.70.mlp.experts.55.up_proj", "model.layers.70.mlp.experts.56.up_proj", "model.layers.70.mlp.experts.57.up_proj", "model.layers.70.mlp.experts.58.up_proj", "model.layers.70.mlp.experts.59.up_proj", "model.layers.70.mlp.experts.60.up_proj", "model.layers.70.mlp.experts.61.up_proj", "model.layers.70.mlp.experts.62.up_proj", "model.layers.70.mlp.experts.63.up_proj", "model.layers.70.mlp.experts.64.up_proj", "model.layers.70.mlp.experts.65.up_proj", "model.layers.70.mlp.experts.66.up_proj", "model.layers.70.mlp.experts.67.up_proj", "model.layers.70.mlp.experts.68.up_proj", "model.layers.70.mlp.experts.69.up_proj", "model.layers.70.mlp.experts.70.up_proj", "model.layers.70.mlp.experts.71.up_proj", "model.layers.70.mlp.experts.72.up_proj", "model.layers.70.mlp.experts.73.up_proj", "model.layers.70.mlp.experts.74.up_proj", "model.layers.70.mlp.experts.75.up_proj", "model.layers.70.mlp.experts.76.up_proj", "model.layers.70.mlp.experts.77.up_proj", "model.layers.70.mlp.experts.78.up_proj", "model.layers.70.mlp.experts.79.up_proj", "model.layers.70.mlp.experts.80.up_proj", "model.layers.70.mlp.experts.81.up_proj", "model.layers.70.mlp.experts.82.up_proj", "model.layers.70.mlp.experts.83.up_proj", "model.layers.70.mlp.experts.84.up_proj", "model.layers.70.mlp.experts.85.up_proj", "model.layers.70.mlp.experts.86.up_proj", "model.layers.70.mlp.experts.87.up_proj", "model.layers.70.mlp.experts.88.up_proj", "model.layers.70.mlp.experts.89.up_proj", "model.layers.70.mlp.experts.90.up_proj", "model.layers.70.mlp.experts.91.up_proj", "model.layers.70.mlp.experts.92.up_proj", "model.layers.70.mlp.experts.93.up_proj", "model.layers.70.mlp.experts.94.up_proj", "model.layers.70.mlp.experts.95.up_proj", "model.layers.70.mlp.experts.96.up_proj", "model.layers.70.mlp.experts.97.up_proj", "model.layers.70.mlp.experts.98.up_proj", "model.layers.70.mlp.experts.99.up_proj", "model.layers.70.mlp.experts.100.up_proj", "model.layers.70.mlp.experts.101.up_proj", "model.layers.70.mlp.experts.102.up_proj", "model.layers.70.mlp.experts.103.up_proj", "model.layers.70.mlp.experts.104.up_proj", "model.layers.70.mlp.experts.105.up_proj", "model.layers.70.mlp.experts.106.up_proj", "model.layers.70.mlp.experts.107.up_proj", "model.layers.70.mlp.experts.108.up_proj", "model.layers.70.mlp.experts.109.up_proj", "model.layers.70.mlp.experts.110.up_proj", "model.layers.70.mlp.experts.111.up_proj", "model.layers.70.mlp.experts.112.up_proj", "model.layers.70.mlp.experts.113.up_proj", "model.layers.70.mlp.experts.114.up_proj", "model.layers.70.mlp.experts.115.up_proj", "model.layers.70.mlp.experts.116.up_proj", "model.layers.70.mlp.experts.117.up_proj", "model.layers.70.mlp.experts.118.up_proj", "model.layers.70.mlp.experts.119.up_proj", "model.layers.70.mlp.experts.120.up_proj", "model.layers.70.mlp.experts.121.up_proj", "model.layers.70.mlp.experts.122.up_proj", "model.layers.70.mlp.experts.123.up_proj", "model.layers.70.mlp.experts.124.up_proj", "model.layers.70.mlp.experts.125.up_proj", "model.layers.70.mlp.experts.126.up_proj", "model.layers.70.mlp.experts.127.up_proj", "model.layers.70.mlp.experts.128.up_proj", "model.layers.70.mlp.experts.129.up_proj", "model.layers.70.mlp.experts.130.up_proj", "model.layers.70.mlp.experts.131.up_proj", "model.layers.70.mlp.experts.132.up_proj", "model.layers.70.mlp.experts.133.up_proj", "model.layers.70.mlp.experts.134.up_proj", "model.layers.70.mlp.experts.135.up_proj", "model.layers.70.mlp.experts.136.up_proj", "model.layers.70.mlp.experts.137.up_proj", "model.layers.70.mlp.experts.138.up_proj", "model.layers.70.mlp.experts.139.up_proj", "model.layers.70.mlp.experts.140.up_proj", "model.layers.70.mlp.experts.141.up_proj", "model.layers.70.mlp.experts.142.up_proj", "model.layers.70.mlp.experts.143.up_proj", "model.layers.70.mlp.experts.144.up_proj", "model.layers.70.mlp.experts.145.up_proj", "model.layers.70.mlp.experts.146.up_proj", "model.layers.70.mlp.experts.147.up_proj", "model.layers.70.mlp.experts.148.up_proj", "model.layers.70.mlp.experts.149.up_proj", "model.layers.70.mlp.experts.150.up_proj", "model.layers.70.mlp.experts.151.up_proj", "model.layers.70.mlp.experts.152.up_proj", "model.layers.70.mlp.experts.153.up_proj", "model.layers.70.mlp.experts.154.up_proj", "model.layers.70.mlp.experts.155.up_proj", "model.layers.70.mlp.experts.156.up_proj", "model.layers.70.mlp.experts.157.up_proj", "model.layers.70.mlp.experts.158.up_proj", "model.layers.70.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -7.972382009030499e-05, "dbits": 2516582400 }, { "dkld": -9.65408980846405e-05, "dbits": 5033164800 }, { "dkld": -2.5655236095198375e-05, "dbits": 7549747200 }, { "dkld": -5.110055208207287e-05, "dbits": 12582912000 } ] }, { "idx": 422, "layers": [ "model.layers.70.mlp.experts.0.down_proj", "model.layers.70.mlp.experts.1.down_proj", "model.layers.70.mlp.experts.2.down_proj", "model.layers.70.mlp.experts.3.down_proj", "model.layers.70.mlp.experts.4.down_proj", "model.layers.70.mlp.experts.5.down_proj", "model.layers.70.mlp.experts.6.down_proj", "model.layers.70.mlp.experts.7.down_proj", "model.layers.70.mlp.experts.8.down_proj", "model.layers.70.mlp.experts.9.down_proj", "model.layers.70.mlp.experts.10.down_proj", "model.layers.70.mlp.experts.11.down_proj", "model.layers.70.mlp.experts.12.down_proj", "model.layers.70.mlp.experts.13.down_proj", "model.layers.70.mlp.experts.14.down_proj", "model.layers.70.mlp.experts.15.down_proj", "model.layers.70.mlp.experts.16.down_proj", "model.layers.70.mlp.experts.17.down_proj", "model.layers.70.mlp.experts.18.down_proj", "model.layers.70.mlp.experts.19.down_proj", "model.layers.70.mlp.experts.20.down_proj", "model.layers.70.mlp.experts.21.down_proj", "model.layers.70.mlp.experts.22.down_proj", "model.layers.70.mlp.experts.23.down_proj", "model.layers.70.mlp.experts.24.down_proj", "model.layers.70.mlp.experts.25.down_proj", "model.layers.70.mlp.experts.26.down_proj", "model.layers.70.mlp.experts.27.down_proj", "model.layers.70.mlp.experts.28.down_proj", "model.layers.70.mlp.experts.29.down_proj", "model.layers.70.mlp.experts.30.down_proj", "model.layers.70.mlp.experts.31.down_proj", "model.layers.70.mlp.experts.32.down_proj", "model.layers.70.mlp.experts.33.down_proj", "model.layers.70.mlp.experts.34.down_proj", "model.layers.70.mlp.experts.35.down_proj", "model.layers.70.mlp.experts.36.down_proj", "model.layers.70.mlp.experts.37.down_proj", "model.layers.70.mlp.experts.38.down_proj", "model.layers.70.mlp.experts.39.down_proj", "model.layers.70.mlp.experts.40.down_proj", "model.layers.70.mlp.experts.41.down_proj", "model.layers.70.mlp.experts.42.down_proj", "model.layers.70.mlp.experts.43.down_proj", "model.layers.70.mlp.experts.44.down_proj", "model.layers.70.mlp.experts.45.down_proj", "model.layers.70.mlp.experts.46.down_proj", "model.layers.70.mlp.experts.47.down_proj", "model.layers.70.mlp.experts.48.down_proj", "model.layers.70.mlp.experts.49.down_proj", "model.layers.70.mlp.experts.50.down_proj", "model.layers.70.mlp.experts.51.down_proj", "model.layers.70.mlp.experts.52.down_proj", "model.layers.70.mlp.experts.53.down_proj", "model.layers.70.mlp.experts.54.down_proj", "model.layers.70.mlp.experts.55.down_proj", "model.layers.70.mlp.experts.56.down_proj", "model.layers.70.mlp.experts.57.down_proj", "model.layers.70.mlp.experts.58.down_proj", "model.layers.70.mlp.experts.59.down_proj", "model.layers.70.mlp.experts.60.down_proj", "model.layers.70.mlp.experts.61.down_proj", "model.layers.70.mlp.experts.62.down_proj", "model.layers.70.mlp.experts.63.down_proj", "model.layers.70.mlp.experts.64.down_proj", "model.layers.70.mlp.experts.65.down_proj", "model.layers.70.mlp.experts.66.down_proj", "model.layers.70.mlp.experts.67.down_proj", "model.layers.70.mlp.experts.68.down_proj", "model.layers.70.mlp.experts.69.down_proj", "model.layers.70.mlp.experts.70.down_proj", "model.layers.70.mlp.experts.71.down_proj", "model.layers.70.mlp.experts.72.down_proj", "model.layers.70.mlp.experts.73.down_proj", "model.layers.70.mlp.experts.74.down_proj", "model.layers.70.mlp.experts.75.down_proj", "model.layers.70.mlp.experts.76.down_proj", "model.layers.70.mlp.experts.77.down_proj", "model.layers.70.mlp.experts.78.down_proj", "model.layers.70.mlp.experts.79.down_proj", "model.layers.70.mlp.experts.80.down_proj", "model.layers.70.mlp.experts.81.down_proj", "model.layers.70.mlp.experts.82.down_proj", "model.layers.70.mlp.experts.83.down_proj", "model.layers.70.mlp.experts.84.down_proj", "model.layers.70.mlp.experts.85.down_proj", "model.layers.70.mlp.experts.86.down_proj", "model.layers.70.mlp.experts.87.down_proj", "model.layers.70.mlp.experts.88.down_proj", "model.layers.70.mlp.experts.89.down_proj", "model.layers.70.mlp.experts.90.down_proj", "model.layers.70.mlp.experts.91.down_proj", "model.layers.70.mlp.experts.92.down_proj", "model.layers.70.mlp.experts.93.down_proj", "model.layers.70.mlp.experts.94.down_proj", "model.layers.70.mlp.experts.95.down_proj", "model.layers.70.mlp.experts.96.down_proj", "model.layers.70.mlp.experts.97.down_proj", "model.layers.70.mlp.experts.98.down_proj", "model.layers.70.mlp.experts.99.down_proj", "model.layers.70.mlp.experts.100.down_proj", "model.layers.70.mlp.experts.101.down_proj", "model.layers.70.mlp.experts.102.down_proj", "model.layers.70.mlp.experts.103.down_proj", "model.layers.70.mlp.experts.104.down_proj", "model.layers.70.mlp.experts.105.down_proj", "model.layers.70.mlp.experts.106.down_proj", "model.layers.70.mlp.experts.107.down_proj", "model.layers.70.mlp.experts.108.down_proj", "model.layers.70.mlp.experts.109.down_proj", "model.layers.70.mlp.experts.110.down_proj", "model.layers.70.mlp.experts.111.down_proj", "model.layers.70.mlp.experts.112.down_proj", "model.layers.70.mlp.experts.113.down_proj", "model.layers.70.mlp.experts.114.down_proj", "model.layers.70.mlp.experts.115.down_proj", "model.layers.70.mlp.experts.116.down_proj", "model.layers.70.mlp.experts.117.down_proj", "model.layers.70.mlp.experts.118.down_proj", "model.layers.70.mlp.experts.119.down_proj", "model.layers.70.mlp.experts.120.down_proj", "model.layers.70.mlp.experts.121.down_proj", "model.layers.70.mlp.experts.122.down_proj", "model.layers.70.mlp.experts.123.down_proj", "model.layers.70.mlp.experts.124.down_proj", "model.layers.70.mlp.experts.125.down_proj", "model.layers.70.mlp.experts.126.down_proj", "model.layers.70.mlp.experts.127.down_proj", "model.layers.70.mlp.experts.128.down_proj", "model.layers.70.mlp.experts.129.down_proj", "model.layers.70.mlp.experts.130.down_proj", "model.layers.70.mlp.experts.131.down_proj", "model.layers.70.mlp.experts.132.down_proj", "model.layers.70.mlp.experts.133.down_proj", "model.layers.70.mlp.experts.134.down_proj", "model.layers.70.mlp.experts.135.down_proj", "model.layers.70.mlp.experts.136.down_proj", "model.layers.70.mlp.experts.137.down_proj", "model.layers.70.mlp.experts.138.down_proj", "model.layers.70.mlp.experts.139.down_proj", "model.layers.70.mlp.experts.140.down_proj", "model.layers.70.mlp.experts.141.down_proj", "model.layers.70.mlp.experts.142.down_proj", "model.layers.70.mlp.experts.143.down_proj", "model.layers.70.mlp.experts.144.down_proj", "model.layers.70.mlp.experts.145.down_proj", "model.layers.70.mlp.experts.146.down_proj", "model.layers.70.mlp.experts.147.down_proj", "model.layers.70.mlp.experts.148.down_proj", "model.layers.70.mlp.experts.149.down_proj", "model.layers.70.mlp.experts.150.down_proj", "model.layers.70.mlp.experts.151.down_proj", "model.layers.70.mlp.experts.152.down_proj", "model.layers.70.mlp.experts.153.down_proj", "model.layers.70.mlp.experts.154.down_proj", "model.layers.70.mlp.experts.155.down_proj", "model.layers.70.mlp.experts.156.down_proj", "model.layers.70.mlp.experts.157.down_proj", "model.layers.70.mlp.experts.158.down_proj", "model.layers.70.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -3.272164613009332e-05, "dbits": 1258291200 }, { "dkld": -7.565468549729226e-05, "dbits": 2516582400 }, { "dkld": -8.065830916166583e-05, "dbits": 3774873600 }, { "dkld": -8.243098855019448e-05, "dbits": 6291456000 } ] }, { "idx": 423, "layers": [ "model.layers.71.self_attn.q_proj" ], "candidates": [ { "dkld": -1.8782913684955993e-06, "dbits": 62914560 }, { "dkld": 8.999276906251907e-05, "dbits": 125829120 }, { "dkld": 0.00010798024013637664, "dbits": 188743680 }, { "dkld": 7.28217884898158e-05, "dbits": 314572800 } ] }, { "idx": 424, "layers": [ "model.layers.71.self_attn.k_proj", "model.layers.71.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00032623354345560907, "dbits": 10485760 }, { "dkld": -0.00032526832073927203, "dbits": 20971520 }, { "dkld": -0.00039116088300944485, "dbits": 31457280 }, { "dkld": -0.0003633255138993374, "dbits": 52428800 } ] }, { "idx": 425, "layers": [ "model.layers.71.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00023001078516245166, "dbits": 62914560 }, { "dkld": -0.0003582755103707369, "dbits": 125829120 }, { "dkld": -0.00030824765563011725, "dbits": 188743680 }, { "dkld": -0.00035587158054113666, "dbits": 314572800 } ] }, { "idx": 426, "layers": [ "model.layers.71.mlp.shared_experts.gate_proj", "model.layers.71.mlp.shared_experts.up_proj", "model.layers.71.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0005682813003659165, "dbits": 23592960 }, { "dkld": 0.00040262741968034865, "dbits": 47185920 }, { "dkld": 0.00045929886400698783, "dbits": 70778880 }, { "dkld": 0.00043604541569947086, "dbits": 117964800 } ] }, { "idx": 427, "layers": [ "model.layers.71.mlp.experts.0.gate_proj", "model.layers.71.mlp.experts.1.gate_proj", "model.layers.71.mlp.experts.2.gate_proj", "model.layers.71.mlp.experts.3.gate_proj", "model.layers.71.mlp.experts.4.gate_proj", "model.layers.71.mlp.experts.5.gate_proj", "model.layers.71.mlp.experts.6.gate_proj", "model.layers.71.mlp.experts.7.gate_proj", "model.layers.71.mlp.experts.8.gate_proj", "model.layers.71.mlp.experts.9.gate_proj", "model.layers.71.mlp.experts.10.gate_proj", "model.layers.71.mlp.experts.11.gate_proj", "model.layers.71.mlp.experts.12.gate_proj", "model.layers.71.mlp.experts.13.gate_proj", "model.layers.71.mlp.experts.14.gate_proj", "model.layers.71.mlp.experts.15.gate_proj", "model.layers.71.mlp.experts.16.gate_proj", "model.layers.71.mlp.experts.17.gate_proj", "model.layers.71.mlp.experts.18.gate_proj", "model.layers.71.mlp.experts.19.gate_proj", "model.layers.71.mlp.experts.20.gate_proj", "model.layers.71.mlp.experts.21.gate_proj", "model.layers.71.mlp.experts.22.gate_proj", "model.layers.71.mlp.experts.23.gate_proj", "model.layers.71.mlp.experts.24.gate_proj", "model.layers.71.mlp.experts.25.gate_proj", "model.layers.71.mlp.experts.26.gate_proj", "model.layers.71.mlp.experts.27.gate_proj", "model.layers.71.mlp.experts.28.gate_proj", "model.layers.71.mlp.experts.29.gate_proj", "model.layers.71.mlp.experts.30.gate_proj", "model.layers.71.mlp.experts.31.gate_proj", "model.layers.71.mlp.experts.32.gate_proj", "model.layers.71.mlp.experts.33.gate_proj", "model.layers.71.mlp.experts.34.gate_proj", "model.layers.71.mlp.experts.35.gate_proj", "model.layers.71.mlp.experts.36.gate_proj", "model.layers.71.mlp.experts.37.gate_proj", "model.layers.71.mlp.experts.38.gate_proj", "model.layers.71.mlp.experts.39.gate_proj", "model.layers.71.mlp.experts.40.gate_proj", "model.layers.71.mlp.experts.41.gate_proj", "model.layers.71.mlp.experts.42.gate_proj", "model.layers.71.mlp.experts.43.gate_proj", "model.layers.71.mlp.experts.44.gate_proj", "model.layers.71.mlp.experts.45.gate_proj", "model.layers.71.mlp.experts.46.gate_proj", "model.layers.71.mlp.experts.47.gate_proj", "model.layers.71.mlp.experts.48.gate_proj", "model.layers.71.mlp.experts.49.gate_proj", "model.layers.71.mlp.experts.50.gate_proj", "model.layers.71.mlp.experts.51.gate_proj", "model.layers.71.mlp.experts.52.gate_proj", "model.layers.71.mlp.experts.53.gate_proj", "model.layers.71.mlp.experts.54.gate_proj", "model.layers.71.mlp.experts.55.gate_proj", "model.layers.71.mlp.experts.56.gate_proj", "model.layers.71.mlp.experts.57.gate_proj", "model.layers.71.mlp.experts.58.gate_proj", "model.layers.71.mlp.experts.59.gate_proj", "model.layers.71.mlp.experts.60.gate_proj", "model.layers.71.mlp.experts.61.gate_proj", "model.layers.71.mlp.experts.62.gate_proj", "model.layers.71.mlp.experts.63.gate_proj", "model.layers.71.mlp.experts.64.gate_proj", "model.layers.71.mlp.experts.65.gate_proj", "model.layers.71.mlp.experts.66.gate_proj", "model.layers.71.mlp.experts.67.gate_proj", "model.layers.71.mlp.experts.68.gate_proj", "model.layers.71.mlp.experts.69.gate_proj", "model.layers.71.mlp.experts.70.gate_proj", "model.layers.71.mlp.experts.71.gate_proj", "model.layers.71.mlp.experts.72.gate_proj", "model.layers.71.mlp.experts.73.gate_proj", "model.layers.71.mlp.experts.74.gate_proj", "model.layers.71.mlp.experts.75.gate_proj", "model.layers.71.mlp.experts.76.gate_proj", "model.layers.71.mlp.experts.77.gate_proj", "model.layers.71.mlp.experts.78.gate_proj", "model.layers.71.mlp.experts.79.gate_proj", "model.layers.71.mlp.experts.80.gate_proj", "model.layers.71.mlp.experts.81.gate_proj", "model.layers.71.mlp.experts.82.gate_proj", "model.layers.71.mlp.experts.83.gate_proj", "model.layers.71.mlp.experts.84.gate_proj", "model.layers.71.mlp.experts.85.gate_proj", "model.layers.71.mlp.experts.86.gate_proj", "model.layers.71.mlp.experts.87.gate_proj", "model.layers.71.mlp.experts.88.gate_proj", "model.layers.71.mlp.experts.89.gate_proj", "model.layers.71.mlp.experts.90.gate_proj", "model.layers.71.mlp.experts.91.gate_proj", "model.layers.71.mlp.experts.92.gate_proj", "model.layers.71.mlp.experts.93.gate_proj", "model.layers.71.mlp.experts.94.gate_proj", "model.layers.71.mlp.experts.95.gate_proj", "model.layers.71.mlp.experts.96.gate_proj", "model.layers.71.mlp.experts.97.gate_proj", "model.layers.71.mlp.experts.98.gate_proj", "model.layers.71.mlp.experts.99.gate_proj", "model.layers.71.mlp.experts.100.gate_proj", "model.layers.71.mlp.experts.101.gate_proj", "model.layers.71.mlp.experts.102.gate_proj", "model.layers.71.mlp.experts.103.gate_proj", "model.layers.71.mlp.experts.104.gate_proj", "model.layers.71.mlp.experts.105.gate_proj", "model.layers.71.mlp.experts.106.gate_proj", "model.layers.71.mlp.experts.107.gate_proj", "model.layers.71.mlp.experts.108.gate_proj", "model.layers.71.mlp.experts.109.gate_proj", "model.layers.71.mlp.experts.110.gate_proj", "model.layers.71.mlp.experts.111.gate_proj", "model.layers.71.mlp.experts.112.gate_proj", "model.layers.71.mlp.experts.113.gate_proj", "model.layers.71.mlp.experts.114.gate_proj", "model.layers.71.mlp.experts.115.gate_proj", "model.layers.71.mlp.experts.116.gate_proj", "model.layers.71.mlp.experts.117.gate_proj", "model.layers.71.mlp.experts.118.gate_proj", "model.layers.71.mlp.experts.119.gate_proj", "model.layers.71.mlp.experts.120.gate_proj", "model.layers.71.mlp.experts.121.gate_proj", "model.layers.71.mlp.experts.122.gate_proj", "model.layers.71.mlp.experts.123.gate_proj", "model.layers.71.mlp.experts.124.gate_proj", "model.layers.71.mlp.experts.125.gate_proj", "model.layers.71.mlp.experts.126.gate_proj", "model.layers.71.mlp.experts.127.gate_proj", "model.layers.71.mlp.experts.128.gate_proj", "model.layers.71.mlp.experts.129.gate_proj", "model.layers.71.mlp.experts.130.gate_proj", "model.layers.71.mlp.experts.131.gate_proj", "model.layers.71.mlp.experts.132.gate_proj", "model.layers.71.mlp.experts.133.gate_proj", "model.layers.71.mlp.experts.134.gate_proj", "model.layers.71.mlp.experts.135.gate_proj", "model.layers.71.mlp.experts.136.gate_proj", "model.layers.71.mlp.experts.137.gate_proj", "model.layers.71.mlp.experts.138.gate_proj", "model.layers.71.mlp.experts.139.gate_proj", "model.layers.71.mlp.experts.140.gate_proj", "model.layers.71.mlp.experts.141.gate_proj", "model.layers.71.mlp.experts.142.gate_proj", "model.layers.71.mlp.experts.143.gate_proj", "model.layers.71.mlp.experts.144.gate_proj", "model.layers.71.mlp.experts.145.gate_proj", "model.layers.71.mlp.experts.146.gate_proj", "model.layers.71.mlp.experts.147.gate_proj", "model.layers.71.mlp.experts.148.gate_proj", "model.layers.71.mlp.experts.149.gate_proj", "model.layers.71.mlp.experts.150.gate_proj", "model.layers.71.mlp.experts.151.gate_proj", "model.layers.71.mlp.experts.152.gate_proj", "model.layers.71.mlp.experts.153.gate_proj", "model.layers.71.mlp.experts.154.gate_proj", "model.layers.71.mlp.experts.155.gate_proj", "model.layers.71.mlp.experts.156.gate_proj", "model.layers.71.mlp.experts.157.gate_proj", "model.layers.71.mlp.experts.158.gate_proj", "model.layers.71.mlp.experts.159.gate_proj", "model.layers.71.mlp.experts.0.up_proj", "model.layers.71.mlp.experts.1.up_proj", "model.layers.71.mlp.experts.2.up_proj", "model.layers.71.mlp.experts.3.up_proj", "model.layers.71.mlp.experts.4.up_proj", "model.layers.71.mlp.experts.5.up_proj", "model.layers.71.mlp.experts.6.up_proj", "model.layers.71.mlp.experts.7.up_proj", "model.layers.71.mlp.experts.8.up_proj", "model.layers.71.mlp.experts.9.up_proj", "model.layers.71.mlp.experts.10.up_proj", "model.layers.71.mlp.experts.11.up_proj", "model.layers.71.mlp.experts.12.up_proj", "model.layers.71.mlp.experts.13.up_proj", "model.layers.71.mlp.experts.14.up_proj", "model.layers.71.mlp.experts.15.up_proj", "model.layers.71.mlp.experts.16.up_proj", "model.layers.71.mlp.experts.17.up_proj", "model.layers.71.mlp.experts.18.up_proj", "model.layers.71.mlp.experts.19.up_proj", "model.layers.71.mlp.experts.20.up_proj", "model.layers.71.mlp.experts.21.up_proj", "model.layers.71.mlp.experts.22.up_proj", "model.layers.71.mlp.experts.23.up_proj", "model.layers.71.mlp.experts.24.up_proj", "model.layers.71.mlp.experts.25.up_proj", "model.layers.71.mlp.experts.26.up_proj", "model.layers.71.mlp.experts.27.up_proj", "model.layers.71.mlp.experts.28.up_proj", "model.layers.71.mlp.experts.29.up_proj", "model.layers.71.mlp.experts.30.up_proj", "model.layers.71.mlp.experts.31.up_proj", "model.layers.71.mlp.experts.32.up_proj", "model.layers.71.mlp.experts.33.up_proj", "model.layers.71.mlp.experts.34.up_proj", "model.layers.71.mlp.experts.35.up_proj", "model.layers.71.mlp.experts.36.up_proj", "model.layers.71.mlp.experts.37.up_proj", "model.layers.71.mlp.experts.38.up_proj", "model.layers.71.mlp.experts.39.up_proj", "model.layers.71.mlp.experts.40.up_proj", "model.layers.71.mlp.experts.41.up_proj", "model.layers.71.mlp.experts.42.up_proj", "model.layers.71.mlp.experts.43.up_proj", "model.layers.71.mlp.experts.44.up_proj", "model.layers.71.mlp.experts.45.up_proj", "model.layers.71.mlp.experts.46.up_proj", "model.layers.71.mlp.experts.47.up_proj", "model.layers.71.mlp.experts.48.up_proj", "model.layers.71.mlp.experts.49.up_proj", "model.layers.71.mlp.experts.50.up_proj", "model.layers.71.mlp.experts.51.up_proj", "model.layers.71.mlp.experts.52.up_proj", "model.layers.71.mlp.experts.53.up_proj", "model.layers.71.mlp.experts.54.up_proj", "model.layers.71.mlp.experts.55.up_proj", "model.layers.71.mlp.experts.56.up_proj", "model.layers.71.mlp.experts.57.up_proj", "model.layers.71.mlp.experts.58.up_proj", "model.layers.71.mlp.experts.59.up_proj", "model.layers.71.mlp.experts.60.up_proj", "model.layers.71.mlp.experts.61.up_proj", "model.layers.71.mlp.experts.62.up_proj", "model.layers.71.mlp.experts.63.up_proj", "model.layers.71.mlp.experts.64.up_proj", "model.layers.71.mlp.experts.65.up_proj", "model.layers.71.mlp.experts.66.up_proj", "model.layers.71.mlp.experts.67.up_proj", "model.layers.71.mlp.experts.68.up_proj", "model.layers.71.mlp.experts.69.up_proj", "model.layers.71.mlp.experts.70.up_proj", "model.layers.71.mlp.experts.71.up_proj", "model.layers.71.mlp.experts.72.up_proj", "model.layers.71.mlp.experts.73.up_proj", "model.layers.71.mlp.experts.74.up_proj", "model.layers.71.mlp.experts.75.up_proj", "model.layers.71.mlp.experts.76.up_proj", "model.layers.71.mlp.experts.77.up_proj", "model.layers.71.mlp.experts.78.up_proj", "model.layers.71.mlp.experts.79.up_proj", "model.layers.71.mlp.experts.80.up_proj", "model.layers.71.mlp.experts.81.up_proj", "model.layers.71.mlp.experts.82.up_proj", "model.layers.71.mlp.experts.83.up_proj", "model.layers.71.mlp.experts.84.up_proj", "model.layers.71.mlp.experts.85.up_proj", "model.layers.71.mlp.experts.86.up_proj", "model.layers.71.mlp.experts.87.up_proj", "model.layers.71.mlp.experts.88.up_proj", "model.layers.71.mlp.experts.89.up_proj", "model.layers.71.mlp.experts.90.up_proj", "model.layers.71.mlp.experts.91.up_proj", "model.layers.71.mlp.experts.92.up_proj", "model.layers.71.mlp.experts.93.up_proj", "model.layers.71.mlp.experts.94.up_proj", "model.layers.71.mlp.experts.95.up_proj", "model.layers.71.mlp.experts.96.up_proj", "model.layers.71.mlp.experts.97.up_proj", "model.layers.71.mlp.experts.98.up_proj", "model.layers.71.mlp.experts.99.up_proj", "model.layers.71.mlp.experts.100.up_proj", "model.layers.71.mlp.experts.101.up_proj", "model.layers.71.mlp.experts.102.up_proj", "model.layers.71.mlp.experts.103.up_proj", "model.layers.71.mlp.experts.104.up_proj", "model.layers.71.mlp.experts.105.up_proj", "model.layers.71.mlp.experts.106.up_proj", "model.layers.71.mlp.experts.107.up_proj", "model.layers.71.mlp.experts.108.up_proj", "model.layers.71.mlp.experts.109.up_proj", "model.layers.71.mlp.experts.110.up_proj", "model.layers.71.mlp.experts.111.up_proj", "model.layers.71.mlp.experts.112.up_proj", "model.layers.71.mlp.experts.113.up_proj", "model.layers.71.mlp.experts.114.up_proj", "model.layers.71.mlp.experts.115.up_proj", "model.layers.71.mlp.experts.116.up_proj", "model.layers.71.mlp.experts.117.up_proj", "model.layers.71.mlp.experts.118.up_proj", "model.layers.71.mlp.experts.119.up_proj", "model.layers.71.mlp.experts.120.up_proj", "model.layers.71.mlp.experts.121.up_proj", "model.layers.71.mlp.experts.122.up_proj", "model.layers.71.mlp.experts.123.up_proj", "model.layers.71.mlp.experts.124.up_proj", "model.layers.71.mlp.experts.125.up_proj", "model.layers.71.mlp.experts.126.up_proj", "model.layers.71.mlp.experts.127.up_proj", "model.layers.71.mlp.experts.128.up_proj", "model.layers.71.mlp.experts.129.up_proj", "model.layers.71.mlp.experts.130.up_proj", "model.layers.71.mlp.experts.131.up_proj", "model.layers.71.mlp.experts.132.up_proj", "model.layers.71.mlp.experts.133.up_proj", "model.layers.71.mlp.experts.134.up_proj", "model.layers.71.mlp.experts.135.up_proj", "model.layers.71.mlp.experts.136.up_proj", "model.layers.71.mlp.experts.137.up_proj", "model.layers.71.mlp.experts.138.up_proj", "model.layers.71.mlp.experts.139.up_proj", "model.layers.71.mlp.experts.140.up_proj", "model.layers.71.mlp.experts.141.up_proj", "model.layers.71.mlp.experts.142.up_proj", "model.layers.71.mlp.experts.143.up_proj", "model.layers.71.mlp.experts.144.up_proj", "model.layers.71.mlp.experts.145.up_proj", "model.layers.71.mlp.experts.146.up_proj", "model.layers.71.mlp.experts.147.up_proj", "model.layers.71.mlp.experts.148.up_proj", "model.layers.71.mlp.experts.149.up_proj", "model.layers.71.mlp.experts.150.up_proj", "model.layers.71.mlp.experts.151.up_proj", "model.layers.71.mlp.experts.152.up_proj", "model.layers.71.mlp.experts.153.up_proj", "model.layers.71.mlp.experts.154.up_proj", "model.layers.71.mlp.experts.155.up_proj", "model.layers.71.mlp.experts.156.up_proj", "model.layers.71.mlp.experts.157.up_proj", "model.layers.71.mlp.experts.158.up_proj", "model.layers.71.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00038291020318866054, "dbits": 2516582400 }, { "dkld": -0.00044766571372747976, "dbits": 5033164800 }, { "dkld": -0.00044913319870830415, "dbits": 7549747200 }, { "dkld": -0.0004707679152488736, "dbits": 12582912000 } ] }, { "idx": 428, "layers": [ "model.layers.71.mlp.experts.0.down_proj", "model.layers.71.mlp.experts.1.down_proj", "model.layers.71.mlp.experts.2.down_proj", "model.layers.71.mlp.experts.3.down_proj", "model.layers.71.mlp.experts.4.down_proj", "model.layers.71.mlp.experts.5.down_proj", "model.layers.71.mlp.experts.6.down_proj", "model.layers.71.mlp.experts.7.down_proj", "model.layers.71.mlp.experts.8.down_proj", "model.layers.71.mlp.experts.9.down_proj", "model.layers.71.mlp.experts.10.down_proj", "model.layers.71.mlp.experts.11.down_proj", "model.layers.71.mlp.experts.12.down_proj", "model.layers.71.mlp.experts.13.down_proj", "model.layers.71.mlp.experts.14.down_proj", "model.layers.71.mlp.experts.15.down_proj", "model.layers.71.mlp.experts.16.down_proj", "model.layers.71.mlp.experts.17.down_proj", "model.layers.71.mlp.experts.18.down_proj", "model.layers.71.mlp.experts.19.down_proj", "model.layers.71.mlp.experts.20.down_proj", "model.layers.71.mlp.experts.21.down_proj", "model.layers.71.mlp.experts.22.down_proj", "model.layers.71.mlp.experts.23.down_proj", "model.layers.71.mlp.experts.24.down_proj", "model.layers.71.mlp.experts.25.down_proj", "model.layers.71.mlp.experts.26.down_proj", "model.layers.71.mlp.experts.27.down_proj", "model.layers.71.mlp.experts.28.down_proj", "model.layers.71.mlp.experts.29.down_proj", "model.layers.71.mlp.experts.30.down_proj", "model.layers.71.mlp.experts.31.down_proj", "model.layers.71.mlp.experts.32.down_proj", "model.layers.71.mlp.experts.33.down_proj", "model.layers.71.mlp.experts.34.down_proj", "model.layers.71.mlp.experts.35.down_proj", "model.layers.71.mlp.experts.36.down_proj", "model.layers.71.mlp.experts.37.down_proj", "model.layers.71.mlp.experts.38.down_proj", "model.layers.71.mlp.experts.39.down_proj", "model.layers.71.mlp.experts.40.down_proj", "model.layers.71.mlp.experts.41.down_proj", "model.layers.71.mlp.experts.42.down_proj", "model.layers.71.mlp.experts.43.down_proj", "model.layers.71.mlp.experts.44.down_proj", "model.layers.71.mlp.experts.45.down_proj", "model.layers.71.mlp.experts.46.down_proj", "model.layers.71.mlp.experts.47.down_proj", "model.layers.71.mlp.experts.48.down_proj", "model.layers.71.mlp.experts.49.down_proj", "model.layers.71.mlp.experts.50.down_proj", "model.layers.71.mlp.experts.51.down_proj", "model.layers.71.mlp.experts.52.down_proj", "model.layers.71.mlp.experts.53.down_proj", "model.layers.71.mlp.experts.54.down_proj", "model.layers.71.mlp.experts.55.down_proj", "model.layers.71.mlp.experts.56.down_proj", "model.layers.71.mlp.experts.57.down_proj", "model.layers.71.mlp.experts.58.down_proj", "model.layers.71.mlp.experts.59.down_proj", "model.layers.71.mlp.experts.60.down_proj", "model.layers.71.mlp.experts.61.down_proj", "model.layers.71.mlp.experts.62.down_proj", "model.layers.71.mlp.experts.63.down_proj", "model.layers.71.mlp.experts.64.down_proj", "model.layers.71.mlp.experts.65.down_proj", "model.layers.71.mlp.experts.66.down_proj", "model.layers.71.mlp.experts.67.down_proj", "model.layers.71.mlp.experts.68.down_proj", "model.layers.71.mlp.experts.69.down_proj", "model.layers.71.mlp.experts.70.down_proj", "model.layers.71.mlp.experts.71.down_proj", "model.layers.71.mlp.experts.72.down_proj", "model.layers.71.mlp.experts.73.down_proj", "model.layers.71.mlp.experts.74.down_proj", "model.layers.71.mlp.experts.75.down_proj", "model.layers.71.mlp.experts.76.down_proj", "model.layers.71.mlp.experts.77.down_proj", "model.layers.71.mlp.experts.78.down_proj", "model.layers.71.mlp.experts.79.down_proj", "model.layers.71.mlp.experts.80.down_proj", "model.layers.71.mlp.experts.81.down_proj", "model.layers.71.mlp.experts.82.down_proj", "model.layers.71.mlp.experts.83.down_proj", "model.layers.71.mlp.experts.84.down_proj", "model.layers.71.mlp.experts.85.down_proj", "model.layers.71.mlp.experts.86.down_proj", "model.layers.71.mlp.experts.87.down_proj", "model.layers.71.mlp.experts.88.down_proj", "model.layers.71.mlp.experts.89.down_proj", "model.layers.71.mlp.experts.90.down_proj", "model.layers.71.mlp.experts.91.down_proj", "model.layers.71.mlp.experts.92.down_proj", "model.layers.71.mlp.experts.93.down_proj", "model.layers.71.mlp.experts.94.down_proj", "model.layers.71.mlp.experts.95.down_proj", "model.layers.71.mlp.experts.96.down_proj", "model.layers.71.mlp.experts.97.down_proj", "model.layers.71.mlp.experts.98.down_proj", "model.layers.71.mlp.experts.99.down_proj", "model.layers.71.mlp.experts.100.down_proj", "model.layers.71.mlp.experts.101.down_proj", "model.layers.71.mlp.experts.102.down_proj", "model.layers.71.mlp.experts.103.down_proj", "model.layers.71.mlp.experts.104.down_proj", "model.layers.71.mlp.experts.105.down_proj", "model.layers.71.mlp.experts.106.down_proj", "model.layers.71.mlp.experts.107.down_proj", "model.layers.71.mlp.experts.108.down_proj", "model.layers.71.mlp.experts.109.down_proj", "model.layers.71.mlp.experts.110.down_proj", "model.layers.71.mlp.experts.111.down_proj", "model.layers.71.mlp.experts.112.down_proj", "model.layers.71.mlp.experts.113.down_proj", "model.layers.71.mlp.experts.114.down_proj", "model.layers.71.mlp.experts.115.down_proj", "model.layers.71.mlp.experts.116.down_proj", "model.layers.71.mlp.experts.117.down_proj", "model.layers.71.mlp.experts.118.down_proj", "model.layers.71.mlp.experts.119.down_proj", "model.layers.71.mlp.experts.120.down_proj", "model.layers.71.mlp.experts.121.down_proj", "model.layers.71.mlp.experts.122.down_proj", "model.layers.71.mlp.experts.123.down_proj", "model.layers.71.mlp.experts.124.down_proj", "model.layers.71.mlp.experts.125.down_proj", "model.layers.71.mlp.experts.126.down_proj", "model.layers.71.mlp.experts.127.down_proj", "model.layers.71.mlp.experts.128.down_proj", "model.layers.71.mlp.experts.129.down_proj", "model.layers.71.mlp.experts.130.down_proj", "model.layers.71.mlp.experts.131.down_proj", "model.layers.71.mlp.experts.132.down_proj", "model.layers.71.mlp.experts.133.down_proj", "model.layers.71.mlp.experts.134.down_proj", "model.layers.71.mlp.experts.135.down_proj", "model.layers.71.mlp.experts.136.down_proj", "model.layers.71.mlp.experts.137.down_proj", "model.layers.71.mlp.experts.138.down_proj", "model.layers.71.mlp.experts.139.down_proj", "model.layers.71.mlp.experts.140.down_proj", "model.layers.71.mlp.experts.141.down_proj", "model.layers.71.mlp.experts.142.down_proj", "model.layers.71.mlp.experts.143.down_proj", "model.layers.71.mlp.experts.144.down_proj", "model.layers.71.mlp.experts.145.down_proj", "model.layers.71.mlp.experts.146.down_proj", "model.layers.71.mlp.experts.147.down_proj", "model.layers.71.mlp.experts.148.down_proj", "model.layers.71.mlp.experts.149.down_proj", "model.layers.71.mlp.experts.150.down_proj", "model.layers.71.mlp.experts.151.down_proj", "model.layers.71.mlp.experts.152.down_proj", "model.layers.71.mlp.experts.153.down_proj", "model.layers.71.mlp.experts.154.down_proj", "model.layers.71.mlp.experts.155.down_proj", "model.layers.71.mlp.experts.156.down_proj", "model.layers.71.mlp.experts.157.down_proj", "model.layers.71.mlp.experts.158.down_proj", "model.layers.71.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.0001703741028905026, "dbits": 1258291200 }, { "dkld": -0.0001366671174764661, "dbits": 2516582400 }, { "dkld": -0.00013466011732816974, "dbits": 3774873600 }, { "dkld": -0.00012939106673003076, "dbits": 6291456000 } ] }, { "idx": 429, "layers": [ "model.layers.72.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00013829600065946857, "dbits": 62914560 }, { "dkld": -0.00011123716831207275, "dbits": 125829120 }, { "dkld": -0.00010088747367263673, "dbits": 188743680 }, { "dkld": -0.00010567829012871066, "dbits": 314572800 } ] }, { "idx": 430, "layers": [ "model.layers.72.self_attn.k_proj", "model.layers.72.self_attn.v_proj" ], "candidates": [ { "dkld": -6.853006780148385e-05, "dbits": 10485760 }, { "dkld": -3.14688310027178e-05, "dbits": 20971520 }, { "dkld": -2.3661274462946635e-05, "dbits": 31457280 }, { "dkld": -8.628983050595895e-06, "dbits": 52428800 } ] }, { "idx": 431, "layers": [ "model.layers.72.self_attn.o_proj" ], "candidates": [ { "dkld": 4.573259502649307e-05, "dbits": 62914560 }, { "dkld": -8.66254791617449e-05, "dbits": 125829120 }, { "dkld": -7.152082398534376e-05, "dbits": 188743680 }, { "dkld": -8.466783910990316e-05, "dbits": 314572800 } ] }, { "idx": 432, "layers": [ "model.layers.72.mlp.shared_experts.gate_proj", "model.layers.72.mlp.shared_experts.up_proj", "model.layers.72.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00031754942610860704, "dbits": 23592960 }, { "dkld": -0.0005067488178610857, "dbits": 47185920 }, { "dkld": -0.0004073618911206722, "dbits": 70778880 }, { "dkld": -0.00042662397027015686, "dbits": 117964800 } ] }, { "idx": 433, "layers": [ "model.layers.72.mlp.experts.0.gate_proj", "model.layers.72.mlp.experts.1.gate_proj", "model.layers.72.mlp.experts.2.gate_proj", "model.layers.72.mlp.experts.3.gate_proj", "model.layers.72.mlp.experts.4.gate_proj", "model.layers.72.mlp.experts.5.gate_proj", "model.layers.72.mlp.experts.6.gate_proj", "model.layers.72.mlp.experts.7.gate_proj", "model.layers.72.mlp.experts.8.gate_proj", "model.layers.72.mlp.experts.9.gate_proj", "model.layers.72.mlp.experts.10.gate_proj", "model.layers.72.mlp.experts.11.gate_proj", "model.layers.72.mlp.experts.12.gate_proj", "model.layers.72.mlp.experts.13.gate_proj", "model.layers.72.mlp.experts.14.gate_proj", "model.layers.72.mlp.experts.15.gate_proj", "model.layers.72.mlp.experts.16.gate_proj", "model.layers.72.mlp.experts.17.gate_proj", "model.layers.72.mlp.experts.18.gate_proj", "model.layers.72.mlp.experts.19.gate_proj", "model.layers.72.mlp.experts.20.gate_proj", "model.layers.72.mlp.experts.21.gate_proj", "model.layers.72.mlp.experts.22.gate_proj", "model.layers.72.mlp.experts.23.gate_proj", "model.layers.72.mlp.experts.24.gate_proj", "model.layers.72.mlp.experts.25.gate_proj", "model.layers.72.mlp.experts.26.gate_proj", "model.layers.72.mlp.experts.27.gate_proj", "model.layers.72.mlp.experts.28.gate_proj", "model.layers.72.mlp.experts.29.gate_proj", "model.layers.72.mlp.experts.30.gate_proj", "model.layers.72.mlp.experts.31.gate_proj", "model.layers.72.mlp.experts.32.gate_proj", "model.layers.72.mlp.experts.33.gate_proj", "model.layers.72.mlp.experts.34.gate_proj", "model.layers.72.mlp.experts.35.gate_proj", "model.layers.72.mlp.experts.36.gate_proj", "model.layers.72.mlp.experts.37.gate_proj", "model.layers.72.mlp.experts.38.gate_proj", "model.layers.72.mlp.experts.39.gate_proj", "model.layers.72.mlp.experts.40.gate_proj", "model.layers.72.mlp.experts.41.gate_proj", "model.layers.72.mlp.experts.42.gate_proj", "model.layers.72.mlp.experts.43.gate_proj", "model.layers.72.mlp.experts.44.gate_proj", "model.layers.72.mlp.experts.45.gate_proj", "model.layers.72.mlp.experts.46.gate_proj", "model.layers.72.mlp.experts.47.gate_proj", "model.layers.72.mlp.experts.48.gate_proj", "model.layers.72.mlp.experts.49.gate_proj", "model.layers.72.mlp.experts.50.gate_proj", "model.layers.72.mlp.experts.51.gate_proj", "model.layers.72.mlp.experts.52.gate_proj", "model.layers.72.mlp.experts.53.gate_proj", "model.layers.72.mlp.experts.54.gate_proj", "model.layers.72.mlp.experts.55.gate_proj", "model.layers.72.mlp.experts.56.gate_proj", "model.layers.72.mlp.experts.57.gate_proj", "model.layers.72.mlp.experts.58.gate_proj", "model.layers.72.mlp.experts.59.gate_proj", "model.layers.72.mlp.experts.60.gate_proj", "model.layers.72.mlp.experts.61.gate_proj", "model.layers.72.mlp.experts.62.gate_proj", "model.layers.72.mlp.experts.63.gate_proj", "model.layers.72.mlp.experts.64.gate_proj", "model.layers.72.mlp.experts.65.gate_proj", "model.layers.72.mlp.experts.66.gate_proj", "model.layers.72.mlp.experts.67.gate_proj", "model.layers.72.mlp.experts.68.gate_proj", "model.layers.72.mlp.experts.69.gate_proj", "model.layers.72.mlp.experts.70.gate_proj", "model.layers.72.mlp.experts.71.gate_proj", "model.layers.72.mlp.experts.72.gate_proj", "model.layers.72.mlp.experts.73.gate_proj", "model.layers.72.mlp.experts.74.gate_proj", "model.layers.72.mlp.experts.75.gate_proj", "model.layers.72.mlp.experts.76.gate_proj", "model.layers.72.mlp.experts.77.gate_proj", "model.layers.72.mlp.experts.78.gate_proj", "model.layers.72.mlp.experts.79.gate_proj", "model.layers.72.mlp.experts.80.gate_proj", "model.layers.72.mlp.experts.81.gate_proj", "model.layers.72.mlp.experts.82.gate_proj", "model.layers.72.mlp.experts.83.gate_proj", "model.layers.72.mlp.experts.84.gate_proj", "model.layers.72.mlp.experts.85.gate_proj", "model.layers.72.mlp.experts.86.gate_proj", "model.layers.72.mlp.experts.87.gate_proj", "model.layers.72.mlp.experts.88.gate_proj", "model.layers.72.mlp.experts.89.gate_proj", "model.layers.72.mlp.experts.90.gate_proj", "model.layers.72.mlp.experts.91.gate_proj", "model.layers.72.mlp.experts.92.gate_proj", "model.layers.72.mlp.experts.93.gate_proj", "model.layers.72.mlp.experts.94.gate_proj", "model.layers.72.mlp.experts.95.gate_proj", "model.layers.72.mlp.experts.96.gate_proj", "model.layers.72.mlp.experts.97.gate_proj", "model.layers.72.mlp.experts.98.gate_proj", "model.layers.72.mlp.experts.99.gate_proj", "model.layers.72.mlp.experts.100.gate_proj", "model.layers.72.mlp.experts.101.gate_proj", "model.layers.72.mlp.experts.102.gate_proj", "model.layers.72.mlp.experts.103.gate_proj", "model.layers.72.mlp.experts.104.gate_proj", "model.layers.72.mlp.experts.105.gate_proj", "model.layers.72.mlp.experts.106.gate_proj", "model.layers.72.mlp.experts.107.gate_proj", "model.layers.72.mlp.experts.108.gate_proj", "model.layers.72.mlp.experts.109.gate_proj", "model.layers.72.mlp.experts.110.gate_proj", "model.layers.72.mlp.experts.111.gate_proj", "model.layers.72.mlp.experts.112.gate_proj", "model.layers.72.mlp.experts.113.gate_proj", "model.layers.72.mlp.experts.114.gate_proj", "model.layers.72.mlp.experts.115.gate_proj", "model.layers.72.mlp.experts.116.gate_proj", "model.layers.72.mlp.experts.117.gate_proj", "model.layers.72.mlp.experts.118.gate_proj", "model.layers.72.mlp.experts.119.gate_proj", "model.layers.72.mlp.experts.120.gate_proj", "model.layers.72.mlp.experts.121.gate_proj", "model.layers.72.mlp.experts.122.gate_proj", "model.layers.72.mlp.experts.123.gate_proj", "model.layers.72.mlp.experts.124.gate_proj", "model.layers.72.mlp.experts.125.gate_proj", "model.layers.72.mlp.experts.126.gate_proj", "model.layers.72.mlp.experts.127.gate_proj", "model.layers.72.mlp.experts.128.gate_proj", "model.layers.72.mlp.experts.129.gate_proj", "model.layers.72.mlp.experts.130.gate_proj", "model.layers.72.mlp.experts.131.gate_proj", "model.layers.72.mlp.experts.132.gate_proj", "model.layers.72.mlp.experts.133.gate_proj", "model.layers.72.mlp.experts.134.gate_proj", "model.layers.72.mlp.experts.135.gate_proj", "model.layers.72.mlp.experts.136.gate_proj", "model.layers.72.mlp.experts.137.gate_proj", "model.layers.72.mlp.experts.138.gate_proj", "model.layers.72.mlp.experts.139.gate_proj", "model.layers.72.mlp.experts.140.gate_proj", "model.layers.72.mlp.experts.141.gate_proj", "model.layers.72.mlp.experts.142.gate_proj", "model.layers.72.mlp.experts.143.gate_proj", "model.layers.72.mlp.experts.144.gate_proj", "model.layers.72.mlp.experts.145.gate_proj", "model.layers.72.mlp.experts.146.gate_proj", "model.layers.72.mlp.experts.147.gate_proj", "model.layers.72.mlp.experts.148.gate_proj", "model.layers.72.mlp.experts.149.gate_proj", "model.layers.72.mlp.experts.150.gate_proj", "model.layers.72.mlp.experts.151.gate_proj", "model.layers.72.mlp.experts.152.gate_proj", "model.layers.72.mlp.experts.153.gate_proj", "model.layers.72.mlp.experts.154.gate_proj", "model.layers.72.mlp.experts.155.gate_proj", "model.layers.72.mlp.experts.156.gate_proj", "model.layers.72.mlp.experts.157.gate_proj", "model.layers.72.mlp.experts.158.gate_proj", "model.layers.72.mlp.experts.159.gate_proj", "model.layers.72.mlp.experts.0.up_proj", "model.layers.72.mlp.experts.1.up_proj", "model.layers.72.mlp.experts.2.up_proj", "model.layers.72.mlp.experts.3.up_proj", "model.layers.72.mlp.experts.4.up_proj", "model.layers.72.mlp.experts.5.up_proj", "model.layers.72.mlp.experts.6.up_proj", "model.layers.72.mlp.experts.7.up_proj", "model.layers.72.mlp.experts.8.up_proj", "model.layers.72.mlp.experts.9.up_proj", "model.layers.72.mlp.experts.10.up_proj", "model.layers.72.mlp.experts.11.up_proj", "model.layers.72.mlp.experts.12.up_proj", "model.layers.72.mlp.experts.13.up_proj", "model.layers.72.mlp.experts.14.up_proj", "model.layers.72.mlp.experts.15.up_proj", "model.layers.72.mlp.experts.16.up_proj", "model.layers.72.mlp.experts.17.up_proj", "model.layers.72.mlp.experts.18.up_proj", "model.layers.72.mlp.experts.19.up_proj", "model.layers.72.mlp.experts.20.up_proj", "model.layers.72.mlp.experts.21.up_proj", "model.layers.72.mlp.experts.22.up_proj", "model.layers.72.mlp.experts.23.up_proj", "model.layers.72.mlp.experts.24.up_proj", "model.layers.72.mlp.experts.25.up_proj", "model.layers.72.mlp.experts.26.up_proj", "model.layers.72.mlp.experts.27.up_proj", "model.layers.72.mlp.experts.28.up_proj", "model.layers.72.mlp.experts.29.up_proj", "model.layers.72.mlp.experts.30.up_proj", "model.layers.72.mlp.experts.31.up_proj", "model.layers.72.mlp.experts.32.up_proj", "model.layers.72.mlp.experts.33.up_proj", "model.layers.72.mlp.experts.34.up_proj", "model.layers.72.mlp.experts.35.up_proj", "model.layers.72.mlp.experts.36.up_proj", "model.layers.72.mlp.experts.37.up_proj", "model.layers.72.mlp.experts.38.up_proj", "model.layers.72.mlp.experts.39.up_proj", "model.layers.72.mlp.experts.40.up_proj", "model.layers.72.mlp.experts.41.up_proj", "model.layers.72.mlp.experts.42.up_proj", "model.layers.72.mlp.experts.43.up_proj", "model.layers.72.mlp.experts.44.up_proj", "model.layers.72.mlp.experts.45.up_proj", "model.layers.72.mlp.experts.46.up_proj", "model.layers.72.mlp.experts.47.up_proj", "model.layers.72.mlp.experts.48.up_proj", "model.layers.72.mlp.experts.49.up_proj", "model.layers.72.mlp.experts.50.up_proj", "model.layers.72.mlp.experts.51.up_proj", "model.layers.72.mlp.experts.52.up_proj", "model.layers.72.mlp.experts.53.up_proj", "model.layers.72.mlp.experts.54.up_proj", "model.layers.72.mlp.experts.55.up_proj", "model.layers.72.mlp.experts.56.up_proj", "model.layers.72.mlp.experts.57.up_proj", "model.layers.72.mlp.experts.58.up_proj", "model.layers.72.mlp.experts.59.up_proj", "model.layers.72.mlp.experts.60.up_proj", "model.layers.72.mlp.experts.61.up_proj", "model.layers.72.mlp.experts.62.up_proj", "model.layers.72.mlp.experts.63.up_proj", "model.layers.72.mlp.experts.64.up_proj", "model.layers.72.mlp.experts.65.up_proj", "model.layers.72.mlp.experts.66.up_proj", "model.layers.72.mlp.experts.67.up_proj", "model.layers.72.mlp.experts.68.up_proj", "model.layers.72.mlp.experts.69.up_proj", "model.layers.72.mlp.experts.70.up_proj", "model.layers.72.mlp.experts.71.up_proj", "model.layers.72.mlp.experts.72.up_proj", "model.layers.72.mlp.experts.73.up_proj", "model.layers.72.mlp.experts.74.up_proj", "model.layers.72.mlp.experts.75.up_proj", "model.layers.72.mlp.experts.76.up_proj", "model.layers.72.mlp.experts.77.up_proj", "model.layers.72.mlp.experts.78.up_proj", "model.layers.72.mlp.experts.79.up_proj", "model.layers.72.mlp.experts.80.up_proj", "model.layers.72.mlp.experts.81.up_proj", "model.layers.72.mlp.experts.82.up_proj", "model.layers.72.mlp.experts.83.up_proj", "model.layers.72.mlp.experts.84.up_proj", "model.layers.72.mlp.experts.85.up_proj", "model.layers.72.mlp.experts.86.up_proj", "model.layers.72.mlp.experts.87.up_proj", "model.layers.72.mlp.experts.88.up_proj", "model.layers.72.mlp.experts.89.up_proj", "model.layers.72.mlp.experts.90.up_proj", "model.layers.72.mlp.experts.91.up_proj", "model.layers.72.mlp.experts.92.up_proj", "model.layers.72.mlp.experts.93.up_proj", "model.layers.72.mlp.experts.94.up_proj", "model.layers.72.mlp.experts.95.up_proj", "model.layers.72.mlp.experts.96.up_proj", "model.layers.72.mlp.experts.97.up_proj", "model.layers.72.mlp.experts.98.up_proj", "model.layers.72.mlp.experts.99.up_proj", "model.layers.72.mlp.experts.100.up_proj", "model.layers.72.mlp.experts.101.up_proj", "model.layers.72.mlp.experts.102.up_proj", "model.layers.72.mlp.experts.103.up_proj", "model.layers.72.mlp.experts.104.up_proj", "model.layers.72.mlp.experts.105.up_proj", "model.layers.72.mlp.experts.106.up_proj", "model.layers.72.mlp.experts.107.up_proj", "model.layers.72.mlp.experts.108.up_proj", "model.layers.72.mlp.experts.109.up_proj", "model.layers.72.mlp.experts.110.up_proj", "model.layers.72.mlp.experts.111.up_proj", "model.layers.72.mlp.experts.112.up_proj", "model.layers.72.mlp.experts.113.up_proj", "model.layers.72.mlp.experts.114.up_proj", "model.layers.72.mlp.experts.115.up_proj", "model.layers.72.mlp.experts.116.up_proj", "model.layers.72.mlp.experts.117.up_proj", "model.layers.72.mlp.experts.118.up_proj", "model.layers.72.mlp.experts.119.up_proj", "model.layers.72.mlp.experts.120.up_proj", "model.layers.72.mlp.experts.121.up_proj", "model.layers.72.mlp.experts.122.up_proj", "model.layers.72.mlp.experts.123.up_proj", "model.layers.72.mlp.experts.124.up_proj", "model.layers.72.mlp.experts.125.up_proj", "model.layers.72.mlp.experts.126.up_proj", "model.layers.72.mlp.experts.127.up_proj", "model.layers.72.mlp.experts.128.up_proj", "model.layers.72.mlp.experts.129.up_proj", "model.layers.72.mlp.experts.130.up_proj", "model.layers.72.mlp.experts.131.up_proj", "model.layers.72.mlp.experts.132.up_proj", "model.layers.72.mlp.experts.133.up_proj", "model.layers.72.mlp.experts.134.up_proj", "model.layers.72.mlp.experts.135.up_proj", "model.layers.72.mlp.experts.136.up_proj", "model.layers.72.mlp.experts.137.up_proj", "model.layers.72.mlp.experts.138.up_proj", "model.layers.72.mlp.experts.139.up_proj", "model.layers.72.mlp.experts.140.up_proj", "model.layers.72.mlp.experts.141.up_proj", "model.layers.72.mlp.experts.142.up_proj", "model.layers.72.mlp.experts.143.up_proj", "model.layers.72.mlp.experts.144.up_proj", "model.layers.72.mlp.experts.145.up_proj", "model.layers.72.mlp.experts.146.up_proj", "model.layers.72.mlp.experts.147.up_proj", "model.layers.72.mlp.experts.148.up_proj", "model.layers.72.mlp.experts.149.up_proj", "model.layers.72.mlp.experts.150.up_proj", "model.layers.72.mlp.experts.151.up_proj", "model.layers.72.mlp.experts.152.up_proj", "model.layers.72.mlp.experts.153.up_proj", "model.layers.72.mlp.experts.154.up_proj", "model.layers.72.mlp.experts.155.up_proj", "model.layers.72.mlp.experts.156.up_proj", "model.layers.72.mlp.experts.157.up_proj", "model.layers.72.mlp.experts.158.up_proj", "model.layers.72.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 0.0002147591672837651, "dbits": 2516582400 }, { "dkld": 0.00012964829802512845, "dbits": 5033164800 }, { "dkld": 0.00012745950371026438, "dbits": 7549747200 }, { "dkld": 0.00010389192029833239, "dbits": 12582912000 } ] }, { "idx": 434, "layers": [ "model.layers.72.mlp.experts.0.down_proj", "model.layers.72.mlp.experts.1.down_proj", "model.layers.72.mlp.experts.2.down_proj", "model.layers.72.mlp.experts.3.down_proj", "model.layers.72.mlp.experts.4.down_proj", "model.layers.72.mlp.experts.5.down_proj", "model.layers.72.mlp.experts.6.down_proj", "model.layers.72.mlp.experts.7.down_proj", "model.layers.72.mlp.experts.8.down_proj", "model.layers.72.mlp.experts.9.down_proj", "model.layers.72.mlp.experts.10.down_proj", "model.layers.72.mlp.experts.11.down_proj", "model.layers.72.mlp.experts.12.down_proj", "model.layers.72.mlp.experts.13.down_proj", "model.layers.72.mlp.experts.14.down_proj", "model.layers.72.mlp.experts.15.down_proj", "model.layers.72.mlp.experts.16.down_proj", "model.layers.72.mlp.experts.17.down_proj", "model.layers.72.mlp.experts.18.down_proj", "model.layers.72.mlp.experts.19.down_proj", "model.layers.72.mlp.experts.20.down_proj", "model.layers.72.mlp.experts.21.down_proj", "model.layers.72.mlp.experts.22.down_proj", "model.layers.72.mlp.experts.23.down_proj", "model.layers.72.mlp.experts.24.down_proj", "model.layers.72.mlp.experts.25.down_proj", "model.layers.72.mlp.experts.26.down_proj", "model.layers.72.mlp.experts.27.down_proj", "model.layers.72.mlp.experts.28.down_proj", "model.layers.72.mlp.experts.29.down_proj", "model.layers.72.mlp.experts.30.down_proj", "model.layers.72.mlp.experts.31.down_proj", "model.layers.72.mlp.experts.32.down_proj", "model.layers.72.mlp.experts.33.down_proj", "model.layers.72.mlp.experts.34.down_proj", "model.layers.72.mlp.experts.35.down_proj", "model.layers.72.mlp.experts.36.down_proj", "model.layers.72.mlp.experts.37.down_proj", "model.layers.72.mlp.experts.38.down_proj", "model.layers.72.mlp.experts.39.down_proj", "model.layers.72.mlp.experts.40.down_proj", "model.layers.72.mlp.experts.41.down_proj", "model.layers.72.mlp.experts.42.down_proj", "model.layers.72.mlp.experts.43.down_proj", "model.layers.72.mlp.experts.44.down_proj", "model.layers.72.mlp.experts.45.down_proj", "model.layers.72.mlp.experts.46.down_proj", "model.layers.72.mlp.experts.47.down_proj", "model.layers.72.mlp.experts.48.down_proj", "model.layers.72.mlp.experts.49.down_proj", "model.layers.72.mlp.experts.50.down_proj", "model.layers.72.mlp.experts.51.down_proj", "model.layers.72.mlp.experts.52.down_proj", "model.layers.72.mlp.experts.53.down_proj", "model.layers.72.mlp.experts.54.down_proj", "model.layers.72.mlp.experts.55.down_proj", "model.layers.72.mlp.experts.56.down_proj", "model.layers.72.mlp.experts.57.down_proj", "model.layers.72.mlp.experts.58.down_proj", "model.layers.72.mlp.experts.59.down_proj", "model.layers.72.mlp.experts.60.down_proj", "model.layers.72.mlp.experts.61.down_proj", "model.layers.72.mlp.experts.62.down_proj", "model.layers.72.mlp.experts.63.down_proj", "model.layers.72.mlp.experts.64.down_proj", "model.layers.72.mlp.experts.65.down_proj", "model.layers.72.mlp.experts.66.down_proj", "model.layers.72.mlp.experts.67.down_proj", "model.layers.72.mlp.experts.68.down_proj", "model.layers.72.mlp.experts.69.down_proj", "model.layers.72.mlp.experts.70.down_proj", "model.layers.72.mlp.experts.71.down_proj", "model.layers.72.mlp.experts.72.down_proj", "model.layers.72.mlp.experts.73.down_proj", "model.layers.72.mlp.experts.74.down_proj", "model.layers.72.mlp.experts.75.down_proj", "model.layers.72.mlp.experts.76.down_proj", "model.layers.72.mlp.experts.77.down_proj", "model.layers.72.mlp.experts.78.down_proj", "model.layers.72.mlp.experts.79.down_proj", "model.layers.72.mlp.experts.80.down_proj", "model.layers.72.mlp.experts.81.down_proj", "model.layers.72.mlp.experts.82.down_proj", "model.layers.72.mlp.experts.83.down_proj", "model.layers.72.mlp.experts.84.down_proj", "model.layers.72.mlp.experts.85.down_proj", "model.layers.72.mlp.experts.86.down_proj", "model.layers.72.mlp.experts.87.down_proj", "model.layers.72.mlp.experts.88.down_proj", "model.layers.72.mlp.experts.89.down_proj", "model.layers.72.mlp.experts.90.down_proj", "model.layers.72.mlp.experts.91.down_proj", "model.layers.72.mlp.experts.92.down_proj", "model.layers.72.mlp.experts.93.down_proj", "model.layers.72.mlp.experts.94.down_proj", "model.layers.72.mlp.experts.95.down_proj", "model.layers.72.mlp.experts.96.down_proj", "model.layers.72.mlp.experts.97.down_proj", "model.layers.72.mlp.experts.98.down_proj", "model.layers.72.mlp.experts.99.down_proj", "model.layers.72.mlp.experts.100.down_proj", "model.layers.72.mlp.experts.101.down_proj", "model.layers.72.mlp.experts.102.down_proj", "model.layers.72.mlp.experts.103.down_proj", "model.layers.72.mlp.experts.104.down_proj", "model.layers.72.mlp.experts.105.down_proj", "model.layers.72.mlp.experts.106.down_proj", "model.layers.72.mlp.experts.107.down_proj", "model.layers.72.mlp.experts.108.down_proj", "model.layers.72.mlp.experts.109.down_proj", "model.layers.72.mlp.experts.110.down_proj", "model.layers.72.mlp.experts.111.down_proj", "model.layers.72.mlp.experts.112.down_proj", "model.layers.72.mlp.experts.113.down_proj", "model.layers.72.mlp.experts.114.down_proj", "model.layers.72.mlp.experts.115.down_proj", "model.layers.72.mlp.experts.116.down_proj", "model.layers.72.mlp.experts.117.down_proj", "model.layers.72.mlp.experts.118.down_proj", "model.layers.72.mlp.experts.119.down_proj", "model.layers.72.mlp.experts.120.down_proj", "model.layers.72.mlp.experts.121.down_proj", "model.layers.72.mlp.experts.122.down_proj", "model.layers.72.mlp.experts.123.down_proj", "model.layers.72.mlp.experts.124.down_proj", "model.layers.72.mlp.experts.125.down_proj", "model.layers.72.mlp.experts.126.down_proj", "model.layers.72.mlp.experts.127.down_proj", "model.layers.72.mlp.experts.128.down_proj", "model.layers.72.mlp.experts.129.down_proj", "model.layers.72.mlp.experts.130.down_proj", "model.layers.72.mlp.experts.131.down_proj", "model.layers.72.mlp.experts.132.down_proj", "model.layers.72.mlp.experts.133.down_proj", "model.layers.72.mlp.experts.134.down_proj", "model.layers.72.mlp.experts.135.down_proj", "model.layers.72.mlp.experts.136.down_proj", "model.layers.72.mlp.experts.137.down_proj", "model.layers.72.mlp.experts.138.down_proj", "model.layers.72.mlp.experts.139.down_proj", "model.layers.72.mlp.experts.140.down_proj", "model.layers.72.mlp.experts.141.down_proj", "model.layers.72.mlp.experts.142.down_proj", "model.layers.72.mlp.experts.143.down_proj", "model.layers.72.mlp.experts.144.down_proj", "model.layers.72.mlp.experts.145.down_proj", "model.layers.72.mlp.experts.146.down_proj", "model.layers.72.mlp.experts.147.down_proj", "model.layers.72.mlp.experts.148.down_proj", "model.layers.72.mlp.experts.149.down_proj", "model.layers.72.mlp.experts.150.down_proj", "model.layers.72.mlp.experts.151.down_proj", "model.layers.72.mlp.experts.152.down_proj", "model.layers.72.mlp.experts.153.down_proj", "model.layers.72.mlp.experts.154.down_proj", "model.layers.72.mlp.experts.155.down_proj", "model.layers.72.mlp.experts.156.down_proj", "model.layers.72.mlp.experts.157.down_proj", "model.layers.72.mlp.experts.158.down_proj", "model.layers.72.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00018082372844219208, "dbits": 1258291200 }, { "dkld": -0.0002242259681224934, "dbits": 2516582400 }, { "dkld": -0.00025263447314501086, "dbits": 3774873600 }, { "dkld": -0.0002618903294205749, "dbits": 6291456000 } ] }, { "idx": 435, "layers": [ "model.layers.73.self_attn.q_proj" ], "candidates": [ { "dkld": -6.800973787904341e-05, "dbits": 62914560 }, { "dkld": -0.0001408737152814865, "dbits": 125829120 }, { "dkld": -9.855814278125763e-05, "dbits": 188743680 }, { "dkld": -0.00010819528251887478, "dbits": 314572800 } ] }, { "idx": 436, "layers": [ "model.layers.73.self_attn.k_proj", "model.layers.73.self_attn.v_proj" ], "candidates": [ { "dkld": -5.175797268748561e-05, "dbits": 10485760 }, { "dkld": 5.004862323403081e-05, "dbits": 20971520 }, { "dkld": 2.6604812592256888e-05, "dbits": 31457280 }, { "dkld": 1.548994332550846e-05, "dbits": 52428800 } ] }, { "idx": 437, "layers": [ "model.layers.73.self_attn.o_proj" ], "candidates": [ { "dkld": -0.000253116711974144, "dbits": 62914560 }, { "dkld": -0.00021683145314455032, "dbits": 125829120 }, { "dkld": -0.000133599154651165, "dbits": 188743680 }, { "dkld": -0.0001803200691938428, "dbits": 314572800 } ] }, { "idx": 438, "layers": [ "model.layers.73.mlp.shared_experts.gate_proj", "model.layers.73.mlp.shared_experts.up_proj", "model.layers.73.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0001889070495963152, "dbits": 23592960 }, { "dkld": -0.00028641344979406236, "dbits": 47185920 }, { "dkld": -0.000304486230015763, "dbits": 70778880 }, { "dkld": -0.00034198435023427565, "dbits": 117964800 } ] }, { "idx": 439, "layers": [ "model.layers.73.mlp.experts.0.gate_proj", "model.layers.73.mlp.experts.1.gate_proj", "model.layers.73.mlp.experts.2.gate_proj", "model.layers.73.mlp.experts.3.gate_proj", "model.layers.73.mlp.experts.4.gate_proj", "model.layers.73.mlp.experts.5.gate_proj", "model.layers.73.mlp.experts.6.gate_proj", "model.layers.73.mlp.experts.7.gate_proj", "model.layers.73.mlp.experts.8.gate_proj", "model.layers.73.mlp.experts.9.gate_proj", "model.layers.73.mlp.experts.10.gate_proj", "model.layers.73.mlp.experts.11.gate_proj", "model.layers.73.mlp.experts.12.gate_proj", "model.layers.73.mlp.experts.13.gate_proj", "model.layers.73.mlp.experts.14.gate_proj", "model.layers.73.mlp.experts.15.gate_proj", "model.layers.73.mlp.experts.16.gate_proj", "model.layers.73.mlp.experts.17.gate_proj", "model.layers.73.mlp.experts.18.gate_proj", "model.layers.73.mlp.experts.19.gate_proj", "model.layers.73.mlp.experts.20.gate_proj", "model.layers.73.mlp.experts.21.gate_proj", "model.layers.73.mlp.experts.22.gate_proj", "model.layers.73.mlp.experts.23.gate_proj", "model.layers.73.mlp.experts.24.gate_proj", "model.layers.73.mlp.experts.25.gate_proj", "model.layers.73.mlp.experts.26.gate_proj", "model.layers.73.mlp.experts.27.gate_proj", "model.layers.73.mlp.experts.28.gate_proj", "model.layers.73.mlp.experts.29.gate_proj", "model.layers.73.mlp.experts.30.gate_proj", "model.layers.73.mlp.experts.31.gate_proj", "model.layers.73.mlp.experts.32.gate_proj", "model.layers.73.mlp.experts.33.gate_proj", "model.layers.73.mlp.experts.34.gate_proj", "model.layers.73.mlp.experts.35.gate_proj", "model.layers.73.mlp.experts.36.gate_proj", "model.layers.73.mlp.experts.37.gate_proj", "model.layers.73.mlp.experts.38.gate_proj", "model.layers.73.mlp.experts.39.gate_proj", "model.layers.73.mlp.experts.40.gate_proj", "model.layers.73.mlp.experts.41.gate_proj", "model.layers.73.mlp.experts.42.gate_proj", "model.layers.73.mlp.experts.43.gate_proj", "model.layers.73.mlp.experts.44.gate_proj", "model.layers.73.mlp.experts.45.gate_proj", "model.layers.73.mlp.experts.46.gate_proj", "model.layers.73.mlp.experts.47.gate_proj", "model.layers.73.mlp.experts.48.gate_proj", "model.layers.73.mlp.experts.49.gate_proj", "model.layers.73.mlp.experts.50.gate_proj", "model.layers.73.mlp.experts.51.gate_proj", "model.layers.73.mlp.experts.52.gate_proj", "model.layers.73.mlp.experts.53.gate_proj", "model.layers.73.mlp.experts.54.gate_proj", "model.layers.73.mlp.experts.55.gate_proj", "model.layers.73.mlp.experts.56.gate_proj", "model.layers.73.mlp.experts.57.gate_proj", "model.layers.73.mlp.experts.58.gate_proj", "model.layers.73.mlp.experts.59.gate_proj", "model.layers.73.mlp.experts.60.gate_proj", "model.layers.73.mlp.experts.61.gate_proj", "model.layers.73.mlp.experts.62.gate_proj", "model.layers.73.mlp.experts.63.gate_proj", "model.layers.73.mlp.experts.64.gate_proj", "model.layers.73.mlp.experts.65.gate_proj", "model.layers.73.mlp.experts.66.gate_proj", "model.layers.73.mlp.experts.67.gate_proj", "model.layers.73.mlp.experts.68.gate_proj", "model.layers.73.mlp.experts.69.gate_proj", "model.layers.73.mlp.experts.70.gate_proj", "model.layers.73.mlp.experts.71.gate_proj", "model.layers.73.mlp.experts.72.gate_proj", "model.layers.73.mlp.experts.73.gate_proj", "model.layers.73.mlp.experts.74.gate_proj", "model.layers.73.mlp.experts.75.gate_proj", "model.layers.73.mlp.experts.76.gate_proj", "model.layers.73.mlp.experts.77.gate_proj", "model.layers.73.mlp.experts.78.gate_proj", "model.layers.73.mlp.experts.79.gate_proj", "model.layers.73.mlp.experts.80.gate_proj", "model.layers.73.mlp.experts.81.gate_proj", "model.layers.73.mlp.experts.82.gate_proj", "model.layers.73.mlp.experts.83.gate_proj", "model.layers.73.mlp.experts.84.gate_proj", "model.layers.73.mlp.experts.85.gate_proj", "model.layers.73.mlp.experts.86.gate_proj", "model.layers.73.mlp.experts.87.gate_proj", "model.layers.73.mlp.experts.88.gate_proj", "model.layers.73.mlp.experts.89.gate_proj", "model.layers.73.mlp.experts.90.gate_proj", "model.layers.73.mlp.experts.91.gate_proj", "model.layers.73.mlp.experts.92.gate_proj", "model.layers.73.mlp.experts.93.gate_proj", "model.layers.73.mlp.experts.94.gate_proj", "model.layers.73.mlp.experts.95.gate_proj", "model.layers.73.mlp.experts.96.gate_proj", "model.layers.73.mlp.experts.97.gate_proj", "model.layers.73.mlp.experts.98.gate_proj", "model.layers.73.mlp.experts.99.gate_proj", "model.layers.73.mlp.experts.100.gate_proj", "model.layers.73.mlp.experts.101.gate_proj", "model.layers.73.mlp.experts.102.gate_proj", "model.layers.73.mlp.experts.103.gate_proj", "model.layers.73.mlp.experts.104.gate_proj", "model.layers.73.mlp.experts.105.gate_proj", "model.layers.73.mlp.experts.106.gate_proj", "model.layers.73.mlp.experts.107.gate_proj", "model.layers.73.mlp.experts.108.gate_proj", "model.layers.73.mlp.experts.109.gate_proj", "model.layers.73.mlp.experts.110.gate_proj", "model.layers.73.mlp.experts.111.gate_proj", "model.layers.73.mlp.experts.112.gate_proj", "model.layers.73.mlp.experts.113.gate_proj", "model.layers.73.mlp.experts.114.gate_proj", "model.layers.73.mlp.experts.115.gate_proj", "model.layers.73.mlp.experts.116.gate_proj", "model.layers.73.mlp.experts.117.gate_proj", "model.layers.73.mlp.experts.118.gate_proj", "model.layers.73.mlp.experts.119.gate_proj", "model.layers.73.mlp.experts.120.gate_proj", "model.layers.73.mlp.experts.121.gate_proj", "model.layers.73.mlp.experts.122.gate_proj", "model.layers.73.mlp.experts.123.gate_proj", "model.layers.73.mlp.experts.124.gate_proj", "model.layers.73.mlp.experts.125.gate_proj", "model.layers.73.mlp.experts.126.gate_proj", "model.layers.73.mlp.experts.127.gate_proj", "model.layers.73.mlp.experts.128.gate_proj", "model.layers.73.mlp.experts.129.gate_proj", "model.layers.73.mlp.experts.130.gate_proj", "model.layers.73.mlp.experts.131.gate_proj", "model.layers.73.mlp.experts.132.gate_proj", "model.layers.73.mlp.experts.133.gate_proj", "model.layers.73.mlp.experts.134.gate_proj", "model.layers.73.mlp.experts.135.gate_proj", "model.layers.73.mlp.experts.136.gate_proj", "model.layers.73.mlp.experts.137.gate_proj", "model.layers.73.mlp.experts.138.gate_proj", "model.layers.73.mlp.experts.139.gate_proj", "model.layers.73.mlp.experts.140.gate_proj", "model.layers.73.mlp.experts.141.gate_proj", "model.layers.73.mlp.experts.142.gate_proj", "model.layers.73.mlp.experts.143.gate_proj", "model.layers.73.mlp.experts.144.gate_proj", "model.layers.73.mlp.experts.145.gate_proj", "model.layers.73.mlp.experts.146.gate_proj", "model.layers.73.mlp.experts.147.gate_proj", "model.layers.73.mlp.experts.148.gate_proj", "model.layers.73.mlp.experts.149.gate_proj", "model.layers.73.mlp.experts.150.gate_proj", "model.layers.73.mlp.experts.151.gate_proj", "model.layers.73.mlp.experts.152.gate_proj", "model.layers.73.mlp.experts.153.gate_proj", "model.layers.73.mlp.experts.154.gate_proj", "model.layers.73.mlp.experts.155.gate_proj", "model.layers.73.mlp.experts.156.gate_proj", "model.layers.73.mlp.experts.157.gate_proj", "model.layers.73.mlp.experts.158.gate_proj", "model.layers.73.mlp.experts.159.gate_proj", "model.layers.73.mlp.experts.0.up_proj", "model.layers.73.mlp.experts.1.up_proj", "model.layers.73.mlp.experts.2.up_proj", "model.layers.73.mlp.experts.3.up_proj", "model.layers.73.mlp.experts.4.up_proj", "model.layers.73.mlp.experts.5.up_proj", "model.layers.73.mlp.experts.6.up_proj", "model.layers.73.mlp.experts.7.up_proj", "model.layers.73.mlp.experts.8.up_proj", "model.layers.73.mlp.experts.9.up_proj", "model.layers.73.mlp.experts.10.up_proj", "model.layers.73.mlp.experts.11.up_proj", "model.layers.73.mlp.experts.12.up_proj", "model.layers.73.mlp.experts.13.up_proj", "model.layers.73.mlp.experts.14.up_proj", "model.layers.73.mlp.experts.15.up_proj", "model.layers.73.mlp.experts.16.up_proj", "model.layers.73.mlp.experts.17.up_proj", "model.layers.73.mlp.experts.18.up_proj", "model.layers.73.mlp.experts.19.up_proj", "model.layers.73.mlp.experts.20.up_proj", "model.layers.73.mlp.experts.21.up_proj", "model.layers.73.mlp.experts.22.up_proj", "model.layers.73.mlp.experts.23.up_proj", "model.layers.73.mlp.experts.24.up_proj", "model.layers.73.mlp.experts.25.up_proj", "model.layers.73.mlp.experts.26.up_proj", "model.layers.73.mlp.experts.27.up_proj", "model.layers.73.mlp.experts.28.up_proj", "model.layers.73.mlp.experts.29.up_proj", "model.layers.73.mlp.experts.30.up_proj", "model.layers.73.mlp.experts.31.up_proj", "model.layers.73.mlp.experts.32.up_proj", "model.layers.73.mlp.experts.33.up_proj", "model.layers.73.mlp.experts.34.up_proj", "model.layers.73.mlp.experts.35.up_proj", "model.layers.73.mlp.experts.36.up_proj", "model.layers.73.mlp.experts.37.up_proj", "model.layers.73.mlp.experts.38.up_proj", "model.layers.73.mlp.experts.39.up_proj", "model.layers.73.mlp.experts.40.up_proj", "model.layers.73.mlp.experts.41.up_proj", "model.layers.73.mlp.experts.42.up_proj", "model.layers.73.mlp.experts.43.up_proj", "model.layers.73.mlp.experts.44.up_proj", "model.layers.73.mlp.experts.45.up_proj", "model.layers.73.mlp.experts.46.up_proj", "model.layers.73.mlp.experts.47.up_proj", "model.layers.73.mlp.experts.48.up_proj", "model.layers.73.mlp.experts.49.up_proj", "model.layers.73.mlp.experts.50.up_proj", "model.layers.73.mlp.experts.51.up_proj", "model.layers.73.mlp.experts.52.up_proj", "model.layers.73.mlp.experts.53.up_proj", "model.layers.73.mlp.experts.54.up_proj", "model.layers.73.mlp.experts.55.up_proj", "model.layers.73.mlp.experts.56.up_proj", "model.layers.73.mlp.experts.57.up_proj", "model.layers.73.mlp.experts.58.up_proj", "model.layers.73.mlp.experts.59.up_proj", "model.layers.73.mlp.experts.60.up_proj", "model.layers.73.mlp.experts.61.up_proj", "model.layers.73.mlp.experts.62.up_proj", "model.layers.73.mlp.experts.63.up_proj", "model.layers.73.mlp.experts.64.up_proj", "model.layers.73.mlp.experts.65.up_proj", "model.layers.73.mlp.experts.66.up_proj", "model.layers.73.mlp.experts.67.up_proj", "model.layers.73.mlp.experts.68.up_proj", "model.layers.73.mlp.experts.69.up_proj", "model.layers.73.mlp.experts.70.up_proj", "model.layers.73.mlp.experts.71.up_proj", "model.layers.73.mlp.experts.72.up_proj", "model.layers.73.mlp.experts.73.up_proj", "model.layers.73.mlp.experts.74.up_proj", "model.layers.73.mlp.experts.75.up_proj", "model.layers.73.mlp.experts.76.up_proj", "model.layers.73.mlp.experts.77.up_proj", "model.layers.73.mlp.experts.78.up_proj", "model.layers.73.mlp.experts.79.up_proj", "model.layers.73.mlp.experts.80.up_proj", "model.layers.73.mlp.experts.81.up_proj", "model.layers.73.mlp.experts.82.up_proj", "model.layers.73.mlp.experts.83.up_proj", "model.layers.73.mlp.experts.84.up_proj", "model.layers.73.mlp.experts.85.up_proj", "model.layers.73.mlp.experts.86.up_proj", "model.layers.73.mlp.experts.87.up_proj", "model.layers.73.mlp.experts.88.up_proj", "model.layers.73.mlp.experts.89.up_proj", "model.layers.73.mlp.experts.90.up_proj", "model.layers.73.mlp.experts.91.up_proj", "model.layers.73.mlp.experts.92.up_proj", "model.layers.73.mlp.experts.93.up_proj", "model.layers.73.mlp.experts.94.up_proj", "model.layers.73.mlp.experts.95.up_proj", "model.layers.73.mlp.experts.96.up_proj", "model.layers.73.mlp.experts.97.up_proj", "model.layers.73.mlp.experts.98.up_proj", "model.layers.73.mlp.experts.99.up_proj", "model.layers.73.mlp.experts.100.up_proj", "model.layers.73.mlp.experts.101.up_proj", "model.layers.73.mlp.experts.102.up_proj", "model.layers.73.mlp.experts.103.up_proj", "model.layers.73.mlp.experts.104.up_proj", "model.layers.73.mlp.experts.105.up_proj", "model.layers.73.mlp.experts.106.up_proj", "model.layers.73.mlp.experts.107.up_proj", "model.layers.73.mlp.experts.108.up_proj", "model.layers.73.mlp.experts.109.up_proj", "model.layers.73.mlp.experts.110.up_proj", "model.layers.73.mlp.experts.111.up_proj", "model.layers.73.mlp.experts.112.up_proj", "model.layers.73.mlp.experts.113.up_proj", "model.layers.73.mlp.experts.114.up_proj", "model.layers.73.mlp.experts.115.up_proj", "model.layers.73.mlp.experts.116.up_proj", "model.layers.73.mlp.experts.117.up_proj", "model.layers.73.mlp.experts.118.up_proj", "model.layers.73.mlp.experts.119.up_proj", "model.layers.73.mlp.experts.120.up_proj", "model.layers.73.mlp.experts.121.up_proj", "model.layers.73.mlp.experts.122.up_proj", "model.layers.73.mlp.experts.123.up_proj", "model.layers.73.mlp.experts.124.up_proj", "model.layers.73.mlp.experts.125.up_proj", "model.layers.73.mlp.experts.126.up_proj", "model.layers.73.mlp.experts.127.up_proj", "model.layers.73.mlp.experts.128.up_proj", "model.layers.73.mlp.experts.129.up_proj", "model.layers.73.mlp.experts.130.up_proj", "model.layers.73.mlp.experts.131.up_proj", "model.layers.73.mlp.experts.132.up_proj", "model.layers.73.mlp.experts.133.up_proj", "model.layers.73.mlp.experts.134.up_proj", "model.layers.73.mlp.experts.135.up_proj", "model.layers.73.mlp.experts.136.up_proj", "model.layers.73.mlp.experts.137.up_proj", "model.layers.73.mlp.experts.138.up_proj", "model.layers.73.mlp.experts.139.up_proj", "model.layers.73.mlp.experts.140.up_proj", "model.layers.73.mlp.experts.141.up_proj", "model.layers.73.mlp.experts.142.up_proj", "model.layers.73.mlp.experts.143.up_proj", "model.layers.73.mlp.experts.144.up_proj", "model.layers.73.mlp.experts.145.up_proj", "model.layers.73.mlp.experts.146.up_proj", "model.layers.73.mlp.experts.147.up_proj", "model.layers.73.mlp.experts.148.up_proj", "model.layers.73.mlp.experts.149.up_proj", "model.layers.73.mlp.experts.150.up_proj", "model.layers.73.mlp.experts.151.up_proj", "model.layers.73.mlp.experts.152.up_proj", "model.layers.73.mlp.experts.153.up_proj", "model.layers.73.mlp.experts.154.up_proj", "model.layers.73.mlp.experts.155.up_proj", "model.layers.73.mlp.experts.156.up_proj", "model.layers.73.mlp.experts.157.up_proj", "model.layers.73.mlp.experts.158.up_proj", "model.layers.73.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00020697722211480973, "dbits": 2516582400 }, { "dkld": -0.00016780951991678117, "dbits": 5033164800 }, { "dkld": -0.00017785625532269755, "dbits": 7549747200 }, { "dkld": -0.00018410971388221897, "dbits": 12582912000 } ] }, { "idx": 440, "layers": [ "model.layers.73.mlp.experts.0.down_proj", "model.layers.73.mlp.experts.1.down_proj", "model.layers.73.mlp.experts.2.down_proj", "model.layers.73.mlp.experts.3.down_proj", "model.layers.73.mlp.experts.4.down_proj", "model.layers.73.mlp.experts.5.down_proj", "model.layers.73.mlp.experts.6.down_proj", "model.layers.73.mlp.experts.7.down_proj", "model.layers.73.mlp.experts.8.down_proj", "model.layers.73.mlp.experts.9.down_proj", "model.layers.73.mlp.experts.10.down_proj", "model.layers.73.mlp.experts.11.down_proj", "model.layers.73.mlp.experts.12.down_proj", "model.layers.73.mlp.experts.13.down_proj", "model.layers.73.mlp.experts.14.down_proj", "model.layers.73.mlp.experts.15.down_proj", "model.layers.73.mlp.experts.16.down_proj", "model.layers.73.mlp.experts.17.down_proj", "model.layers.73.mlp.experts.18.down_proj", "model.layers.73.mlp.experts.19.down_proj", "model.layers.73.mlp.experts.20.down_proj", "model.layers.73.mlp.experts.21.down_proj", "model.layers.73.mlp.experts.22.down_proj", "model.layers.73.mlp.experts.23.down_proj", "model.layers.73.mlp.experts.24.down_proj", "model.layers.73.mlp.experts.25.down_proj", "model.layers.73.mlp.experts.26.down_proj", "model.layers.73.mlp.experts.27.down_proj", "model.layers.73.mlp.experts.28.down_proj", "model.layers.73.mlp.experts.29.down_proj", "model.layers.73.mlp.experts.30.down_proj", "model.layers.73.mlp.experts.31.down_proj", "model.layers.73.mlp.experts.32.down_proj", "model.layers.73.mlp.experts.33.down_proj", "model.layers.73.mlp.experts.34.down_proj", "model.layers.73.mlp.experts.35.down_proj", "model.layers.73.mlp.experts.36.down_proj", "model.layers.73.mlp.experts.37.down_proj", "model.layers.73.mlp.experts.38.down_proj", "model.layers.73.mlp.experts.39.down_proj", "model.layers.73.mlp.experts.40.down_proj", "model.layers.73.mlp.experts.41.down_proj", "model.layers.73.mlp.experts.42.down_proj", "model.layers.73.mlp.experts.43.down_proj", "model.layers.73.mlp.experts.44.down_proj", "model.layers.73.mlp.experts.45.down_proj", "model.layers.73.mlp.experts.46.down_proj", "model.layers.73.mlp.experts.47.down_proj", "model.layers.73.mlp.experts.48.down_proj", "model.layers.73.mlp.experts.49.down_proj", "model.layers.73.mlp.experts.50.down_proj", "model.layers.73.mlp.experts.51.down_proj", "model.layers.73.mlp.experts.52.down_proj", "model.layers.73.mlp.experts.53.down_proj", "model.layers.73.mlp.experts.54.down_proj", "model.layers.73.mlp.experts.55.down_proj", "model.layers.73.mlp.experts.56.down_proj", "model.layers.73.mlp.experts.57.down_proj", "model.layers.73.mlp.experts.58.down_proj", "model.layers.73.mlp.experts.59.down_proj", "model.layers.73.mlp.experts.60.down_proj", "model.layers.73.mlp.experts.61.down_proj", "model.layers.73.mlp.experts.62.down_proj", "model.layers.73.mlp.experts.63.down_proj", "model.layers.73.mlp.experts.64.down_proj", "model.layers.73.mlp.experts.65.down_proj", "model.layers.73.mlp.experts.66.down_proj", "model.layers.73.mlp.experts.67.down_proj", "model.layers.73.mlp.experts.68.down_proj", "model.layers.73.mlp.experts.69.down_proj", "model.layers.73.mlp.experts.70.down_proj", "model.layers.73.mlp.experts.71.down_proj", "model.layers.73.mlp.experts.72.down_proj", "model.layers.73.mlp.experts.73.down_proj", "model.layers.73.mlp.experts.74.down_proj", "model.layers.73.mlp.experts.75.down_proj", "model.layers.73.mlp.experts.76.down_proj", "model.layers.73.mlp.experts.77.down_proj", "model.layers.73.mlp.experts.78.down_proj", "model.layers.73.mlp.experts.79.down_proj", "model.layers.73.mlp.experts.80.down_proj", "model.layers.73.mlp.experts.81.down_proj", "model.layers.73.mlp.experts.82.down_proj", "model.layers.73.mlp.experts.83.down_proj", "model.layers.73.mlp.experts.84.down_proj", "model.layers.73.mlp.experts.85.down_proj", "model.layers.73.mlp.experts.86.down_proj", "model.layers.73.mlp.experts.87.down_proj", "model.layers.73.mlp.experts.88.down_proj", "model.layers.73.mlp.experts.89.down_proj", "model.layers.73.mlp.experts.90.down_proj", "model.layers.73.mlp.experts.91.down_proj", "model.layers.73.mlp.experts.92.down_proj", "model.layers.73.mlp.experts.93.down_proj", "model.layers.73.mlp.experts.94.down_proj", "model.layers.73.mlp.experts.95.down_proj", "model.layers.73.mlp.experts.96.down_proj", "model.layers.73.mlp.experts.97.down_proj", "model.layers.73.mlp.experts.98.down_proj", "model.layers.73.mlp.experts.99.down_proj", "model.layers.73.mlp.experts.100.down_proj", "model.layers.73.mlp.experts.101.down_proj", "model.layers.73.mlp.experts.102.down_proj", "model.layers.73.mlp.experts.103.down_proj", "model.layers.73.mlp.experts.104.down_proj", "model.layers.73.mlp.experts.105.down_proj", "model.layers.73.mlp.experts.106.down_proj", "model.layers.73.mlp.experts.107.down_proj", "model.layers.73.mlp.experts.108.down_proj", "model.layers.73.mlp.experts.109.down_proj", "model.layers.73.mlp.experts.110.down_proj", "model.layers.73.mlp.experts.111.down_proj", "model.layers.73.mlp.experts.112.down_proj", "model.layers.73.mlp.experts.113.down_proj", "model.layers.73.mlp.experts.114.down_proj", "model.layers.73.mlp.experts.115.down_proj", "model.layers.73.mlp.experts.116.down_proj", "model.layers.73.mlp.experts.117.down_proj", "model.layers.73.mlp.experts.118.down_proj", "model.layers.73.mlp.experts.119.down_proj", "model.layers.73.mlp.experts.120.down_proj", "model.layers.73.mlp.experts.121.down_proj", "model.layers.73.mlp.experts.122.down_proj", "model.layers.73.mlp.experts.123.down_proj", "model.layers.73.mlp.experts.124.down_proj", "model.layers.73.mlp.experts.125.down_proj", "model.layers.73.mlp.experts.126.down_proj", "model.layers.73.mlp.experts.127.down_proj", "model.layers.73.mlp.experts.128.down_proj", "model.layers.73.mlp.experts.129.down_proj", "model.layers.73.mlp.experts.130.down_proj", "model.layers.73.mlp.experts.131.down_proj", "model.layers.73.mlp.experts.132.down_proj", "model.layers.73.mlp.experts.133.down_proj", "model.layers.73.mlp.experts.134.down_proj", "model.layers.73.mlp.experts.135.down_proj", "model.layers.73.mlp.experts.136.down_proj", "model.layers.73.mlp.experts.137.down_proj", "model.layers.73.mlp.experts.138.down_proj", "model.layers.73.mlp.experts.139.down_proj", "model.layers.73.mlp.experts.140.down_proj", "model.layers.73.mlp.experts.141.down_proj", "model.layers.73.mlp.experts.142.down_proj", "model.layers.73.mlp.experts.143.down_proj", "model.layers.73.mlp.experts.144.down_proj", "model.layers.73.mlp.experts.145.down_proj", "model.layers.73.mlp.experts.146.down_proj", "model.layers.73.mlp.experts.147.down_proj", "model.layers.73.mlp.experts.148.down_proj", "model.layers.73.mlp.experts.149.down_proj", "model.layers.73.mlp.experts.150.down_proj", "model.layers.73.mlp.experts.151.down_proj", "model.layers.73.mlp.experts.152.down_proj", "model.layers.73.mlp.experts.153.down_proj", "model.layers.73.mlp.experts.154.down_proj", "model.layers.73.mlp.experts.155.down_proj", "model.layers.73.mlp.experts.156.down_proj", "model.layers.73.mlp.experts.157.down_proj", "model.layers.73.mlp.experts.158.down_proj", "model.layers.73.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00011574346572161553, "dbits": 1258291200 }, { "dkld": -0.00014568241313100971, "dbits": 2516582400 }, { "dkld": -0.00013454295694828589, "dbits": 3774873600 }, { "dkld": -0.00014976430684328357, "dbits": 6291456000 } ] }, { "idx": 441, "layers": [ "model.layers.74.self_attn.q_proj" ], "candidates": [ { "dkld": -5.6324806064378397e-05, "dbits": 62914560 }, { "dkld": -1.855595037342228e-05, "dbits": 125829120 }, { "dkld": -1.604724675417779e-05, "dbits": 188743680 }, { "dkld": -1.6021821647893564e-05, "dbits": 314572800 } ] }, { "idx": 442, "layers": [ "model.layers.74.self_attn.k_proj", "model.layers.74.self_attn.v_proj" ], "candidates": [ { "dkld": -8.753873407840729e-05, "dbits": 10485760 }, { "dkld": -0.00012207403779029846, "dbits": 20971520 }, { "dkld": -8.31768848001957e-05, "dbits": 31457280 }, { "dkld": -0.00010654041543603499, "dbits": 52428800 } ] }, { "idx": 443, "layers": [ "model.layers.74.self_attn.o_proj" ], "candidates": [ { "dkld": -7.100217044364587e-06, "dbits": 62914560 }, { "dkld": 4.315488040446125e-05, "dbits": 125829120 }, { "dkld": 6.505027413367115e-05, "dbits": 188743680 }, { "dkld": 6.44514337181934e-05, "dbits": 314572800 } ] }, { "idx": 444, "layers": [ "model.layers.74.mlp.shared_experts.gate_proj", "model.layers.74.mlp.shared_experts.up_proj", "model.layers.74.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00018845731392502785, "dbits": 23592960 }, { "dkld": -0.00010099820792675018, "dbits": 47185920 }, { "dkld": -0.00011910106986762203, "dbits": 70778880 }, { "dkld": -0.00013827886432410397, "dbits": 117964800 } ] }, { "idx": 445, "layers": [ "model.layers.74.mlp.experts.0.gate_proj", "model.layers.74.mlp.experts.1.gate_proj", "model.layers.74.mlp.experts.2.gate_proj", "model.layers.74.mlp.experts.3.gate_proj", "model.layers.74.mlp.experts.4.gate_proj", "model.layers.74.mlp.experts.5.gate_proj", "model.layers.74.mlp.experts.6.gate_proj", "model.layers.74.mlp.experts.7.gate_proj", "model.layers.74.mlp.experts.8.gate_proj", "model.layers.74.mlp.experts.9.gate_proj", "model.layers.74.mlp.experts.10.gate_proj", "model.layers.74.mlp.experts.11.gate_proj", "model.layers.74.mlp.experts.12.gate_proj", "model.layers.74.mlp.experts.13.gate_proj", "model.layers.74.mlp.experts.14.gate_proj", "model.layers.74.mlp.experts.15.gate_proj", "model.layers.74.mlp.experts.16.gate_proj", "model.layers.74.mlp.experts.17.gate_proj", "model.layers.74.mlp.experts.18.gate_proj", "model.layers.74.mlp.experts.19.gate_proj", "model.layers.74.mlp.experts.20.gate_proj", "model.layers.74.mlp.experts.21.gate_proj", "model.layers.74.mlp.experts.22.gate_proj", "model.layers.74.mlp.experts.23.gate_proj", "model.layers.74.mlp.experts.24.gate_proj", "model.layers.74.mlp.experts.25.gate_proj", "model.layers.74.mlp.experts.26.gate_proj", "model.layers.74.mlp.experts.27.gate_proj", "model.layers.74.mlp.experts.28.gate_proj", "model.layers.74.mlp.experts.29.gate_proj", "model.layers.74.mlp.experts.30.gate_proj", "model.layers.74.mlp.experts.31.gate_proj", "model.layers.74.mlp.experts.32.gate_proj", "model.layers.74.mlp.experts.33.gate_proj", "model.layers.74.mlp.experts.34.gate_proj", "model.layers.74.mlp.experts.35.gate_proj", "model.layers.74.mlp.experts.36.gate_proj", "model.layers.74.mlp.experts.37.gate_proj", "model.layers.74.mlp.experts.38.gate_proj", "model.layers.74.mlp.experts.39.gate_proj", "model.layers.74.mlp.experts.40.gate_proj", "model.layers.74.mlp.experts.41.gate_proj", "model.layers.74.mlp.experts.42.gate_proj", "model.layers.74.mlp.experts.43.gate_proj", "model.layers.74.mlp.experts.44.gate_proj", "model.layers.74.mlp.experts.45.gate_proj", "model.layers.74.mlp.experts.46.gate_proj", "model.layers.74.mlp.experts.47.gate_proj", "model.layers.74.mlp.experts.48.gate_proj", "model.layers.74.mlp.experts.49.gate_proj", "model.layers.74.mlp.experts.50.gate_proj", "model.layers.74.mlp.experts.51.gate_proj", "model.layers.74.mlp.experts.52.gate_proj", "model.layers.74.mlp.experts.53.gate_proj", "model.layers.74.mlp.experts.54.gate_proj", "model.layers.74.mlp.experts.55.gate_proj", "model.layers.74.mlp.experts.56.gate_proj", "model.layers.74.mlp.experts.57.gate_proj", "model.layers.74.mlp.experts.58.gate_proj", "model.layers.74.mlp.experts.59.gate_proj", "model.layers.74.mlp.experts.60.gate_proj", "model.layers.74.mlp.experts.61.gate_proj", "model.layers.74.mlp.experts.62.gate_proj", "model.layers.74.mlp.experts.63.gate_proj", "model.layers.74.mlp.experts.64.gate_proj", "model.layers.74.mlp.experts.65.gate_proj", "model.layers.74.mlp.experts.66.gate_proj", "model.layers.74.mlp.experts.67.gate_proj", "model.layers.74.mlp.experts.68.gate_proj", "model.layers.74.mlp.experts.69.gate_proj", "model.layers.74.mlp.experts.70.gate_proj", "model.layers.74.mlp.experts.71.gate_proj", "model.layers.74.mlp.experts.72.gate_proj", "model.layers.74.mlp.experts.73.gate_proj", "model.layers.74.mlp.experts.74.gate_proj", "model.layers.74.mlp.experts.75.gate_proj", "model.layers.74.mlp.experts.76.gate_proj", "model.layers.74.mlp.experts.77.gate_proj", "model.layers.74.mlp.experts.78.gate_proj", "model.layers.74.mlp.experts.79.gate_proj", "model.layers.74.mlp.experts.80.gate_proj", "model.layers.74.mlp.experts.81.gate_proj", "model.layers.74.mlp.experts.82.gate_proj", "model.layers.74.mlp.experts.83.gate_proj", "model.layers.74.mlp.experts.84.gate_proj", "model.layers.74.mlp.experts.85.gate_proj", "model.layers.74.mlp.experts.86.gate_proj", "model.layers.74.mlp.experts.87.gate_proj", "model.layers.74.mlp.experts.88.gate_proj", "model.layers.74.mlp.experts.89.gate_proj", "model.layers.74.mlp.experts.90.gate_proj", "model.layers.74.mlp.experts.91.gate_proj", "model.layers.74.mlp.experts.92.gate_proj", "model.layers.74.mlp.experts.93.gate_proj", "model.layers.74.mlp.experts.94.gate_proj", "model.layers.74.mlp.experts.95.gate_proj", "model.layers.74.mlp.experts.96.gate_proj", "model.layers.74.mlp.experts.97.gate_proj", "model.layers.74.mlp.experts.98.gate_proj", "model.layers.74.mlp.experts.99.gate_proj", "model.layers.74.mlp.experts.100.gate_proj", "model.layers.74.mlp.experts.101.gate_proj", "model.layers.74.mlp.experts.102.gate_proj", "model.layers.74.mlp.experts.103.gate_proj", "model.layers.74.mlp.experts.104.gate_proj", "model.layers.74.mlp.experts.105.gate_proj", "model.layers.74.mlp.experts.106.gate_proj", "model.layers.74.mlp.experts.107.gate_proj", "model.layers.74.mlp.experts.108.gate_proj", "model.layers.74.mlp.experts.109.gate_proj", "model.layers.74.mlp.experts.110.gate_proj", "model.layers.74.mlp.experts.111.gate_proj", "model.layers.74.mlp.experts.112.gate_proj", "model.layers.74.mlp.experts.113.gate_proj", "model.layers.74.mlp.experts.114.gate_proj", "model.layers.74.mlp.experts.115.gate_proj", "model.layers.74.mlp.experts.116.gate_proj", "model.layers.74.mlp.experts.117.gate_proj", "model.layers.74.mlp.experts.118.gate_proj", "model.layers.74.mlp.experts.119.gate_proj", "model.layers.74.mlp.experts.120.gate_proj", "model.layers.74.mlp.experts.121.gate_proj", "model.layers.74.mlp.experts.122.gate_proj", "model.layers.74.mlp.experts.123.gate_proj", "model.layers.74.mlp.experts.124.gate_proj", "model.layers.74.mlp.experts.125.gate_proj", "model.layers.74.mlp.experts.126.gate_proj", "model.layers.74.mlp.experts.127.gate_proj", "model.layers.74.mlp.experts.128.gate_proj", "model.layers.74.mlp.experts.129.gate_proj", "model.layers.74.mlp.experts.130.gate_proj", "model.layers.74.mlp.experts.131.gate_proj", "model.layers.74.mlp.experts.132.gate_proj", "model.layers.74.mlp.experts.133.gate_proj", "model.layers.74.mlp.experts.134.gate_proj", "model.layers.74.mlp.experts.135.gate_proj", "model.layers.74.mlp.experts.136.gate_proj", "model.layers.74.mlp.experts.137.gate_proj", "model.layers.74.mlp.experts.138.gate_proj", "model.layers.74.mlp.experts.139.gate_proj", "model.layers.74.mlp.experts.140.gate_proj", "model.layers.74.mlp.experts.141.gate_proj", "model.layers.74.mlp.experts.142.gate_proj", "model.layers.74.mlp.experts.143.gate_proj", "model.layers.74.mlp.experts.144.gate_proj", "model.layers.74.mlp.experts.145.gate_proj", "model.layers.74.mlp.experts.146.gate_proj", "model.layers.74.mlp.experts.147.gate_proj", "model.layers.74.mlp.experts.148.gate_proj", "model.layers.74.mlp.experts.149.gate_proj", "model.layers.74.mlp.experts.150.gate_proj", "model.layers.74.mlp.experts.151.gate_proj", "model.layers.74.mlp.experts.152.gate_proj", "model.layers.74.mlp.experts.153.gate_proj", "model.layers.74.mlp.experts.154.gate_proj", "model.layers.74.mlp.experts.155.gate_proj", "model.layers.74.mlp.experts.156.gate_proj", "model.layers.74.mlp.experts.157.gate_proj", "model.layers.74.mlp.experts.158.gate_proj", "model.layers.74.mlp.experts.159.gate_proj", "model.layers.74.mlp.experts.0.up_proj", "model.layers.74.mlp.experts.1.up_proj", "model.layers.74.mlp.experts.2.up_proj", "model.layers.74.mlp.experts.3.up_proj", "model.layers.74.mlp.experts.4.up_proj", "model.layers.74.mlp.experts.5.up_proj", "model.layers.74.mlp.experts.6.up_proj", "model.layers.74.mlp.experts.7.up_proj", "model.layers.74.mlp.experts.8.up_proj", "model.layers.74.mlp.experts.9.up_proj", "model.layers.74.mlp.experts.10.up_proj", "model.layers.74.mlp.experts.11.up_proj", "model.layers.74.mlp.experts.12.up_proj", "model.layers.74.mlp.experts.13.up_proj", "model.layers.74.mlp.experts.14.up_proj", "model.layers.74.mlp.experts.15.up_proj", "model.layers.74.mlp.experts.16.up_proj", "model.layers.74.mlp.experts.17.up_proj", "model.layers.74.mlp.experts.18.up_proj", "model.layers.74.mlp.experts.19.up_proj", "model.layers.74.mlp.experts.20.up_proj", "model.layers.74.mlp.experts.21.up_proj", "model.layers.74.mlp.experts.22.up_proj", "model.layers.74.mlp.experts.23.up_proj", "model.layers.74.mlp.experts.24.up_proj", "model.layers.74.mlp.experts.25.up_proj", "model.layers.74.mlp.experts.26.up_proj", "model.layers.74.mlp.experts.27.up_proj", "model.layers.74.mlp.experts.28.up_proj", "model.layers.74.mlp.experts.29.up_proj", "model.layers.74.mlp.experts.30.up_proj", "model.layers.74.mlp.experts.31.up_proj", "model.layers.74.mlp.experts.32.up_proj", "model.layers.74.mlp.experts.33.up_proj", "model.layers.74.mlp.experts.34.up_proj", "model.layers.74.mlp.experts.35.up_proj", "model.layers.74.mlp.experts.36.up_proj", "model.layers.74.mlp.experts.37.up_proj", "model.layers.74.mlp.experts.38.up_proj", "model.layers.74.mlp.experts.39.up_proj", "model.layers.74.mlp.experts.40.up_proj", "model.layers.74.mlp.experts.41.up_proj", "model.layers.74.mlp.experts.42.up_proj", "model.layers.74.mlp.experts.43.up_proj", "model.layers.74.mlp.experts.44.up_proj", "model.layers.74.mlp.experts.45.up_proj", "model.layers.74.mlp.experts.46.up_proj", "model.layers.74.mlp.experts.47.up_proj", "model.layers.74.mlp.experts.48.up_proj", "model.layers.74.mlp.experts.49.up_proj", "model.layers.74.mlp.experts.50.up_proj", "model.layers.74.mlp.experts.51.up_proj", "model.layers.74.mlp.experts.52.up_proj", "model.layers.74.mlp.experts.53.up_proj", "model.layers.74.mlp.experts.54.up_proj", "model.layers.74.mlp.experts.55.up_proj", "model.layers.74.mlp.experts.56.up_proj", "model.layers.74.mlp.experts.57.up_proj", "model.layers.74.mlp.experts.58.up_proj", "model.layers.74.mlp.experts.59.up_proj", "model.layers.74.mlp.experts.60.up_proj", "model.layers.74.mlp.experts.61.up_proj", "model.layers.74.mlp.experts.62.up_proj", "model.layers.74.mlp.experts.63.up_proj", "model.layers.74.mlp.experts.64.up_proj", "model.layers.74.mlp.experts.65.up_proj", "model.layers.74.mlp.experts.66.up_proj", "model.layers.74.mlp.experts.67.up_proj", "model.layers.74.mlp.experts.68.up_proj", "model.layers.74.mlp.experts.69.up_proj", "model.layers.74.mlp.experts.70.up_proj", "model.layers.74.mlp.experts.71.up_proj", "model.layers.74.mlp.experts.72.up_proj", "model.layers.74.mlp.experts.73.up_proj", "model.layers.74.mlp.experts.74.up_proj", "model.layers.74.mlp.experts.75.up_proj", "model.layers.74.mlp.experts.76.up_proj", "model.layers.74.mlp.experts.77.up_proj", "model.layers.74.mlp.experts.78.up_proj", "model.layers.74.mlp.experts.79.up_proj", "model.layers.74.mlp.experts.80.up_proj", "model.layers.74.mlp.experts.81.up_proj", "model.layers.74.mlp.experts.82.up_proj", "model.layers.74.mlp.experts.83.up_proj", "model.layers.74.mlp.experts.84.up_proj", "model.layers.74.mlp.experts.85.up_proj", "model.layers.74.mlp.experts.86.up_proj", "model.layers.74.mlp.experts.87.up_proj", "model.layers.74.mlp.experts.88.up_proj", "model.layers.74.mlp.experts.89.up_proj", "model.layers.74.mlp.experts.90.up_proj", "model.layers.74.mlp.experts.91.up_proj", "model.layers.74.mlp.experts.92.up_proj", "model.layers.74.mlp.experts.93.up_proj", "model.layers.74.mlp.experts.94.up_proj", "model.layers.74.mlp.experts.95.up_proj", "model.layers.74.mlp.experts.96.up_proj", "model.layers.74.mlp.experts.97.up_proj", "model.layers.74.mlp.experts.98.up_proj", "model.layers.74.mlp.experts.99.up_proj", "model.layers.74.mlp.experts.100.up_proj", "model.layers.74.mlp.experts.101.up_proj", "model.layers.74.mlp.experts.102.up_proj", "model.layers.74.mlp.experts.103.up_proj", "model.layers.74.mlp.experts.104.up_proj", "model.layers.74.mlp.experts.105.up_proj", "model.layers.74.mlp.experts.106.up_proj", "model.layers.74.mlp.experts.107.up_proj", "model.layers.74.mlp.experts.108.up_proj", "model.layers.74.mlp.experts.109.up_proj", "model.layers.74.mlp.experts.110.up_proj", "model.layers.74.mlp.experts.111.up_proj", "model.layers.74.mlp.experts.112.up_proj", "model.layers.74.mlp.experts.113.up_proj", "model.layers.74.mlp.experts.114.up_proj", "model.layers.74.mlp.experts.115.up_proj", "model.layers.74.mlp.experts.116.up_proj", "model.layers.74.mlp.experts.117.up_proj", "model.layers.74.mlp.experts.118.up_proj", "model.layers.74.mlp.experts.119.up_proj", "model.layers.74.mlp.experts.120.up_proj", "model.layers.74.mlp.experts.121.up_proj", "model.layers.74.mlp.experts.122.up_proj", "model.layers.74.mlp.experts.123.up_proj", "model.layers.74.mlp.experts.124.up_proj", "model.layers.74.mlp.experts.125.up_proj", "model.layers.74.mlp.experts.126.up_proj", "model.layers.74.mlp.experts.127.up_proj", "model.layers.74.mlp.experts.128.up_proj", "model.layers.74.mlp.experts.129.up_proj", "model.layers.74.mlp.experts.130.up_proj", "model.layers.74.mlp.experts.131.up_proj", "model.layers.74.mlp.experts.132.up_proj", "model.layers.74.mlp.experts.133.up_proj", "model.layers.74.mlp.experts.134.up_proj", "model.layers.74.mlp.experts.135.up_proj", "model.layers.74.mlp.experts.136.up_proj", "model.layers.74.mlp.experts.137.up_proj", "model.layers.74.mlp.experts.138.up_proj", "model.layers.74.mlp.experts.139.up_proj", "model.layers.74.mlp.experts.140.up_proj", "model.layers.74.mlp.experts.141.up_proj", "model.layers.74.mlp.experts.142.up_proj", "model.layers.74.mlp.experts.143.up_proj", "model.layers.74.mlp.experts.144.up_proj", "model.layers.74.mlp.experts.145.up_proj", "model.layers.74.mlp.experts.146.up_proj", "model.layers.74.mlp.experts.147.up_proj", "model.layers.74.mlp.experts.148.up_proj", "model.layers.74.mlp.experts.149.up_proj", "model.layers.74.mlp.experts.150.up_proj", "model.layers.74.mlp.experts.151.up_proj", "model.layers.74.mlp.experts.152.up_proj", "model.layers.74.mlp.experts.153.up_proj", "model.layers.74.mlp.experts.154.up_proj", "model.layers.74.mlp.experts.155.up_proj", "model.layers.74.mlp.experts.156.up_proj", "model.layers.74.mlp.experts.157.up_proj", "model.layers.74.mlp.experts.158.up_proj", "model.layers.74.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00014325696974992752, "dbits": 2516582400 }, { "dkld": -0.00017046090215444565, "dbits": 5033164800 }, { "dkld": -0.00017914408817887306, "dbits": 7549747200 }, { "dkld": -0.00016727233305574574, "dbits": 12582912000 } ] }, { "idx": 446, "layers": [ "model.layers.74.mlp.experts.0.down_proj", "model.layers.74.mlp.experts.1.down_proj", "model.layers.74.mlp.experts.2.down_proj", "model.layers.74.mlp.experts.3.down_proj", "model.layers.74.mlp.experts.4.down_proj", "model.layers.74.mlp.experts.5.down_proj", "model.layers.74.mlp.experts.6.down_proj", "model.layers.74.mlp.experts.7.down_proj", "model.layers.74.mlp.experts.8.down_proj", "model.layers.74.mlp.experts.9.down_proj", "model.layers.74.mlp.experts.10.down_proj", "model.layers.74.mlp.experts.11.down_proj", "model.layers.74.mlp.experts.12.down_proj", "model.layers.74.mlp.experts.13.down_proj", "model.layers.74.mlp.experts.14.down_proj", "model.layers.74.mlp.experts.15.down_proj", "model.layers.74.mlp.experts.16.down_proj", "model.layers.74.mlp.experts.17.down_proj", "model.layers.74.mlp.experts.18.down_proj", "model.layers.74.mlp.experts.19.down_proj", "model.layers.74.mlp.experts.20.down_proj", "model.layers.74.mlp.experts.21.down_proj", "model.layers.74.mlp.experts.22.down_proj", "model.layers.74.mlp.experts.23.down_proj", "model.layers.74.mlp.experts.24.down_proj", "model.layers.74.mlp.experts.25.down_proj", "model.layers.74.mlp.experts.26.down_proj", "model.layers.74.mlp.experts.27.down_proj", "model.layers.74.mlp.experts.28.down_proj", "model.layers.74.mlp.experts.29.down_proj", "model.layers.74.mlp.experts.30.down_proj", "model.layers.74.mlp.experts.31.down_proj", "model.layers.74.mlp.experts.32.down_proj", "model.layers.74.mlp.experts.33.down_proj", "model.layers.74.mlp.experts.34.down_proj", "model.layers.74.mlp.experts.35.down_proj", "model.layers.74.mlp.experts.36.down_proj", "model.layers.74.mlp.experts.37.down_proj", "model.layers.74.mlp.experts.38.down_proj", "model.layers.74.mlp.experts.39.down_proj", "model.layers.74.mlp.experts.40.down_proj", "model.layers.74.mlp.experts.41.down_proj", "model.layers.74.mlp.experts.42.down_proj", "model.layers.74.mlp.experts.43.down_proj", "model.layers.74.mlp.experts.44.down_proj", "model.layers.74.mlp.experts.45.down_proj", "model.layers.74.mlp.experts.46.down_proj", "model.layers.74.mlp.experts.47.down_proj", "model.layers.74.mlp.experts.48.down_proj", "model.layers.74.mlp.experts.49.down_proj", "model.layers.74.mlp.experts.50.down_proj", "model.layers.74.mlp.experts.51.down_proj", "model.layers.74.mlp.experts.52.down_proj", "model.layers.74.mlp.experts.53.down_proj", "model.layers.74.mlp.experts.54.down_proj", "model.layers.74.mlp.experts.55.down_proj", "model.layers.74.mlp.experts.56.down_proj", "model.layers.74.mlp.experts.57.down_proj", "model.layers.74.mlp.experts.58.down_proj", "model.layers.74.mlp.experts.59.down_proj", "model.layers.74.mlp.experts.60.down_proj", "model.layers.74.mlp.experts.61.down_proj", "model.layers.74.mlp.experts.62.down_proj", "model.layers.74.mlp.experts.63.down_proj", "model.layers.74.mlp.experts.64.down_proj", "model.layers.74.mlp.experts.65.down_proj", "model.layers.74.mlp.experts.66.down_proj", "model.layers.74.mlp.experts.67.down_proj", "model.layers.74.mlp.experts.68.down_proj", "model.layers.74.mlp.experts.69.down_proj", "model.layers.74.mlp.experts.70.down_proj", "model.layers.74.mlp.experts.71.down_proj", "model.layers.74.mlp.experts.72.down_proj", "model.layers.74.mlp.experts.73.down_proj", "model.layers.74.mlp.experts.74.down_proj", "model.layers.74.mlp.experts.75.down_proj", "model.layers.74.mlp.experts.76.down_proj", "model.layers.74.mlp.experts.77.down_proj", "model.layers.74.mlp.experts.78.down_proj", "model.layers.74.mlp.experts.79.down_proj", "model.layers.74.mlp.experts.80.down_proj", "model.layers.74.mlp.experts.81.down_proj", "model.layers.74.mlp.experts.82.down_proj", "model.layers.74.mlp.experts.83.down_proj", "model.layers.74.mlp.experts.84.down_proj", "model.layers.74.mlp.experts.85.down_proj", "model.layers.74.mlp.experts.86.down_proj", "model.layers.74.mlp.experts.87.down_proj", "model.layers.74.mlp.experts.88.down_proj", "model.layers.74.mlp.experts.89.down_proj", "model.layers.74.mlp.experts.90.down_proj", "model.layers.74.mlp.experts.91.down_proj", "model.layers.74.mlp.experts.92.down_proj", "model.layers.74.mlp.experts.93.down_proj", "model.layers.74.mlp.experts.94.down_proj", "model.layers.74.mlp.experts.95.down_proj", "model.layers.74.mlp.experts.96.down_proj", "model.layers.74.mlp.experts.97.down_proj", "model.layers.74.mlp.experts.98.down_proj", "model.layers.74.mlp.experts.99.down_proj", "model.layers.74.mlp.experts.100.down_proj", "model.layers.74.mlp.experts.101.down_proj", "model.layers.74.mlp.experts.102.down_proj", "model.layers.74.mlp.experts.103.down_proj", "model.layers.74.mlp.experts.104.down_proj", "model.layers.74.mlp.experts.105.down_proj", "model.layers.74.mlp.experts.106.down_proj", "model.layers.74.mlp.experts.107.down_proj", "model.layers.74.mlp.experts.108.down_proj", "model.layers.74.mlp.experts.109.down_proj", "model.layers.74.mlp.experts.110.down_proj", "model.layers.74.mlp.experts.111.down_proj", "model.layers.74.mlp.experts.112.down_proj", "model.layers.74.mlp.experts.113.down_proj", "model.layers.74.mlp.experts.114.down_proj", "model.layers.74.mlp.experts.115.down_proj", "model.layers.74.mlp.experts.116.down_proj", "model.layers.74.mlp.experts.117.down_proj", "model.layers.74.mlp.experts.118.down_proj", "model.layers.74.mlp.experts.119.down_proj", "model.layers.74.mlp.experts.120.down_proj", "model.layers.74.mlp.experts.121.down_proj", "model.layers.74.mlp.experts.122.down_proj", "model.layers.74.mlp.experts.123.down_proj", "model.layers.74.mlp.experts.124.down_proj", "model.layers.74.mlp.experts.125.down_proj", "model.layers.74.mlp.experts.126.down_proj", "model.layers.74.mlp.experts.127.down_proj", "model.layers.74.mlp.experts.128.down_proj", "model.layers.74.mlp.experts.129.down_proj", "model.layers.74.mlp.experts.130.down_proj", "model.layers.74.mlp.experts.131.down_proj", "model.layers.74.mlp.experts.132.down_proj", "model.layers.74.mlp.experts.133.down_proj", "model.layers.74.mlp.experts.134.down_proj", "model.layers.74.mlp.experts.135.down_proj", "model.layers.74.mlp.experts.136.down_proj", "model.layers.74.mlp.experts.137.down_proj", "model.layers.74.mlp.experts.138.down_proj", "model.layers.74.mlp.experts.139.down_proj", "model.layers.74.mlp.experts.140.down_proj", "model.layers.74.mlp.experts.141.down_proj", "model.layers.74.mlp.experts.142.down_proj", "model.layers.74.mlp.experts.143.down_proj", "model.layers.74.mlp.experts.144.down_proj", "model.layers.74.mlp.experts.145.down_proj", "model.layers.74.mlp.experts.146.down_proj", "model.layers.74.mlp.experts.147.down_proj", "model.layers.74.mlp.experts.148.down_proj", "model.layers.74.mlp.experts.149.down_proj", "model.layers.74.mlp.experts.150.down_proj", "model.layers.74.mlp.experts.151.down_proj", "model.layers.74.mlp.experts.152.down_proj", "model.layers.74.mlp.experts.153.down_proj", "model.layers.74.mlp.experts.154.down_proj", "model.layers.74.mlp.experts.155.down_proj", "model.layers.74.mlp.experts.156.down_proj", "model.layers.74.mlp.experts.157.down_proj", "model.layers.74.mlp.experts.158.down_proj", "model.layers.74.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00013078786432743628, "dbits": 1258291200 }, { "dkld": -0.00015968997031451382, "dbits": 2516582400 }, { "dkld": -0.00015896931290626526, "dbits": 3774873600 }, { "dkld": -0.00016332687810063917, "dbits": 6291456000 } ] }, { "idx": 447, "layers": [ "model.layers.75.self_attn.q_proj" ], "candidates": [ { "dkld": 4.208907485007962e-05, "dbits": 62914560 }, { "dkld": -2.3760460317143184e-05, "dbits": 125829120 }, { "dkld": 1.797154545783719e-05, "dbits": 188743680 }, { "dkld": 8.50316137074314e-06, "dbits": 314572800 } ] }, { "idx": 448, "layers": [ "model.layers.75.self_attn.k_proj", "model.layers.75.self_attn.v_proj" ], "candidates": [ { "dkld": -7.385462522507824e-05, "dbits": 10485760 }, { "dkld": -4.500234499574263e-05, "dbits": 20971520 }, { "dkld": -1.1345651000749246e-05, "dbits": 31457280 }, { "dkld": 6.440095603466034e-07, "dbits": 52428800 } ] }, { "idx": 449, "layers": [ "model.layers.75.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002207107841968592, "dbits": 62914560 }, { "dkld": -0.00011501284316182969, "dbits": 125829120 }, { "dkld": -0.00015957048162817955, "dbits": 188743680 }, { "dkld": -0.00014605727046729244, "dbits": 314572800 } ] }, { "idx": 450, "layers": [ "model.layers.75.mlp.shared_experts.gate_proj", "model.layers.75.mlp.shared_experts.up_proj", "model.layers.75.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 2.475222572684288e-05, "dbits": 23592960 }, { "dkld": -2.306709066034196e-05, "dbits": 47185920 }, { "dkld": -6.757164373993874e-05, "dbits": 70778880 }, { "dkld": -8.777249604463577e-05, "dbits": 117964800 } ] }, { "idx": 451, "layers": [ "model.layers.75.mlp.experts.0.gate_proj", "model.layers.75.mlp.experts.1.gate_proj", "model.layers.75.mlp.experts.2.gate_proj", "model.layers.75.mlp.experts.3.gate_proj", "model.layers.75.mlp.experts.4.gate_proj", "model.layers.75.mlp.experts.5.gate_proj", "model.layers.75.mlp.experts.6.gate_proj", "model.layers.75.mlp.experts.7.gate_proj", "model.layers.75.mlp.experts.8.gate_proj", "model.layers.75.mlp.experts.9.gate_proj", "model.layers.75.mlp.experts.10.gate_proj", "model.layers.75.mlp.experts.11.gate_proj", "model.layers.75.mlp.experts.12.gate_proj", "model.layers.75.mlp.experts.13.gate_proj", "model.layers.75.mlp.experts.14.gate_proj", "model.layers.75.mlp.experts.15.gate_proj", "model.layers.75.mlp.experts.16.gate_proj", "model.layers.75.mlp.experts.17.gate_proj", "model.layers.75.mlp.experts.18.gate_proj", "model.layers.75.mlp.experts.19.gate_proj", "model.layers.75.mlp.experts.20.gate_proj", "model.layers.75.mlp.experts.21.gate_proj", "model.layers.75.mlp.experts.22.gate_proj", "model.layers.75.mlp.experts.23.gate_proj", "model.layers.75.mlp.experts.24.gate_proj", "model.layers.75.mlp.experts.25.gate_proj", "model.layers.75.mlp.experts.26.gate_proj", "model.layers.75.mlp.experts.27.gate_proj", "model.layers.75.mlp.experts.28.gate_proj", "model.layers.75.mlp.experts.29.gate_proj", "model.layers.75.mlp.experts.30.gate_proj", "model.layers.75.mlp.experts.31.gate_proj", "model.layers.75.mlp.experts.32.gate_proj", "model.layers.75.mlp.experts.33.gate_proj", "model.layers.75.mlp.experts.34.gate_proj", "model.layers.75.mlp.experts.35.gate_proj", "model.layers.75.mlp.experts.36.gate_proj", "model.layers.75.mlp.experts.37.gate_proj", "model.layers.75.mlp.experts.38.gate_proj", "model.layers.75.mlp.experts.39.gate_proj", "model.layers.75.mlp.experts.40.gate_proj", "model.layers.75.mlp.experts.41.gate_proj", "model.layers.75.mlp.experts.42.gate_proj", "model.layers.75.mlp.experts.43.gate_proj", "model.layers.75.mlp.experts.44.gate_proj", "model.layers.75.mlp.experts.45.gate_proj", "model.layers.75.mlp.experts.46.gate_proj", "model.layers.75.mlp.experts.47.gate_proj", "model.layers.75.mlp.experts.48.gate_proj", "model.layers.75.mlp.experts.49.gate_proj", "model.layers.75.mlp.experts.50.gate_proj", "model.layers.75.mlp.experts.51.gate_proj", "model.layers.75.mlp.experts.52.gate_proj", "model.layers.75.mlp.experts.53.gate_proj", "model.layers.75.mlp.experts.54.gate_proj", "model.layers.75.mlp.experts.55.gate_proj", "model.layers.75.mlp.experts.56.gate_proj", "model.layers.75.mlp.experts.57.gate_proj", "model.layers.75.mlp.experts.58.gate_proj", "model.layers.75.mlp.experts.59.gate_proj", "model.layers.75.mlp.experts.60.gate_proj", "model.layers.75.mlp.experts.61.gate_proj", "model.layers.75.mlp.experts.62.gate_proj", "model.layers.75.mlp.experts.63.gate_proj", "model.layers.75.mlp.experts.64.gate_proj", "model.layers.75.mlp.experts.65.gate_proj", "model.layers.75.mlp.experts.66.gate_proj", "model.layers.75.mlp.experts.67.gate_proj", "model.layers.75.mlp.experts.68.gate_proj", "model.layers.75.mlp.experts.69.gate_proj", "model.layers.75.mlp.experts.70.gate_proj", "model.layers.75.mlp.experts.71.gate_proj", "model.layers.75.mlp.experts.72.gate_proj", "model.layers.75.mlp.experts.73.gate_proj", "model.layers.75.mlp.experts.74.gate_proj", "model.layers.75.mlp.experts.75.gate_proj", "model.layers.75.mlp.experts.76.gate_proj", "model.layers.75.mlp.experts.77.gate_proj", "model.layers.75.mlp.experts.78.gate_proj", "model.layers.75.mlp.experts.79.gate_proj", "model.layers.75.mlp.experts.80.gate_proj", "model.layers.75.mlp.experts.81.gate_proj", "model.layers.75.mlp.experts.82.gate_proj", "model.layers.75.mlp.experts.83.gate_proj", "model.layers.75.mlp.experts.84.gate_proj", "model.layers.75.mlp.experts.85.gate_proj", "model.layers.75.mlp.experts.86.gate_proj", "model.layers.75.mlp.experts.87.gate_proj", "model.layers.75.mlp.experts.88.gate_proj", "model.layers.75.mlp.experts.89.gate_proj", "model.layers.75.mlp.experts.90.gate_proj", "model.layers.75.mlp.experts.91.gate_proj", "model.layers.75.mlp.experts.92.gate_proj", "model.layers.75.mlp.experts.93.gate_proj", "model.layers.75.mlp.experts.94.gate_proj", "model.layers.75.mlp.experts.95.gate_proj", "model.layers.75.mlp.experts.96.gate_proj", "model.layers.75.mlp.experts.97.gate_proj", "model.layers.75.mlp.experts.98.gate_proj", "model.layers.75.mlp.experts.99.gate_proj", "model.layers.75.mlp.experts.100.gate_proj", "model.layers.75.mlp.experts.101.gate_proj", "model.layers.75.mlp.experts.102.gate_proj", "model.layers.75.mlp.experts.103.gate_proj", "model.layers.75.mlp.experts.104.gate_proj", "model.layers.75.mlp.experts.105.gate_proj", "model.layers.75.mlp.experts.106.gate_proj", "model.layers.75.mlp.experts.107.gate_proj", "model.layers.75.mlp.experts.108.gate_proj", "model.layers.75.mlp.experts.109.gate_proj", "model.layers.75.mlp.experts.110.gate_proj", "model.layers.75.mlp.experts.111.gate_proj", "model.layers.75.mlp.experts.112.gate_proj", "model.layers.75.mlp.experts.113.gate_proj", "model.layers.75.mlp.experts.114.gate_proj", "model.layers.75.mlp.experts.115.gate_proj", "model.layers.75.mlp.experts.116.gate_proj", "model.layers.75.mlp.experts.117.gate_proj", "model.layers.75.mlp.experts.118.gate_proj", "model.layers.75.mlp.experts.119.gate_proj", "model.layers.75.mlp.experts.120.gate_proj", "model.layers.75.mlp.experts.121.gate_proj", "model.layers.75.mlp.experts.122.gate_proj", "model.layers.75.mlp.experts.123.gate_proj", "model.layers.75.mlp.experts.124.gate_proj", "model.layers.75.mlp.experts.125.gate_proj", "model.layers.75.mlp.experts.126.gate_proj", "model.layers.75.mlp.experts.127.gate_proj", "model.layers.75.mlp.experts.128.gate_proj", "model.layers.75.mlp.experts.129.gate_proj", "model.layers.75.mlp.experts.130.gate_proj", "model.layers.75.mlp.experts.131.gate_proj", "model.layers.75.mlp.experts.132.gate_proj", "model.layers.75.mlp.experts.133.gate_proj", "model.layers.75.mlp.experts.134.gate_proj", "model.layers.75.mlp.experts.135.gate_proj", "model.layers.75.mlp.experts.136.gate_proj", "model.layers.75.mlp.experts.137.gate_proj", "model.layers.75.mlp.experts.138.gate_proj", "model.layers.75.mlp.experts.139.gate_proj", "model.layers.75.mlp.experts.140.gate_proj", "model.layers.75.mlp.experts.141.gate_proj", "model.layers.75.mlp.experts.142.gate_proj", "model.layers.75.mlp.experts.143.gate_proj", "model.layers.75.mlp.experts.144.gate_proj", "model.layers.75.mlp.experts.145.gate_proj", "model.layers.75.mlp.experts.146.gate_proj", "model.layers.75.mlp.experts.147.gate_proj", "model.layers.75.mlp.experts.148.gate_proj", "model.layers.75.mlp.experts.149.gate_proj", "model.layers.75.mlp.experts.150.gate_proj", "model.layers.75.mlp.experts.151.gate_proj", "model.layers.75.mlp.experts.152.gate_proj", "model.layers.75.mlp.experts.153.gate_proj", "model.layers.75.mlp.experts.154.gate_proj", "model.layers.75.mlp.experts.155.gate_proj", "model.layers.75.mlp.experts.156.gate_proj", "model.layers.75.mlp.experts.157.gate_proj", "model.layers.75.mlp.experts.158.gate_proj", "model.layers.75.mlp.experts.159.gate_proj", "model.layers.75.mlp.experts.0.up_proj", "model.layers.75.mlp.experts.1.up_proj", "model.layers.75.mlp.experts.2.up_proj", "model.layers.75.mlp.experts.3.up_proj", "model.layers.75.mlp.experts.4.up_proj", "model.layers.75.mlp.experts.5.up_proj", "model.layers.75.mlp.experts.6.up_proj", "model.layers.75.mlp.experts.7.up_proj", "model.layers.75.mlp.experts.8.up_proj", "model.layers.75.mlp.experts.9.up_proj", "model.layers.75.mlp.experts.10.up_proj", "model.layers.75.mlp.experts.11.up_proj", "model.layers.75.mlp.experts.12.up_proj", "model.layers.75.mlp.experts.13.up_proj", "model.layers.75.mlp.experts.14.up_proj", "model.layers.75.mlp.experts.15.up_proj", "model.layers.75.mlp.experts.16.up_proj", "model.layers.75.mlp.experts.17.up_proj", "model.layers.75.mlp.experts.18.up_proj", "model.layers.75.mlp.experts.19.up_proj", "model.layers.75.mlp.experts.20.up_proj", "model.layers.75.mlp.experts.21.up_proj", "model.layers.75.mlp.experts.22.up_proj", "model.layers.75.mlp.experts.23.up_proj", "model.layers.75.mlp.experts.24.up_proj", "model.layers.75.mlp.experts.25.up_proj", "model.layers.75.mlp.experts.26.up_proj", "model.layers.75.mlp.experts.27.up_proj", "model.layers.75.mlp.experts.28.up_proj", "model.layers.75.mlp.experts.29.up_proj", "model.layers.75.mlp.experts.30.up_proj", "model.layers.75.mlp.experts.31.up_proj", "model.layers.75.mlp.experts.32.up_proj", "model.layers.75.mlp.experts.33.up_proj", "model.layers.75.mlp.experts.34.up_proj", "model.layers.75.mlp.experts.35.up_proj", "model.layers.75.mlp.experts.36.up_proj", "model.layers.75.mlp.experts.37.up_proj", "model.layers.75.mlp.experts.38.up_proj", "model.layers.75.mlp.experts.39.up_proj", "model.layers.75.mlp.experts.40.up_proj", "model.layers.75.mlp.experts.41.up_proj", "model.layers.75.mlp.experts.42.up_proj", "model.layers.75.mlp.experts.43.up_proj", "model.layers.75.mlp.experts.44.up_proj", "model.layers.75.mlp.experts.45.up_proj", "model.layers.75.mlp.experts.46.up_proj", "model.layers.75.mlp.experts.47.up_proj", "model.layers.75.mlp.experts.48.up_proj", "model.layers.75.mlp.experts.49.up_proj", "model.layers.75.mlp.experts.50.up_proj", "model.layers.75.mlp.experts.51.up_proj", "model.layers.75.mlp.experts.52.up_proj", "model.layers.75.mlp.experts.53.up_proj", "model.layers.75.mlp.experts.54.up_proj", "model.layers.75.mlp.experts.55.up_proj", "model.layers.75.mlp.experts.56.up_proj", "model.layers.75.mlp.experts.57.up_proj", "model.layers.75.mlp.experts.58.up_proj", "model.layers.75.mlp.experts.59.up_proj", "model.layers.75.mlp.experts.60.up_proj", "model.layers.75.mlp.experts.61.up_proj", "model.layers.75.mlp.experts.62.up_proj", "model.layers.75.mlp.experts.63.up_proj", "model.layers.75.mlp.experts.64.up_proj", "model.layers.75.mlp.experts.65.up_proj", "model.layers.75.mlp.experts.66.up_proj", "model.layers.75.mlp.experts.67.up_proj", "model.layers.75.mlp.experts.68.up_proj", "model.layers.75.mlp.experts.69.up_proj", "model.layers.75.mlp.experts.70.up_proj", "model.layers.75.mlp.experts.71.up_proj", "model.layers.75.mlp.experts.72.up_proj", "model.layers.75.mlp.experts.73.up_proj", "model.layers.75.mlp.experts.74.up_proj", "model.layers.75.mlp.experts.75.up_proj", "model.layers.75.mlp.experts.76.up_proj", "model.layers.75.mlp.experts.77.up_proj", "model.layers.75.mlp.experts.78.up_proj", "model.layers.75.mlp.experts.79.up_proj", "model.layers.75.mlp.experts.80.up_proj", "model.layers.75.mlp.experts.81.up_proj", "model.layers.75.mlp.experts.82.up_proj", "model.layers.75.mlp.experts.83.up_proj", "model.layers.75.mlp.experts.84.up_proj", "model.layers.75.mlp.experts.85.up_proj", "model.layers.75.mlp.experts.86.up_proj", "model.layers.75.mlp.experts.87.up_proj", "model.layers.75.mlp.experts.88.up_proj", "model.layers.75.mlp.experts.89.up_proj", "model.layers.75.mlp.experts.90.up_proj", "model.layers.75.mlp.experts.91.up_proj", "model.layers.75.mlp.experts.92.up_proj", "model.layers.75.mlp.experts.93.up_proj", "model.layers.75.mlp.experts.94.up_proj", "model.layers.75.mlp.experts.95.up_proj", "model.layers.75.mlp.experts.96.up_proj", "model.layers.75.mlp.experts.97.up_proj", "model.layers.75.mlp.experts.98.up_proj", "model.layers.75.mlp.experts.99.up_proj", "model.layers.75.mlp.experts.100.up_proj", "model.layers.75.mlp.experts.101.up_proj", "model.layers.75.mlp.experts.102.up_proj", "model.layers.75.mlp.experts.103.up_proj", "model.layers.75.mlp.experts.104.up_proj", "model.layers.75.mlp.experts.105.up_proj", "model.layers.75.mlp.experts.106.up_proj", "model.layers.75.mlp.experts.107.up_proj", "model.layers.75.mlp.experts.108.up_proj", "model.layers.75.mlp.experts.109.up_proj", "model.layers.75.mlp.experts.110.up_proj", "model.layers.75.mlp.experts.111.up_proj", "model.layers.75.mlp.experts.112.up_proj", "model.layers.75.mlp.experts.113.up_proj", "model.layers.75.mlp.experts.114.up_proj", "model.layers.75.mlp.experts.115.up_proj", "model.layers.75.mlp.experts.116.up_proj", "model.layers.75.mlp.experts.117.up_proj", "model.layers.75.mlp.experts.118.up_proj", "model.layers.75.mlp.experts.119.up_proj", "model.layers.75.mlp.experts.120.up_proj", "model.layers.75.mlp.experts.121.up_proj", "model.layers.75.mlp.experts.122.up_proj", "model.layers.75.mlp.experts.123.up_proj", "model.layers.75.mlp.experts.124.up_proj", "model.layers.75.mlp.experts.125.up_proj", "model.layers.75.mlp.experts.126.up_proj", "model.layers.75.mlp.experts.127.up_proj", "model.layers.75.mlp.experts.128.up_proj", "model.layers.75.mlp.experts.129.up_proj", "model.layers.75.mlp.experts.130.up_proj", "model.layers.75.mlp.experts.131.up_proj", "model.layers.75.mlp.experts.132.up_proj", "model.layers.75.mlp.experts.133.up_proj", "model.layers.75.mlp.experts.134.up_proj", "model.layers.75.mlp.experts.135.up_proj", "model.layers.75.mlp.experts.136.up_proj", "model.layers.75.mlp.experts.137.up_proj", "model.layers.75.mlp.experts.138.up_proj", "model.layers.75.mlp.experts.139.up_proj", "model.layers.75.mlp.experts.140.up_proj", "model.layers.75.mlp.experts.141.up_proj", "model.layers.75.mlp.experts.142.up_proj", "model.layers.75.mlp.experts.143.up_proj", "model.layers.75.mlp.experts.144.up_proj", "model.layers.75.mlp.experts.145.up_proj", "model.layers.75.mlp.experts.146.up_proj", "model.layers.75.mlp.experts.147.up_proj", "model.layers.75.mlp.experts.148.up_proj", "model.layers.75.mlp.experts.149.up_proj", "model.layers.75.mlp.experts.150.up_proj", "model.layers.75.mlp.experts.151.up_proj", "model.layers.75.mlp.experts.152.up_proj", "model.layers.75.mlp.experts.153.up_proj", "model.layers.75.mlp.experts.154.up_proj", "model.layers.75.mlp.experts.155.up_proj", "model.layers.75.mlp.experts.156.up_proj", "model.layers.75.mlp.experts.157.up_proj", "model.layers.75.mlp.experts.158.up_proj", "model.layers.75.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0002455906011164244, "dbits": 2516582400 }, { "dkld": -0.00035186987370253164, "dbits": 5033164800 }, { "dkld": -0.00037049083039164543, "dbits": 7549747200 }, { "dkld": -0.0003261410631239525, "dbits": 12582912000 } ] }, { "idx": 452, "layers": [ "model.layers.75.mlp.experts.0.down_proj", "model.layers.75.mlp.experts.1.down_proj", "model.layers.75.mlp.experts.2.down_proj", "model.layers.75.mlp.experts.3.down_proj", "model.layers.75.mlp.experts.4.down_proj", "model.layers.75.mlp.experts.5.down_proj", "model.layers.75.mlp.experts.6.down_proj", "model.layers.75.mlp.experts.7.down_proj", "model.layers.75.mlp.experts.8.down_proj", "model.layers.75.mlp.experts.9.down_proj", "model.layers.75.mlp.experts.10.down_proj", "model.layers.75.mlp.experts.11.down_proj", "model.layers.75.mlp.experts.12.down_proj", "model.layers.75.mlp.experts.13.down_proj", "model.layers.75.mlp.experts.14.down_proj", "model.layers.75.mlp.experts.15.down_proj", "model.layers.75.mlp.experts.16.down_proj", "model.layers.75.mlp.experts.17.down_proj", "model.layers.75.mlp.experts.18.down_proj", "model.layers.75.mlp.experts.19.down_proj", "model.layers.75.mlp.experts.20.down_proj", "model.layers.75.mlp.experts.21.down_proj", "model.layers.75.mlp.experts.22.down_proj", "model.layers.75.mlp.experts.23.down_proj", "model.layers.75.mlp.experts.24.down_proj", "model.layers.75.mlp.experts.25.down_proj", "model.layers.75.mlp.experts.26.down_proj", "model.layers.75.mlp.experts.27.down_proj", "model.layers.75.mlp.experts.28.down_proj", "model.layers.75.mlp.experts.29.down_proj", "model.layers.75.mlp.experts.30.down_proj", "model.layers.75.mlp.experts.31.down_proj", "model.layers.75.mlp.experts.32.down_proj", "model.layers.75.mlp.experts.33.down_proj", "model.layers.75.mlp.experts.34.down_proj", "model.layers.75.mlp.experts.35.down_proj", "model.layers.75.mlp.experts.36.down_proj", "model.layers.75.mlp.experts.37.down_proj", "model.layers.75.mlp.experts.38.down_proj", "model.layers.75.mlp.experts.39.down_proj", "model.layers.75.mlp.experts.40.down_proj", "model.layers.75.mlp.experts.41.down_proj", "model.layers.75.mlp.experts.42.down_proj", "model.layers.75.mlp.experts.43.down_proj", "model.layers.75.mlp.experts.44.down_proj", "model.layers.75.mlp.experts.45.down_proj", "model.layers.75.mlp.experts.46.down_proj", "model.layers.75.mlp.experts.47.down_proj", "model.layers.75.mlp.experts.48.down_proj", "model.layers.75.mlp.experts.49.down_proj", "model.layers.75.mlp.experts.50.down_proj", "model.layers.75.mlp.experts.51.down_proj", "model.layers.75.mlp.experts.52.down_proj", "model.layers.75.mlp.experts.53.down_proj", "model.layers.75.mlp.experts.54.down_proj", "model.layers.75.mlp.experts.55.down_proj", "model.layers.75.mlp.experts.56.down_proj", "model.layers.75.mlp.experts.57.down_proj", "model.layers.75.mlp.experts.58.down_proj", "model.layers.75.mlp.experts.59.down_proj", "model.layers.75.mlp.experts.60.down_proj", "model.layers.75.mlp.experts.61.down_proj", "model.layers.75.mlp.experts.62.down_proj", "model.layers.75.mlp.experts.63.down_proj", "model.layers.75.mlp.experts.64.down_proj", "model.layers.75.mlp.experts.65.down_proj", "model.layers.75.mlp.experts.66.down_proj", "model.layers.75.mlp.experts.67.down_proj", "model.layers.75.mlp.experts.68.down_proj", "model.layers.75.mlp.experts.69.down_proj", "model.layers.75.mlp.experts.70.down_proj", "model.layers.75.mlp.experts.71.down_proj", "model.layers.75.mlp.experts.72.down_proj", "model.layers.75.mlp.experts.73.down_proj", "model.layers.75.mlp.experts.74.down_proj", "model.layers.75.mlp.experts.75.down_proj", "model.layers.75.mlp.experts.76.down_proj", "model.layers.75.mlp.experts.77.down_proj", "model.layers.75.mlp.experts.78.down_proj", "model.layers.75.mlp.experts.79.down_proj", "model.layers.75.mlp.experts.80.down_proj", "model.layers.75.mlp.experts.81.down_proj", "model.layers.75.mlp.experts.82.down_proj", "model.layers.75.mlp.experts.83.down_proj", "model.layers.75.mlp.experts.84.down_proj", "model.layers.75.mlp.experts.85.down_proj", "model.layers.75.mlp.experts.86.down_proj", "model.layers.75.mlp.experts.87.down_proj", "model.layers.75.mlp.experts.88.down_proj", "model.layers.75.mlp.experts.89.down_proj", "model.layers.75.mlp.experts.90.down_proj", "model.layers.75.mlp.experts.91.down_proj", "model.layers.75.mlp.experts.92.down_proj", "model.layers.75.mlp.experts.93.down_proj", "model.layers.75.mlp.experts.94.down_proj", "model.layers.75.mlp.experts.95.down_proj", "model.layers.75.mlp.experts.96.down_proj", "model.layers.75.mlp.experts.97.down_proj", "model.layers.75.mlp.experts.98.down_proj", "model.layers.75.mlp.experts.99.down_proj", "model.layers.75.mlp.experts.100.down_proj", "model.layers.75.mlp.experts.101.down_proj", "model.layers.75.mlp.experts.102.down_proj", "model.layers.75.mlp.experts.103.down_proj", "model.layers.75.mlp.experts.104.down_proj", "model.layers.75.mlp.experts.105.down_proj", "model.layers.75.mlp.experts.106.down_proj", "model.layers.75.mlp.experts.107.down_proj", "model.layers.75.mlp.experts.108.down_proj", "model.layers.75.mlp.experts.109.down_proj", "model.layers.75.mlp.experts.110.down_proj", "model.layers.75.mlp.experts.111.down_proj", "model.layers.75.mlp.experts.112.down_proj", "model.layers.75.mlp.experts.113.down_proj", "model.layers.75.mlp.experts.114.down_proj", "model.layers.75.mlp.experts.115.down_proj", "model.layers.75.mlp.experts.116.down_proj", "model.layers.75.mlp.experts.117.down_proj", "model.layers.75.mlp.experts.118.down_proj", "model.layers.75.mlp.experts.119.down_proj", "model.layers.75.mlp.experts.120.down_proj", "model.layers.75.mlp.experts.121.down_proj", "model.layers.75.mlp.experts.122.down_proj", "model.layers.75.mlp.experts.123.down_proj", "model.layers.75.mlp.experts.124.down_proj", "model.layers.75.mlp.experts.125.down_proj", "model.layers.75.mlp.experts.126.down_proj", "model.layers.75.mlp.experts.127.down_proj", "model.layers.75.mlp.experts.128.down_proj", "model.layers.75.mlp.experts.129.down_proj", "model.layers.75.mlp.experts.130.down_proj", "model.layers.75.mlp.experts.131.down_proj", "model.layers.75.mlp.experts.132.down_proj", "model.layers.75.mlp.experts.133.down_proj", "model.layers.75.mlp.experts.134.down_proj", "model.layers.75.mlp.experts.135.down_proj", "model.layers.75.mlp.experts.136.down_proj", "model.layers.75.mlp.experts.137.down_proj", "model.layers.75.mlp.experts.138.down_proj", "model.layers.75.mlp.experts.139.down_proj", "model.layers.75.mlp.experts.140.down_proj", "model.layers.75.mlp.experts.141.down_proj", "model.layers.75.mlp.experts.142.down_proj", "model.layers.75.mlp.experts.143.down_proj", "model.layers.75.mlp.experts.144.down_proj", "model.layers.75.mlp.experts.145.down_proj", "model.layers.75.mlp.experts.146.down_proj", "model.layers.75.mlp.experts.147.down_proj", "model.layers.75.mlp.experts.148.down_proj", "model.layers.75.mlp.experts.149.down_proj", "model.layers.75.mlp.experts.150.down_proj", "model.layers.75.mlp.experts.151.down_proj", "model.layers.75.mlp.experts.152.down_proj", "model.layers.75.mlp.experts.153.down_proj", "model.layers.75.mlp.experts.154.down_proj", "model.layers.75.mlp.experts.155.down_proj", "model.layers.75.mlp.experts.156.down_proj", "model.layers.75.mlp.experts.157.down_proj", "model.layers.75.mlp.experts.158.down_proj", "model.layers.75.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00013963030651212294, "dbits": 1258291200 }, { "dkld": -0.00015229750424623767, "dbits": 2516582400 }, { "dkld": -0.00010167537257076142, "dbits": 3774873600 }, { "dkld": -0.00013349587097764848, "dbits": 6291456000 } ] }, { "idx": 453, "layers": [ "model.layers.76.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00013017710298299234, "dbits": 62914560 }, { "dkld": -0.00011207237839699902, "dbits": 125829120 }, { "dkld": -0.00011362805962562839, "dbits": 188743680 }, { "dkld": -0.00014710836112499792, "dbits": 314572800 } ] }, { "idx": 454, "layers": [ "model.layers.76.self_attn.k_proj", "model.layers.76.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00015917494893075146, "dbits": 10485760 }, { "dkld": -0.00028963778167963306, "dbits": 20971520 }, { "dkld": -0.0003214678727090359, "dbits": 31457280 }, { "dkld": -0.0002642361447215136, "dbits": 52428800 } ] }, { "idx": 455, "layers": [ "model.layers.76.self_attn.o_proj" ], "candidates": [ { "dkld": 9.571015834805574e-06, "dbits": 62914560 }, { "dkld": -0.0001042359508574009, "dbits": 125829120 }, { "dkld": -0.00013966560363770641, "dbits": 188743680 }, { "dkld": -0.00014089234173297882, "dbits": 314572800 } ] }, { "idx": 456, "layers": [ "model.layers.76.mlp.shared_experts.gate_proj", "model.layers.76.mlp.shared_experts.up_proj", "model.layers.76.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 1.9917357712978534e-05, "dbits": 23592960 }, { "dkld": -6.315670907497406e-05, "dbits": 47185920 }, { "dkld": -5.1221251487737485e-05, "dbits": 70778880 }, { "dkld": -9.869821369648535e-05, "dbits": 117964800 } ] }, { "idx": 457, "layers": [ "model.layers.76.mlp.experts.0.gate_proj", "model.layers.76.mlp.experts.1.gate_proj", "model.layers.76.mlp.experts.2.gate_proj", "model.layers.76.mlp.experts.3.gate_proj", "model.layers.76.mlp.experts.4.gate_proj", "model.layers.76.mlp.experts.5.gate_proj", "model.layers.76.mlp.experts.6.gate_proj", "model.layers.76.mlp.experts.7.gate_proj", "model.layers.76.mlp.experts.8.gate_proj", "model.layers.76.mlp.experts.9.gate_proj", "model.layers.76.mlp.experts.10.gate_proj", "model.layers.76.mlp.experts.11.gate_proj", "model.layers.76.mlp.experts.12.gate_proj", "model.layers.76.mlp.experts.13.gate_proj", "model.layers.76.mlp.experts.14.gate_proj", "model.layers.76.mlp.experts.15.gate_proj", "model.layers.76.mlp.experts.16.gate_proj", "model.layers.76.mlp.experts.17.gate_proj", "model.layers.76.mlp.experts.18.gate_proj", "model.layers.76.mlp.experts.19.gate_proj", "model.layers.76.mlp.experts.20.gate_proj", "model.layers.76.mlp.experts.21.gate_proj", "model.layers.76.mlp.experts.22.gate_proj", "model.layers.76.mlp.experts.23.gate_proj", "model.layers.76.mlp.experts.24.gate_proj", "model.layers.76.mlp.experts.25.gate_proj", "model.layers.76.mlp.experts.26.gate_proj", "model.layers.76.mlp.experts.27.gate_proj", "model.layers.76.mlp.experts.28.gate_proj", "model.layers.76.mlp.experts.29.gate_proj", "model.layers.76.mlp.experts.30.gate_proj", "model.layers.76.mlp.experts.31.gate_proj", "model.layers.76.mlp.experts.32.gate_proj", "model.layers.76.mlp.experts.33.gate_proj", "model.layers.76.mlp.experts.34.gate_proj", "model.layers.76.mlp.experts.35.gate_proj", "model.layers.76.mlp.experts.36.gate_proj", "model.layers.76.mlp.experts.37.gate_proj", "model.layers.76.mlp.experts.38.gate_proj", "model.layers.76.mlp.experts.39.gate_proj", "model.layers.76.mlp.experts.40.gate_proj", "model.layers.76.mlp.experts.41.gate_proj", "model.layers.76.mlp.experts.42.gate_proj", "model.layers.76.mlp.experts.43.gate_proj", "model.layers.76.mlp.experts.44.gate_proj", "model.layers.76.mlp.experts.45.gate_proj", "model.layers.76.mlp.experts.46.gate_proj", "model.layers.76.mlp.experts.47.gate_proj", "model.layers.76.mlp.experts.48.gate_proj", "model.layers.76.mlp.experts.49.gate_proj", "model.layers.76.mlp.experts.50.gate_proj", "model.layers.76.mlp.experts.51.gate_proj", "model.layers.76.mlp.experts.52.gate_proj", "model.layers.76.mlp.experts.53.gate_proj", "model.layers.76.mlp.experts.54.gate_proj", "model.layers.76.mlp.experts.55.gate_proj", "model.layers.76.mlp.experts.56.gate_proj", "model.layers.76.mlp.experts.57.gate_proj", "model.layers.76.mlp.experts.58.gate_proj", "model.layers.76.mlp.experts.59.gate_proj", "model.layers.76.mlp.experts.60.gate_proj", "model.layers.76.mlp.experts.61.gate_proj", "model.layers.76.mlp.experts.62.gate_proj", "model.layers.76.mlp.experts.63.gate_proj", "model.layers.76.mlp.experts.64.gate_proj", "model.layers.76.mlp.experts.65.gate_proj", "model.layers.76.mlp.experts.66.gate_proj", "model.layers.76.mlp.experts.67.gate_proj", "model.layers.76.mlp.experts.68.gate_proj", "model.layers.76.mlp.experts.69.gate_proj", "model.layers.76.mlp.experts.70.gate_proj", "model.layers.76.mlp.experts.71.gate_proj", "model.layers.76.mlp.experts.72.gate_proj", "model.layers.76.mlp.experts.73.gate_proj", "model.layers.76.mlp.experts.74.gate_proj", "model.layers.76.mlp.experts.75.gate_proj", "model.layers.76.mlp.experts.76.gate_proj", "model.layers.76.mlp.experts.77.gate_proj", "model.layers.76.mlp.experts.78.gate_proj", "model.layers.76.mlp.experts.79.gate_proj", "model.layers.76.mlp.experts.80.gate_proj", "model.layers.76.mlp.experts.81.gate_proj", "model.layers.76.mlp.experts.82.gate_proj", "model.layers.76.mlp.experts.83.gate_proj", "model.layers.76.mlp.experts.84.gate_proj", "model.layers.76.mlp.experts.85.gate_proj", "model.layers.76.mlp.experts.86.gate_proj", "model.layers.76.mlp.experts.87.gate_proj", "model.layers.76.mlp.experts.88.gate_proj", "model.layers.76.mlp.experts.89.gate_proj", "model.layers.76.mlp.experts.90.gate_proj", "model.layers.76.mlp.experts.91.gate_proj", "model.layers.76.mlp.experts.92.gate_proj", "model.layers.76.mlp.experts.93.gate_proj", "model.layers.76.mlp.experts.94.gate_proj", "model.layers.76.mlp.experts.95.gate_proj", "model.layers.76.mlp.experts.96.gate_proj", "model.layers.76.mlp.experts.97.gate_proj", "model.layers.76.mlp.experts.98.gate_proj", "model.layers.76.mlp.experts.99.gate_proj", "model.layers.76.mlp.experts.100.gate_proj", "model.layers.76.mlp.experts.101.gate_proj", "model.layers.76.mlp.experts.102.gate_proj", "model.layers.76.mlp.experts.103.gate_proj", "model.layers.76.mlp.experts.104.gate_proj", "model.layers.76.mlp.experts.105.gate_proj", "model.layers.76.mlp.experts.106.gate_proj", "model.layers.76.mlp.experts.107.gate_proj", "model.layers.76.mlp.experts.108.gate_proj", "model.layers.76.mlp.experts.109.gate_proj", "model.layers.76.mlp.experts.110.gate_proj", "model.layers.76.mlp.experts.111.gate_proj", "model.layers.76.mlp.experts.112.gate_proj", "model.layers.76.mlp.experts.113.gate_proj", "model.layers.76.mlp.experts.114.gate_proj", "model.layers.76.mlp.experts.115.gate_proj", "model.layers.76.mlp.experts.116.gate_proj", "model.layers.76.mlp.experts.117.gate_proj", "model.layers.76.mlp.experts.118.gate_proj", "model.layers.76.mlp.experts.119.gate_proj", "model.layers.76.mlp.experts.120.gate_proj", "model.layers.76.mlp.experts.121.gate_proj", "model.layers.76.mlp.experts.122.gate_proj", "model.layers.76.mlp.experts.123.gate_proj", "model.layers.76.mlp.experts.124.gate_proj", "model.layers.76.mlp.experts.125.gate_proj", "model.layers.76.mlp.experts.126.gate_proj", "model.layers.76.mlp.experts.127.gate_proj", "model.layers.76.mlp.experts.128.gate_proj", "model.layers.76.mlp.experts.129.gate_proj", "model.layers.76.mlp.experts.130.gate_proj", "model.layers.76.mlp.experts.131.gate_proj", "model.layers.76.mlp.experts.132.gate_proj", "model.layers.76.mlp.experts.133.gate_proj", "model.layers.76.mlp.experts.134.gate_proj", "model.layers.76.mlp.experts.135.gate_proj", "model.layers.76.mlp.experts.136.gate_proj", "model.layers.76.mlp.experts.137.gate_proj", "model.layers.76.mlp.experts.138.gate_proj", "model.layers.76.mlp.experts.139.gate_proj", "model.layers.76.mlp.experts.140.gate_proj", "model.layers.76.mlp.experts.141.gate_proj", "model.layers.76.mlp.experts.142.gate_proj", "model.layers.76.mlp.experts.143.gate_proj", "model.layers.76.mlp.experts.144.gate_proj", "model.layers.76.mlp.experts.145.gate_proj", "model.layers.76.mlp.experts.146.gate_proj", "model.layers.76.mlp.experts.147.gate_proj", "model.layers.76.mlp.experts.148.gate_proj", "model.layers.76.mlp.experts.149.gate_proj", "model.layers.76.mlp.experts.150.gate_proj", "model.layers.76.mlp.experts.151.gate_proj", "model.layers.76.mlp.experts.152.gate_proj", "model.layers.76.mlp.experts.153.gate_proj", "model.layers.76.mlp.experts.154.gate_proj", "model.layers.76.mlp.experts.155.gate_proj", "model.layers.76.mlp.experts.156.gate_proj", "model.layers.76.mlp.experts.157.gate_proj", "model.layers.76.mlp.experts.158.gate_proj", "model.layers.76.mlp.experts.159.gate_proj", "model.layers.76.mlp.experts.0.up_proj", "model.layers.76.mlp.experts.1.up_proj", "model.layers.76.mlp.experts.2.up_proj", "model.layers.76.mlp.experts.3.up_proj", "model.layers.76.mlp.experts.4.up_proj", "model.layers.76.mlp.experts.5.up_proj", "model.layers.76.mlp.experts.6.up_proj", "model.layers.76.mlp.experts.7.up_proj", "model.layers.76.mlp.experts.8.up_proj", "model.layers.76.mlp.experts.9.up_proj", "model.layers.76.mlp.experts.10.up_proj", "model.layers.76.mlp.experts.11.up_proj", "model.layers.76.mlp.experts.12.up_proj", "model.layers.76.mlp.experts.13.up_proj", "model.layers.76.mlp.experts.14.up_proj", "model.layers.76.mlp.experts.15.up_proj", "model.layers.76.mlp.experts.16.up_proj", "model.layers.76.mlp.experts.17.up_proj", "model.layers.76.mlp.experts.18.up_proj", "model.layers.76.mlp.experts.19.up_proj", "model.layers.76.mlp.experts.20.up_proj", "model.layers.76.mlp.experts.21.up_proj", "model.layers.76.mlp.experts.22.up_proj", "model.layers.76.mlp.experts.23.up_proj", "model.layers.76.mlp.experts.24.up_proj", "model.layers.76.mlp.experts.25.up_proj", "model.layers.76.mlp.experts.26.up_proj", "model.layers.76.mlp.experts.27.up_proj", "model.layers.76.mlp.experts.28.up_proj", "model.layers.76.mlp.experts.29.up_proj", "model.layers.76.mlp.experts.30.up_proj", "model.layers.76.mlp.experts.31.up_proj", "model.layers.76.mlp.experts.32.up_proj", "model.layers.76.mlp.experts.33.up_proj", "model.layers.76.mlp.experts.34.up_proj", "model.layers.76.mlp.experts.35.up_proj", "model.layers.76.mlp.experts.36.up_proj", "model.layers.76.mlp.experts.37.up_proj", "model.layers.76.mlp.experts.38.up_proj", "model.layers.76.mlp.experts.39.up_proj", "model.layers.76.mlp.experts.40.up_proj", "model.layers.76.mlp.experts.41.up_proj", "model.layers.76.mlp.experts.42.up_proj", "model.layers.76.mlp.experts.43.up_proj", "model.layers.76.mlp.experts.44.up_proj", "model.layers.76.mlp.experts.45.up_proj", "model.layers.76.mlp.experts.46.up_proj", "model.layers.76.mlp.experts.47.up_proj", "model.layers.76.mlp.experts.48.up_proj", "model.layers.76.mlp.experts.49.up_proj", "model.layers.76.mlp.experts.50.up_proj", "model.layers.76.mlp.experts.51.up_proj", "model.layers.76.mlp.experts.52.up_proj", "model.layers.76.mlp.experts.53.up_proj", "model.layers.76.mlp.experts.54.up_proj", "model.layers.76.mlp.experts.55.up_proj", "model.layers.76.mlp.experts.56.up_proj", "model.layers.76.mlp.experts.57.up_proj", "model.layers.76.mlp.experts.58.up_proj", "model.layers.76.mlp.experts.59.up_proj", "model.layers.76.mlp.experts.60.up_proj", "model.layers.76.mlp.experts.61.up_proj", "model.layers.76.mlp.experts.62.up_proj", "model.layers.76.mlp.experts.63.up_proj", "model.layers.76.mlp.experts.64.up_proj", "model.layers.76.mlp.experts.65.up_proj", "model.layers.76.mlp.experts.66.up_proj", "model.layers.76.mlp.experts.67.up_proj", "model.layers.76.mlp.experts.68.up_proj", "model.layers.76.mlp.experts.69.up_proj", "model.layers.76.mlp.experts.70.up_proj", "model.layers.76.mlp.experts.71.up_proj", "model.layers.76.mlp.experts.72.up_proj", "model.layers.76.mlp.experts.73.up_proj", "model.layers.76.mlp.experts.74.up_proj", "model.layers.76.mlp.experts.75.up_proj", "model.layers.76.mlp.experts.76.up_proj", "model.layers.76.mlp.experts.77.up_proj", "model.layers.76.mlp.experts.78.up_proj", "model.layers.76.mlp.experts.79.up_proj", "model.layers.76.mlp.experts.80.up_proj", "model.layers.76.mlp.experts.81.up_proj", "model.layers.76.mlp.experts.82.up_proj", "model.layers.76.mlp.experts.83.up_proj", "model.layers.76.mlp.experts.84.up_proj", "model.layers.76.mlp.experts.85.up_proj", "model.layers.76.mlp.experts.86.up_proj", "model.layers.76.mlp.experts.87.up_proj", "model.layers.76.mlp.experts.88.up_proj", "model.layers.76.mlp.experts.89.up_proj", "model.layers.76.mlp.experts.90.up_proj", "model.layers.76.mlp.experts.91.up_proj", "model.layers.76.mlp.experts.92.up_proj", "model.layers.76.mlp.experts.93.up_proj", "model.layers.76.mlp.experts.94.up_proj", "model.layers.76.mlp.experts.95.up_proj", "model.layers.76.mlp.experts.96.up_proj", "model.layers.76.mlp.experts.97.up_proj", "model.layers.76.mlp.experts.98.up_proj", "model.layers.76.mlp.experts.99.up_proj", "model.layers.76.mlp.experts.100.up_proj", "model.layers.76.mlp.experts.101.up_proj", "model.layers.76.mlp.experts.102.up_proj", "model.layers.76.mlp.experts.103.up_proj", "model.layers.76.mlp.experts.104.up_proj", "model.layers.76.mlp.experts.105.up_proj", "model.layers.76.mlp.experts.106.up_proj", "model.layers.76.mlp.experts.107.up_proj", "model.layers.76.mlp.experts.108.up_proj", "model.layers.76.mlp.experts.109.up_proj", "model.layers.76.mlp.experts.110.up_proj", "model.layers.76.mlp.experts.111.up_proj", "model.layers.76.mlp.experts.112.up_proj", "model.layers.76.mlp.experts.113.up_proj", "model.layers.76.mlp.experts.114.up_proj", "model.layers.76.mlp.experts.115.up_proj", "model.layers.76.mlp.experts.116.up_proj", "model.layers.76.mlp.experts.117.up_proj", "model.layers.76.mlp.experts.118.up_proj", "model.layers.76.mlp.experts.119.up_proj", "model.layers.76.mlp.experts.120.up_proj", "model.layers.76.mlp.experts.121.up_proj", "model.layers.76.mlp.experts.122.up_proj", "model.layers.76.mlp.experts.123.up_proj", "model.layers.76.mlp.experts.124.up_proj", "model.layers.76.mlp.experts.125.up_proj", "model.layers.76.mlp.experts.126.up_proj", "model.layers.76.mlp.experts.127.up_proj", "model.layers.76.mlp.experts.128.up_proj", "model.layers.76.mlp.experts.129.up_proj", "model.layers.76.mlp.experts.130.up_proj", "model.layers.76.mlp.experts.131.up_proj", "model.layers.76.mlp.experts.132.up_proj", "model.layers.76.mlp.experts.133.up_proj", "model.layers.76.mlp.experts.134.up_proj", "model.layers.76.mlp.experts.135.up_proj", "model.layers.76.mlp.experts.136.up_proj", "model.layers.76.mlp.experts.137.up_proj", "model.layers.76.mlp.experts.138.up_proj", "model.layers.76.mlp.experts.139.up_proj", "model.layers.76.mlp.experts.140.up_proj", "model.layers.76.mlp.experts.141.up_proj", "model.layers.76.mlp.experts.142.up_proj", "model.layers.76.mlp.experts.143.up_proj", "model.layers.76.mlp.experts.144.up_proj", "model.layers.76.mlp.experts.145.up_proj", "model.layers.76.mlp.experts.146.up_proj", "model.layers.76.mlp.experts.147.up_proj", "model.layers.76.mlp.experts.148.up_proj", "model.layers.76.mlp.experts.149.up_proj", "model.layers.76.mlp.experts.150.up_proj", "model.layers.76.mlp.experts.151.up_proj", "model.layers.76.mlp.experts.152.up_proj", "model.layers.76.mlp.experts.153.up_proj", "model.layers.76.mlp.experts.154.up_proj", "model.layers.76.mlp.experts.155.up_proj", "model.layers.76.mlp.experts.156.up_proj", "model.layers.76.mlp.experts.157.up_proj", "model.layers.76.mlp.experts.158.up_proj", "model.layers.76.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 0.000231365393847216, "dbits": 2516582400 }, { "dkld": 0.00023508118465542793, "dbits": 5033164800 }, { "dkld": 0.00026599746197461527, "dbits": 7549747200 }, { "dkld": 0.0002207936719059861, "dbits": 12582912000 } ] }, { "idx": 458, "layers": [ "model.layers.76.mlp.experts.0.down_proj", "model.layers.76.mlp.experts.1.down_proj", "model.layers.76.mlp.experts.2.down_proj", "model.layers.76.mlp.experts.3.down_proj", "model.layers.76.mlp.experts.4.down_proj", "model.layers.76.mlp.experts.5.down_proj", "model.layers.76.mlp.experts.6.down_proj", "model.layers.76.mlp.experts.7.down_proj", "model.layers.76.mlp.experts.8.down_proj", "model.layers.76.mlp.experts.9.down_proj", "model.layers.76.mlp.experts.10.down_proj", "model.layers.76.mlp.experts.11.down_proj", "model.layers.76.mlp.experts.12.down_proj", "model.layers.76.mlp.experts.13.down_proj", "model.layers.76.mlp.experts.14.down_proj", "model.layers.76.mlp.experts.15.down_proj", "model.layers.76.mlp.experts.16.down_proj", "model.layers.76.mlp.experts.17.down_proj", "model.layers.76.mlp.experts.18.down_proj", "model.layers.76.mlp.experts.19.down_proj", "model.layers.76.mlp.experts.20.down_proj", "model.layers.76.mlp.experts.21.down_proj", "model.layers.76.mlp.experts.22.down_proj", "model.layers.76.mlp.experts.23.down_proj", "model.layers.76.mlp.experts.24.down_proj", "model.layers.76.mlp.experts.25.down_proj", "model.layers.76.mlp.experts.26.down_proj", "model.layers.76.mlp.experts.27.down_proj", "model.layers.76.mlp.experts.28.down_proj", "model.layers.76.mlp.experts.29.down_proj", "model.layers.76.mlp.experts.30.down_proj", "model.layers.76.mlp.experts.31.down_proj", "model.layers.76.mlp.experts.32.down_proj", "model.layers.76.mlp.experts.33.down_proj", "model.layers.76.mlp.experts.34.down_proj", "model.layers.76.mlp.experts.35.down_proj", "model.layers.76.mlp.experts.36.down_proj", "model.layers.76.mlp.experts.37.down_proj", "model.layers.76.mlp.experts.38.down_proj", "model.layers.76.mlp.experts.39.down_proj", "model.layers.76.mlp.experts.40.down_proj", "model.layers.76.mlp.experts.41.down_proj", "model.layers.76.mlp.experts.42.down_proj", "model.layers.76.mlp.experts.43.down_proj", "model.layers.76.mlp.experts.44.down_proj", "model.layers.76.mlp.experts.45.down_proj", "model.layers.76.mlp.experts.46.down_proj", "model.layers.76.mlp.experts.47.down_proj", "model.layers.76.mlp.experts.48.down_proj", "model.layers.76.mlp.experts.49.down_proj", "model.layers.76.mlp.experts.50.down_proj", "model.layers.76.mlp.experts.51.down_proj", "model.layers.76.mlp.experts.52.down_proj", "model.layers.76.mlp.experts.53.down_proj", "model.layers.76.mlp.experts.54.down_proj", "model.layers.76.mlp.experts.55.down_proj", "model.layers.76.mlp.experts.56.down_proj", "model.layers.76.mlp.experts.57.down_proj", "model.layers.76.mlp.experts.58.down_proj", "model.layers.76.mlp.experts.59.down_proj", "model.layers.76.mlp.experts.60.down_proj", "model.layers.76.mlp.experts.61.down_proj", "model.layers.76.mlp.experts.62.down_proj", "model.layers.76.mlp.experts.63.down_proj", "model.layers.76.mlp.experts.64.down_proj", "model.layers.76.mlp.experts.65.down_proj", "model.layers.76.mlp.experts.66.down_proj", "model.layers.76.mlp.experts.67.down_proj", "model.layers.76.mlp.experts.68.down_proj", "model.layers.76.mlp.experts.69.down_proj", "model.layers.76.mlp.experts.70.down_proj", "model.layers.76.mlp.experts.71.down_proj", "model.layers.76.mlp.experts.72.down_proj", "model.layers.76.mlp.experts.73.down_proj", "model.layers.76.mlp.experts.74.down_proj", "model.layers.76.mlp.experts.75.down_proj", "model.layers.76.mlp.experts.76.down_proj", "model.layers.76.mlp.experts.77.down_proj", "model.layers.76.mlp.experts.78.down_proj", "model.layers.76.mlp.experts.79.down_proj", "model.layers.76.mlp.experts.80.down_proj", "model.layers.76.mlp.experts.81.down_proj", "model.layers.76.mlp.experts.82.down_proj", "model.layers.76.mlp.experts.83.down_proj", "model.layers.76.mlp.experts.84.down_proj", "model.layers.76.mlp.experts.85.down_proj", "model.layers.76.mlp.experts.86.down_proj", "model.layers.76.mlp.experts.87.down_proj", "model.layers.76.mlp.experts.88.down_proj", "model.layers.76.mlp.experts.89.down_proj", "model.layers.76.mlp.experts.90.down_proj", "model.layers.76.mlp.experts.91.down_proj", "model.layers.76.mlp.experts.92.down_proj", "model.layers.76.mlp.experts.93.down_proj", "model.layers.76.mlp.experts.94.down_proj", "model.layers.76.mlp.experts.95.down_proj", "model.layers.76.mlp.experts.96.down_proj", "model.layers.76.mlp.experts.97.down_proj", "model.layers.76.mlp.experts.98.down_proj", "model.layers.76.mlp.experts.99.down_proj", "model.layers.76.mlp.experts.100.down_proj", "model.layers.76.mlp.experts.101.down_proj", "model.layers.76.mlp.experts.102.down_proj", "model.layers.76.mlp.experts.103.down_proj", "model.layers.76.mlp.experts.104.down_proj", "model.layers.76.mlp.experts.105.down_proj", "model.layers.76.mlp.experts.106.down_proj", "model.layers.76.mlp.experts.107.down_proj", "model.layers.76.mlp.experts.108.down_proj", "model.layers.76.mlp.experts.109.down_proj", "model.layers.76.mlp.experts.110.down_proj", "model.layers.76.mlp.experts.111.down_proj", "model.layers.76.mlp.experts.112.down_proj", "model.layers.76.mlp.experts.113.down_proj", "model.layers.76.mlp.experts.114.down_proj", "model.layers.76.mlp.experts.115.down_proj", "model.layers.76.mlp.experts.116.down_proj", "model.layers.76.mlp.experts.117.down_proj", "model.layers.76.mlp.experts.118.down_proj", "model.layers.76.mlp.experts.119.down_proj", "model.layers.76.mlp.experts.120.down_proj", "model.layers.76.mlp.experts.121.down_proj", "model.layers.76.mlp.experts.122.down_proj", "model.layers.76.mlp.experts.123.down_proj", "model.layers.76.mlp.experts.124.down_proj", "model.layers.76.mlp.experts.125.down_proj", "model.layers.76.mlp.experts.126.down_proj", "model.layers.76.mlp.experts.127.down_proj", "model.layers.76.mlp.experts.128.down_proj", "model.layers.76.mlp.experts.129.down_proj", "model.layers.76.mlp.experts.130.down_proj", "model.layers.76.mlp.experts.131.down_proj", "model.layers.76.mlp.experts.132.down_proj", "model.layers.76.mlp.experts.133.down_proj", "model.layers.76.mlp.experts.134.down_proj", "model.layers.76.mlp.experts.135.down_proj", "model.layers.76.mlp.experts.136.down_proj", "model.layers.76.mlp.experts.137.down_proj", "model.layers.76.mlp.experts.138.down_proj", "model.layers.76.mlp.experts.139.down_proj", "model.layers.76.mlp.experts.140.down_proj", "model.layers.76.mlp.experts.141.down_proj", "model.layers.76.mlp.experts.142.down_proj", "model.layers.76.mlp.experts.143.down_proj", "model.layers.76.mlp.experts.144.down_proj", "model.layers.76.mlp.experts.145.down_proj", "model.layers.76.mlp.experts.146.down_proj", "model.layers.76.mlp.experts.147.down_proj", "model.layers.76.mlp.experts.148.down_proj", "model.layers.76.mlp.experts.149.down_proj", "model.layers.76.mlp.experts.150.down_proj", "model.layers.76.mlp.experts.151.down_proj", "model.layers.76.mlp.experts.152.down_proj", "model.layers.76.mlp.experts.153.down_proj", "model.layers.76.mlp.experts.154.down_proj", "model.layers.76.mlp.experts.155.down_proj", "model.layers.76.mlp.experts.156.down_proj", "model.layers.76.mlp.experts.157.down_proj", "model.layers.76.mlp.experts.158.down_proj", "model.layers.76.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -6.481306627392769e-05, "dbits": 1258291200 }, { "dkld": -0.00010262085124851383, "dbits": 2516582400 }, { "dkld": -0.00010199984535574913, "dbits": 3774873600 }, { "dkld": -0.00012462381273509182, "dbits": 6291456000 } ] }, { "idx": 459, "layers": [ "model.layers.77.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00028107045218349336, "dbits": 62914560 }, { "dkld": -0.0003672899678349578, "dbits": 125829120 }, { "dkld": -0.0003994859755039326, "dbits": 188743680 }, { "dkld": -0.00040099984034897007, "dbits": 314572800 } ] }, { "idx": 460, "layers": [ "model.layers.77.self_attn.k_proj", "model.layers.77.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00023970864713192264, "dbits": 10485760 }, { "dkld": -0.0005644025281071746, "dbits": 20971520 }, { "dkld": -0.0006217842921614758, "dbits": 31457280 }, { "dkld": -0.0005734166130423601, "dbits": 52428800 } ] }, { "idx": 461, "layers": [ "model.layers.77.self_attn.o_proj" ], "candidates": [ { "dkld": -6.303563714027405e-05, "dbits": 62914560 }, { "dkld": -0.0001337008550763158, "dbits": 125829120 }, { "dkld": -0.00014754245057702065, "dbits": 188743680 }, { "dkld": -8.8508240878582e-05, "dbits": 314572800 } ] }, { "idx": 462, "layers": [ "model.layers.77.mlp.shared_experts.gate_proj", "model.layers.77.mlp.shared_experts.up_proj", "model.layers.77.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00020230896770954965, "dbits": 23592960 }, { "dkld": -0.0002434002235531807, "dbits": 47185920 }, { "dkld": -0.0002091703936457634, "dbits": 70778880 }, { "dkld": -0.00018019210547209896, "dbits": 117964800 } ] }, { "idx": 463, "layers": [ "model.layers.77.mlp.experts.0.gate_proj", "model.layers.77.mlp.experts.1.gate_proj", "model.layers.77.mlp.experts.2.gate_proj", "model.layers.77.mlp.experts.3.gate_proj", "model.layers.77.mlp.experts.4.gate_proj", "model.layers.77.mlp.experts.5.gate_proj", "model.layers.77.mlp.experts.6.gate_proj", "model.layers.77.mlp.experts.7.gate_proj", "model.layers.77.mlp.experts.8.gate_proj", "model.layers.77.mlp.experts.9.gate_proj", "model.layers.77.mlp.experts.10.gate_proj", "model.layers.77.mlp.experts.11.gate_proj", "model.layers.77.mlp.experts.12.gate_proj", "model.layers.77.mlp.experts.13.gate_proj", "model.layers.77.mlp.experts.14.gate_proj", "model.layers.77.mlp.experts.15.gate_proj", "model.layers.77.mlp.experts.16.gate_proj", "model.layers.77.mlp.experts.17.gate_proj", "model.layers.77.mlp.experts.18.gate_proj", "model.layers.77.mlp.experts.19.gate_proj", "model.layers.77.mlp.experts.20.gate_proj", "model.layers.77.mlp.experts.21.gate_proj", "model.layers.77.mlp.experts.22.gate_proj", "model.layers.77.mlp.experts.23.gate_proj", "model.layers.77.mlp.experts.24.gate_proj", "model.layers.77.mlp.experts.25.gate_proj", "model.layers.77.mlp.experts.26.gate_proj", "model.layers.77.mlp.experts.27.gate_proj", "model.layers.77.mlp.experts.28.gate_proj", "model.layers.77.mlp.experts.29.gate_proj", "model.layers.77.mlp.experts.30.gate_proj", "model.layers.77.mlp.experts.31.gate_proj", "model.layers.77.mlp.experts.32.gate_proj", "model.layers.77.mlp.experts.33.gate_proj", "model.layers.77.mlp.experts.34.gate_proj", "model.layers.77.mlp.experts.35.gate_proj", "model.layers.77.mlp.experts.36.gate_proj", "model.layers.77.mlp.experts.37.gate_proj", "model.layers.77.mlp.experts.38.gate_proj", "model.layers.77.mlp.experts.39.gate_proj", "model.layers.77.mlp.experts.40.gate_proj", "model.layers.77.mlp.experts.41.gate_proj", "model.layers.77.mlp.experts.42.gate_proj", "model.layers.77.mlp.experts.43.gate_proj", "model.layers.77.mlp.experts.44.gate_proj", "model.layers.77.mlp.experts.45.gate_proj", "model.layers.77.mlp.experts.46.gate_proj", "model.layers.77.mlp.experts.47.gate_proj", "model.layers.77.mlp.experts.48.gate_proj", "model.layers.77.mlp.experts.49.gate_proj", "model.layers.77.mlp.experts.50.gate_proj", "model.layers.77.mlp.experts.51.gate_proj", "model.layers.77.mlp.experts.52.gate_proj", "model.layers.77.mlp.experts.53.gate_proj", "model.layers.77.mlp.experts.54.gate_proj", "model.layers.77.mlp.experts.55.gate_proj", "model.layers.77.mlp.experts.56.gate_proj", "model.layers.77.mlp.experts.57.gate_proj", "model.layers.77.mlp.experts.58.gate_proj", "model.layers.77.mlp.experts.59.gate_proj", "model.layers.77.mlp.experts.60.gate_proj", "model.layers.77.mlp.experts.61.gate_proj", "model.layers.77.mlp.experts.62.gate_proj", "model.layers.77.mlp.experts.63.gate_proj", "model.layers.77.mlp.experts.64.gate_proj", "model.layers.77.mlp.experts.65.gate_proj", "model.layers.77.mlp.experts.66.gate_proj", "model.layers.77.mlp.experts.67.gate_proj", "model.layers.77.mlp.experts.68.gate_proj", "model.layers.77.mlp.experts.69.gate_proj", "model.layers.77.mlp.experts.70.gate_proj", "model.layers.77.mlp.experts.71.gate_proj", "model.layers.77.mlp.experts.72.gate_proj", "model.layers.77.mlp.experts.73.gate_proj", "model.layers.77.mlp.experts.74.gate_proj", "model.layers.77.mlp.experts.75.gate_proj", "model.layers.77.mlp.experts.76.gate_proj", "model.layers.77.mlp.experts.77.gate_proj", "model.layers.77.mlp.experts.78.gate_proj", "model.layers.77.mlp.experts.79.gate_proj", "model.layers.77.mlp.experts.80.gate_proj", "model.layers.77.mlp.experts.81.gate_proj", "model.layers.77.mlp.experts.82.gate_proj", "model.layers.77.mlp.experts.83.gate_proj", "model.layers.77.mlp.experts.84.gate_proj", "model.layers.77.mlp.experts.85.gate_proj", "model.layers.77.mlp.experts.86.gate_proj", "model.layers.77.mlp.experts.87.gate_proj", "model.layers.77.mlp.experts.88.gate_proj", "model.layers.77.mlp.experts.89.gate_proj", "model.layers.77.mlp.experts.90.gate_proj", "model.layers.77.mlp.experts.91.gate_proj", "model.layers.77.mlp.experts.92.gate_proj", "model.layers.77.mlp.experts.93.gate_proj", "model.layers.77.mlp.experts.94.gate_proj", "model.layers.77.mlp.experts.95.gate_proj", "model.layers.77.mlp.experts.96.gate_proj", "model.layers.77.mlp.experts.97.gate_proj", "model.layers.77.mlp.experts.98.gate_proj", "model.layers.77.mlp.experts.99.gate_proj", "model.layers.77.mlp.experts.100.gate_proj", "model.layers.77.mlp.experts.101.gate_proj", "model.layers.77.mlp.experts.102.gate_proj", "model.layers.77.mlp.experts.103.gate_proj", "model.layers.77.mlp.experts.104.gate_proj", "model.layers.77.mlp.experts.105.gate_proj", "model.layers.77.mlp.experts.106.gate_proj", "model.layers.77.mlp.experts.107.gate_proj", "model.layers.77.mlp.experts.108.gate_proj", "model.layers.77.mlp.experts.109.gate_proj", "model.layers.77.mlp.experts.110.gate_proj", "model.layers.77.mlp.experts.111.gate_proj", "model.layers.77.mlp.experts.112.gate_proj", "model.layers.77.mlp.experts.113.gate_proj", "model.layers.77.mlp.experts.114.gate_proj", "model.layers.77.mlp.experts.115.gate_proj", "model.layers.77.mlp.experts.116.gate_proj", "model.layers.77.mlp.experts.117.gate_proj", "model.layers.77.mlp.experts.118.gate_proj", "model.layers.77.mlp.experts.119.gate_proj", "model.layers.77.mlp.experts.120.gate_proj", "model.layers.77.mlp.experts.121.gate_proj", "model.layers.77.mlp.experts.122.gate_proj", "model.layers.77.mlp.experts.123.gate_proj", "model.layers.77.mlp.experts.124.gate_proj", "model.layers.77.mlp.experts.125.gate_proj", "model.layers.77.mlp.experts.126.gate_proj", "model.layers.77.mlp.experts.127.gate_proj", "model.layers.77.mlp.experts.128.gate_proj", "model.layers.77.mlp.experts.129.gate_proj", "model.layers.77.mlp.experts.130.gate_proj", "model.layers.77.mlp.experts.131.gate_proj", "model.layers.77.mlp.experts.132.gate_proj", "model.layers.77.mlp.experts.133.gate_proj", "model.layers.77.mlp.experts.134.gate_proj", "model.layers.77.mlp.experts.135.gate_proj", "model.layers.77.mlp.experts.136.gate_proj", "model.layers.77.mlp.experts.137.gate_proj", "model.layers.77.mlp.experts.138.gate_proj", "model.layers.77.mlp.experts.139.gate_proj", "model.layers.77.mlp.experts.140.gate_proj", "model.layers.77.mlp.experts.141.gate_proj", "model.layers.77.mlp.experts.142.gate_proj", "model.layers.77.mlp.experts.143.gate_proj", "model.layers.77.mlp.experts.144.gate_proj", "model.layers.77.mlp.experts.145.gate_proj", "model.layers.77.mlp.experts.146.gate_proj", "model.layers.77.mlp.experts.147.gate_proj", "model.layers.77.mlp.experts.148.gate_proj", "model.layers.77.mlp.experts.149.gate_proj", "model.layers.77.mlp.experts.150.gate_proj", "model.layers.77.mlp.experts.151.gate_proj", "model.layers.77.mlp.experts.152.gate_proj", "model.layers.77.mlp.experts.153.gate_proj", "model.layers.77.mlp.experts.154.gate_proj", "model.layers.77.mlp.experts.155.gate_proj", "model.layers.77.mlp.experts.156.gate_proj", "model.layers.77.mlp.experts.157.gate_proj", "model.layers.77.mlp.experts.158.gate_proj", "model.layers.77.mlp.experts.159.gate_proj", "model.layers.77.mlp.experts.0.up_proj", "model.layers.77.mlp.experts.1.up_proj", "model.layers.77.mlp.experts.2.up_proj", "model.layers.77.mlp.experts.3.up_proj", "model.layers.77.mlp.experts.4.up_proj", "model.layers.77.mlp.experts.5.up_proj", "model.layers.77.mlp.experts.6.up_proj", "model.layers.77.mlp.experts.7.up_proj", "model.layers.77.mlp.experts.8.up_proj", "model.layers.77.mlp.experts.9.up_proj", "model.layers.77.mlp.experts.10.up_proj", "model.layers.77.mlp.experts.11.up_proj", "model.layers.77.mlp.experts.12.up_proj", "model.layers.77.mlp.experts.13.up_proj", "model.layers.77.mlp.experts.14.up_proj", "model.layers.77.mlp.experts.15.up_proj", "model.layers.77.mlp.experts.16.up_proj", "model.layers.77.mlp.experts.17.up_proj", "model.layers.77.mlp.experts.18.up_proj", "model.layers.77.mlp.experts.19.up_proj", "model.layers.77.mlp.experts.20.up_proj", "model.layers.77.mlp.experts.21.up_proj", "model.layers.77.mlp.experts.22.up_proj", "model.layers.77.mlp.experts.23.up_proj", "model.layers.77.mlp.experts.24.up_proj", "model.layers.77.mlp.experts.25.up_proj", "model.layers.77.mlp.experts.26.up_proj", "model.layers.77.mlp.experts.27.up_proj", "model.layers.77.mlp.experts.28.up_proj", "model.layers.77.mlp.experts.29.up_proj", "model.layers.77.mlp.experts.30.up_proj", "model.layers.77.mlp.experts.31.up_proj", "model.layers.77.mlp.experts.32.up_proj", "model.layers.77.mlp.experts.33.up_proj", "model.layers.77.mlp.experts.34.up_proj", "model.layers.77.mlp.experts.35.up_proj", "model.layers.77.mlp.experts.36.up_proj", "model.layers.77.mlp.experts.37.up_proj", "model.layers.77.mlp.experts.38.up_proj", "model.layers.77.mlp.experts.39.up_proj", "model.layers.77.mlp.experts.40.up_proj", "model.layers.77.mlp.experts.41.up_proj", "model.layers.77.mlp.experts.42.up_proj", "model.layers.77.mlp.experts.43.up_proj", "model.layers.77.mlp.experts.44.up_proj", "model.layers.77.mlp.experts.45.up_proj", "model.layers.77.mlp.experts.46.up_proj", "model.layers.77.mlp.experts.47.up_proj", "model.layers.77.mlp.experts.48.up_proj", "model.layers.77.mlp.experts.49.up_proj", "model.layers.77.mlp.experts.50.up_proj", "model.layers.77.mlp.experts.51.up_proj", "model.layers.77.mlp.experts.52.up_proj", "model.layers.77.mlp.experts.53.up_proj", "model.layers.77.mlp.experts.54.up_proj", "model.layers.77.mlp.experts.55.up_proj", "model.layers.77.mlp.experts.56.up_proj", "model.layers.77.mlp.experts.57.up_proj", "model.layers.77.mlp.experts.58.up_proj", "model.layers.77.mlp.experts.59.up_proj", "model.layers.77.mlp.experts.60.up_proj", "model.layers.77.mlp.experts.61.up_proj", "model.layers.77.mlp.experts.62.up_proj", "model.layers.77.mlp.experts.63.up_proj", "model.layers.77.mlp.experts.64.up_proj", "model.layers.77.mlp.experts.65.up_proj", "model.layers.77.mlp.experts.66.up_proj", "model.layers.77.mlp.experts.67.up_proj", "model.layers.77.mlp.experts.68.up_proj", "model.layers.77.mlp.experts.69.up_proj", "model.layers.77.mlp.experts.70.up_proj", "model.layers.77.mlp.experts.71.up_proj", "model.layers.77.mlp.experts.72.up_proj", "model.layers.77.mlp.experts.73.up_proj", "model.layers.77.mlp.experts.74.up_proj", "model.layers.77.mlp.experts.75.up_proj", "model.layers.77.mlp.experts.76.up_proj", "model.layers.77.mlp.experts.77.up_proj", "model.layers.77.mlp.experts.78.up_proj", "model.layers.77.mlp.experts.79.up_proj", "model.layers.77.mlp.experts.80.up_proj", "model.layers.77.mlp.experts.81.up_proj", "model.layers.77.mlp.experts.82.up_proj", "model.layers.77.mlp.experts.83.up_proj", "model.layers.77.mlp.experts.84.up_proj", "model.layers.77.mlp.experts.85.up_proj", "model.layers.77.mlp.experts.86.up_proj", "model.layers.77.mlp.experts.87.up_proj", "model.layers.77.mlp.experts.88.up_proj", "model.layers.77.mlp.experts.89.up_proj", "model.layers.77.mlp.experts.90.up_proj", "model.layers.77.mlp.experts.91.up_proj", "model.layers.77.mlp.experts.92.up_proj", "model.layers.77.mlp.experts.93.up_proj", "model.layers.77.mlp.experts.94.up_proj", "model.layers.77.mlp.experts.95.up_proj", "model.layers.77.mlp.experts.96.up_proj", "model.layers.77.mlp.experts.97.up_proj", "model.layers.77.mlp.experts.98.up_proj", "model.layers.77.mlp.experts.99.up_proj", "model.layers.77.mlp.experts.100.up_proj", "model.layers.77.mlp.experts.101.up_proj", "model.layers.77.mlp.experts.102.up_proj", "model.layers.77.mlp.experts.103.up_proj", "model.layers.77.mlp.experts.104.up_proj", "model.layers.77.mlp.experts.105.up_proj", "model.layers.77.mlp.experts.106.up_proj", "model.layers.77.mlp.experts.107.up_proj", "model.layers.77.mlp.experts.108.up_proj", "model.layers.77.mlp.experts.109.up_proj", "model.layers.77.mlp.experts.110.up_proj", "model.layers.77.mlp.experts.111.up_proj", "model.layers.77.mlp.experts.112.up_proj", "model.layers.77.mlp.experts.113.up_proj", "model.layers.77.mlp.experts.114.up_proj", "model.layers.77.mlp.experts.115.up_proj", "model.layers.77.mlp.experts.116.up_proj", "model.layers.77.mlp.experts.117.up_proj", "model.layers.77.mlp.experts.118.up_proj", "model.layers.77.mlp.experts.119.up_proj", "model.layers.77.mlp.experts.120.up_proj", "model.layers.77.mlp.experts.121.up_proj", "model.layers.77.mlp.experts.122.up_proj", "model.layers.77.mlp.experts.123.up_proj", "model.layers.77.mlp.experts.124.up_proj", "model.layers.77.mlp.experts.125.up_proj", "model.layers.77.mlp.experts.126.up_proj", "model.layers.77.mlp.experts.127.up_proj", "model.layers.77.mlp.experts.128.up_proj", "model.layers.77.mlp.experts.129.up_proj", "model.layers.77.mlp.experts.130.up_proj", "model.layers.77.mlp.experts.131.up_proj", "model.layers.77.mlp.experts.132.up_proj", "model.layers.77.mlp.experts.133.up_proj", "model.layers.77.mlp.experts.134.up_proj", "model.layers.77.mlp.experts.135.up_proj", "model.layers.77.mlp.experts.136.up_proj", "model.layers.77.mlp.experts.137.up_proj", "model.layers.77.mlp.experts.138.up_proj", "model.layers.77.mlp.experts.139.up_proj", "model.layers.77.mlp.experts.140.up_proj", "model.layers.77.mlp.experts.141.up_proj", "model.layers.77.mlp.experts.142.up_proj", "model.layers.77.mlp.experts.143.up_proj", "model.layers.77.mlp.experts.144.up_proj", "model.layers.77.mlp.experts.145.up_proj", "model.layers.77.mlp.experts.146.up_proj", "model.layers.77.mlp.experts.147.up_proj", "model.layers.77.mlp.experts.148.up_proj", "model.layers.77.mlp.experts.149.up_proj", "model.layers.77.mlp.experts.150.up_proj", "model.layers.77.mlp.experts.151.up_proj", "model.layers.77.mlp.experts.152.up_proj", "model.layers.77.mlp.experts.153.up_proj", "model.layers.77.mlp.experts.154.up_proj", "model.layers.77.mlp.experts.155.up_proj", "model.layers.77.mlp.experts.156.up_proj", "model.layers.77.mlp.experts.157.up_proj", "model.layers.77.mlp.experts.158.up_proj", "model.layers.77.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000303406640887266, "dbits": 2516582400 }, { "dkld": -0.00037640603259206373, "dbits": 5033164800 }, { "dkld": -0.00034390445798636315, "dbits": 7549747200 }, { "dkld": -0.00034387242048979083, "dbits": 12582912000 } ] }, { "idx": 464, "layers": [ "model.layers.77.mlp.experts.0.down_proj", "model.layers.77.mlp.experts.1.down_proj", "model.layers.77.mlp.experts.2.down_proj", "model.layers.77.mlp.experts.3.down_proj", "model.layers.77.mlp.experts.4.down_proj", "model.layers.77.mlp.experts.5.down_proj", "model.layers.77.mlp.experts.6.down_proj", "model.layers.77.mlp.experts.7.down_proj", "model.layers.77.mlp.experts.8.down_proj", "model.layers.77.mlp.experts.9.down_proj", "model.layers.77.mlp.experts.10.down_proj", "model.layers.77.mlp.experts.11.down_proj", "model.layers.77.mlp.experts.12.down_proj", "model.layers.77.mlp.experts.13.down_proj", "model.layers.77.mlp.experts.14.down_proj", "model.layers.77.mlp.experts.15.down_proj", "model.layers.77.mlp.experts.16.down_proj", "model.layers.77.mlp.experts.17.down_proj", "model.layers.77.mlp.experts.18.down_proj", "model.layers.77.mlp.experts.19.down_proj", "model.layers.77.mlp.experts.20.down_proj", "model.layers.77.mlp.experts.21.down_proj", "model.layers.77.mlp.experts.22.down_proj", "model.layers.77.mlp.experts.23.down_proj", "model.layers.77.mlp.experts.24.down_proj", "model.layers.77.mlp.experts.25.down_proj", "model.layers.77.mlp.experts.26.down_proj", "model.layers.77.mlp.experts.27.down_proj", "model.layers.77.mlp.experts.28.down_proj", "model.layers.77.mlp.experts.29.down_proj", "model.layers.77.mlp.experts.30.down_proj", "model.layers.77.mlp.experts.31.down_proj", "model.layers.77.mlp.experts.32.down_proj", "model.layers.77.mlp.experts.33.down_proj", "model.layers.77.mlp.experts.34.down_proj", "model.layers.77.mlp.experts.35.down_proj", "model.layers.77.mlp.experts.36.down_proj", "model.layers.77.mlp.experts.37.down_proj", "model.layers.77.mlp.experts.38.down_proj", "model.layers.77.mlp.experts.39.down_proj", "model.layers.77.mlp.experts.40.down_proj", "model.layers.77.mlp.experts.41.down_proj", "model.layers.77.mlp.experts.42.down_proj", "model.layers.77.mlp.experts.43.down_proj", "model.layers.77.mlp.experts.44.down_proj", "model.layers.77.mlp.experts.45.down_proj", "model.layers.77.mlp.experts.46.down_proj", "model.layers.77.mlp.experts.47.down_proj", "model.layers.77.mlp.experts.48.down_proj", "model.layers.77.mlp.experts.49.down_proj", "model.layers.77.mlp.experts.50.down_proj", "model.layers.77.mlp.experts.51.down_proj", "model.layers.77.mlp.experts.52.down_proj", "model.layers.77.mlp.experts.53.down_proj", "model.layers.77.mlp.experts.54.down_proj", "model.layers.77.mlp.experts.55.down_proj", "model.layers.77.mlp.experts.56.down_proj", "model.layers.77.mlp.experts.57.down_proj", "model.layers.77.mlp.experts.58.down_proj", "model.layers.77.mlp.experts.59.down_proj", "model.layers.77.mlp.experts.60.down_proj", "model.layers.77.mlp.experts.61.down_proj", "model.layers.77.mlp.experts.62.down_proj", "model.layers.77.mlp.experts.63.down_proj", "model.layers.77.mlp.experts.64.down_proj", "model.layers.77.mlp.experts.65.down_proj", "model.layers.77.mlp.experts.66.down_proj", "model.layers.77.mlp.experts.67.down_proj", "model.layers.77.mlp.experts.68.down_proj", "model.layers.77.mlp.experts.69.down_proj", "model.layers.77.mlp.experts.70.down_proj", "model.layers.77.mlp.experts.71.down_proj", "model.layers.77.mlp.experts.72.down_proj", "model.layers.77.mlp.experts.73.down_proj", "model.layers.77.mlp.experts.74.down_proj", "model.layers.77.mlp.experts.75.down_proj", "model.layers.77.mlp.experts.76.down_proj", "model.layers.77.mlp.experts.77.down_proj", "model.layers.77.mlp.experts.78.down_proj", "model.layers.77.mlp.experts.79.down_proj", "model.layers.77.mlp.experts.80.down_proj", "model.layers.77.mlp.experts.81.down_proj", "model.layers.77.mlp.experts.82.down_proj", "model.layers.77.mlp.experts.83.down_proj", "model.layers.77.mlp.experts.84.down_proj", "model.layers.77.mlp.experts.85.down_proj", "model.layers.77.mlp.experts.86.down_proj", "model.layers.77.mlp.experts.87.down_proj", "model.layers.77.mlp.experts.88.down_proj", "model.layers.77.mlp.experts.89.down_proj", "model.layers.77.mlp.experts.90.down_proj", "model.layers.77.mlp.experts.91.down_proj", "model.layers.77.mlp.experts.92.down_proj", "model.layers.77.mlp.experts.93.down_proj", "model.layers.77.mlp.experts.94.down_proj", "model.layers.77.mlp.experts.95.down_proj", "model.layers.77.mlp.experts.96.down_proj", "model.layers.77.mlp.experts.97.down_proj", "model.layers.77.mlp.experts.98.down_proj", "model.layers.77.mlp.experts.99.down_proj", "model.layers.77.mlp.experts.100.down_proj", "model.layers.77.mlp.experts.101.down_proj", "model.layers.77.mlp.experts.102.down_proj", "model.layers.77.mlp.experts.103.down_proj", "model.layers.77.mlp.experts.104.down_proj", "model.layers.77.mlp.experts.105.down_proj", "model.layers.77.mlp.experts.106.down_proj", "model.layers.77.mlp.experts.107.down_proj", "model.layers.77.mlp.experts.108.down_proj", "model.layers.77.mlp.experts.109.down_proj", "model.layers.77.mlp.experts.110.down_proj", "model.layers.77.mlp.experts.111.down_proj", "model.layers.77.mlp.experts.112.down_proj", "model.layers.77.mlp.experts.113.down_proj", "model.layers.77.mlp.experts.114.down_proj", "model.layers.77.mlp.experts.115.down_proj", "model.layers.77.mlp.experts.116.down_proj", "model.layers.77.mlp.experts.117.down_proj", "model.layers.77.mlp.experts.118.down_proj", "model.layers.77.mlp.experts.119.down_proj", "model.layers.77.mlp.experts.120.down_proj", "model.layers.77.mlp.experts.121.down_proj", "model.layers.77.mlp.experts.122.down_proj", "model.layers.77.mlp.experts.123.down_proj", "model.layers.77.mlp.experts.124.down_proj", "model.layers.77.mlp.experts.125.down_proj", "model.layers.77.mlp.experts.126.down_proj", "model.layers.77.mlp.experts.127.down_proj", "model.layers.77.mlp.experts.128.down_proj", "model.layers.77.mlp.experts.129.down_proj", "model.layers.77.mlp.experts.130.down_proj", "model.layers.77.mlp.experts.131.down_proj", "model.layers.77.mlp.experts.132.down_proj", "model.layers.77.mlp.experts.133.down_proj", "model.layers.77.mlp.experts.134.down_proj", "model.layers.77.mlp.experts.135.down_proj", "model.layers.77.mlp.experts.136.down_proj", "model.layers.77.mlp.experts.137.down_proj", "model.layers.77.mlp.experts.138.down_proj", "model.layers.77.mlp.experts.139.down_proj", "model.layers.77.mlp.experts.140.down_proj", "model.layers.77.mlp.experts.141.down_proj", "model.layers.77.mlp.experts.142.down_proj", "model.layers.77.mlp.experts.143.down_proj", "model.layers.77.mlp.experts.144.down_proj", "model.layers.77.mlp.experts.145.down_proj", "model.layers.77.mlp.experts.146.down_proj", "model.layers.77.mlp.experts.147.down_proj", "model.layers.77.mlp.experts.148.down_proj", "model.layers.77.mlp.experts.149.down_proj", "model.layers.77.mlp.experts.150.down_proj", "model.layers.77.mlp.experts.151.down_proj", "model.layers.77.mlp.experts.152.down_proj", "model.layers.77.mlp.experts.153.down_proj", "model.layers.77.mlp.experts.154.down_proj", "model.layers.77.mlp.experts.155.down_proj", "model.layers.77.mlp.experts.156.down_proj", "model.layers.77.mlp.experts.157.down_proj", "model.layers.77.mlp.experts.158.down_proj", "model.layers.77.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -2.3428350687026978e-05, "dbits": 1258291200 }, { "dkld": -4.6992022544156686e-05, "dbits": 2516582400 }, { "dkld": -4.7209206968545914e-05, "dbits": 3774873600 }, { "dkld": -4.718676209450323e-05, "dbits": 6291456000 } ] }, { "idx": 465, "layers": [ "model.layers.78.self_attn.q_proj" ], "candidates": [ { "dkld": -9.763129055501141e-05, "dbits": 62914560 }, { "dkld": -1.0660476982601863e-05, "dbits": 125829120 }, { "dkld": -3.477111458778659e-05, "dbits": 188743680 }, { "dkld": -2.132467925548831e-05, "dbits": 314572800 } ] }, { "idx": 466, "layers": [ "model.layers.78.self_attn.k_proj", "model.layers.78.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00040168315172195435, "dbits": 10485760 }, { "dkld": -0.00046522151678801693, "dbits": 20971520 }, { "dkld": -0.0005219470709562357, "dbits": 31457280 }, { "dkld": -0.000486585404723891, "dbits": 52428800 } ] }, { "idx": 467, "layers": [ "model.layers.78.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00016404232010246156, "dbits": 62914560 }, { "dkld": -0.00014531128108502545, "dbits": 125829120 }, { "dkld": -0.00010917056351900101, "dbits": 188743680 }, { "dkld": -0.00012883003801107684, "dbits": 314572800 } ] }, { "idx": 468, "layers": [ "model.layers.78.mlp.shared_experts.gate_proj", "model.layers.78.mlp.shared_experts.up_proj", "model.layers.78.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00010434165596961975, "dbits": 23592960 }, { "dkld": -0.00015214635059238035, "dbits": 47185920 }, { "dkld": -6.169294938446201e-05, "dbits": 70778880 }, { "dkld": -6.661787629128058e-05, "dbits": 117964800 } ] }, { "idx": 469, "layers": [ "model.layers.78.mlp.experts.0.gate_proj", "model.layers.78.mlp.experts.1.gate_proj", "model.layers.78.mlp.experts.2.gate_proj", "model.layers.78.mlp.experts.3.gate_proj", "model.layers.78.mlp.experts.4.gate_proj", "model.layers.78.mlp.experts.5.gate_proj", "model.layers.78.mlp.experts.6.gate_proj", "model.layers.78.mlp.experts.7.gate_proj", "model.layers.78.mlp.experts.8.gate_proj", "model.layers.78.mlp.experts.9.gate_proj", "model.layers.78.mlp.experts.10.gate_proj", "model.layers.78.mlp.experts.11.gate_proj", "model.layers.78.mlp.experts.12.gate_proj", "model.layers.78.mlp.experts.13.gate_proj", "model.layers.78.mlp.experts.14.gate_proj", "model.layers.78.mlp.experts.15.gate_proj", "model.layers.78.mlp.experts.16.gate_proj", "model.layers.78.mlp.experts.17.gate_proj", "model.layers.78.mlp.experts.18.gate_proj", "model.layers.78.mlp.experts.19.gate_proj", "model.layers.78.mlp.experts.20.gate_proj", "model.layers.78.mlp.experts.21.gate_proj", "model.layers.78.mlp.experts.22.gate_proj", "model.layers.78.mlp.experts.23.gate_proj", "model.layers.78.mlp.experts.24.gate_proj", "model.layers.78.mlp.experts.25.gate_proj", "model.layers.78.mlp.experts.26.gate_proj", "model.layers.78.mlp.experts.27.gate_proj", "model.layers.78.mlp.experts.28.gate_proj", "model.layers.78.mlp.experts.29.gate_proj", "model.layers.78.mlp.experts.30.gate_proj", "model.layers.78.mlp.experts.31.gate_proj", "model.layers.78.mlp.experts.32.gate_proj", "model.layers.78.mlp.experts.33.gate_proj", "model.layers.78.mlp.experts.34.gate_proj", "model.layers.78.mlp.experts.35.gate_proj", "model.layers.78.mlp.experts.36.gate_proj", "model.layers.78.mlp.experts.37.gate_proj", "model.layers.78.mlp.experts.38.gate_proj", "model.layers.78.mlp.experts.39.gate_proj", "model.layers.78.mlp.experts.40.gate_proj", "model.layers.78.mlp.experts.41.gate_proj", "model.layers.78.mlp.experts.42.gate_proj", "model.layers.78.mlp.experts.43.gate_proj", "model.layers.78.mlp.experts.44.gate_proj", "model.layers.78.mlp.experts.45.gate_proj", "model.layers.78.mlp.experts.46.gate_proj", "model.layers.78.mlp.experts.47.gate_proj", "model.layers.78.mlp.experts.48.gate_proj", "model.layers.78.mlp.experts.49.gate_proj", "model.layers.78.mlp.experts.50.gate_proj", "model.layers.78.mlp.experts.51.gate_proj", "model.layers.78.mlp.experts.52.gate_proj", "model.layers.78.mlp.experts.53.gate_proj", "model.layers.78.mlp.experts.54.gate_proj", "model.layers.78.mlp.experts.55.gate_proj", "model.layers.78.mlp.experts.56.gate_proj", "model.layers.78.mlp.experts.57.gate_proj", "model.layers.78.mlp.experts.58.gate_proj", "model.layers.78.mlp.experts.59.gate_proj", "model.layers.78.mlp.experts.60.gate_proj", "model.layers.78.mlp.experts.61.gate_proj", "model.layers.78.mlp.experts.62.gate_proj", "model.layers.78.mlp.experts.63.gate_proj", "model.layers.78.mlp.experts.64.gate_proj", "model.layers.78.mlp.experts.65.gate_proj", "model.layers.78.mlp.experts.66.gate_proj", "model.layers.78.mlp.experts.67.gate_proj", "model.layers.78.mlp.experts.68.gate_proj", "model.layers.78.mlp.experts.69.gate_proj", "model.layers.78.mlp.experts.70.gate_proj", "model.layers.78.mlp.experts.71.gate_proj", "model.layers.78.mlp.experts.72.gate_proj", "model.layers.78.mlp.experts.73.gate_proj", "model.layers.78.mlp.experts.74.gate_proj", "model.layers.78.mlp.experts.75.gate_proj", "model.layers.78.mlp.experts.76.gate_proj", "model.layers.78.mlp.experts.77.gate_proj", "model.layers.78.mlp.experts.78.gate_proj", "model.layers.78.mlp.experts.79.gate_proj", "model.layers.78.mlp.experts.80.gate_proj", "model.layers.78.mlp.experts.81.gate_proj", "model.layers.78.mlp.experts.82.gate_proj", "model.layers.78.mlp.experts.83.gate_proj", "model.layers.78.mlp.experts.84.gate_proj", "model.layers.78.mlp.experts.85.gate_proj", "model.layers.78.mlp.experts.86.gate_proj", "model.layers.78.mlp.experts.87.gate_proj", "model.layers.78.mlp.experts.88.gate_proj", "model.layers.78.mlp.experts.89.gate_proj", "model.layers.78.mlp.experts.90.gate_proj", "model.layers.78.mlp.experts.91.gate_proj", "model.layers.78.mlp.experts.92.gate_proj", "model.layers.78.mlp.experts.93.gate_proj", "model.layers.78.mlp.experts.94.gate_proj", "model.layers.78.mlp.experts.95.gate_proj", "model.layers.78.mlp.experts.96.gate_proj", "model.layers.78.mlp.experts.97.gate_proj", "model.layers.78.mlp.experts.98.gate_proj", "model.layers.78.mlp.experts.99.gate_proj", "model.layers.78.mlp.experts.100.gate_proj", "model.layers.78.mlp.experts.101.gate_proj", "model.layers.78.mlp.experts.102.gate_proj", "model.layers.78.mlp.experts.103.gate_proj", "model.layers.78.mlp.experts.104.gate_proj", "model.layers.78.mlp.experts.105.gate_proj", "model.layers.78.mlp.experts.106.gate_proj", "model.layers.78.mlp.experts.107.gate_proj", "model.layers.78.mlp.experts.108.gate_proj", "model.layers.78.mlp.experts.109.gate_proj", "model.layers.78.mlp.experts.110.gate_proj", "model.layers.78.mlp.experts.111.gate_proj", "model.layers.78.mlp.experts.112.gate_proj", "model.layers.78.mlp.experts.113.gate_proj", "model.layers.78.mlp.experts.114.gate_proj", "model.layers.78.mlp.experts.115.gate_proj", "model.layers.78.mlp.experts.116.gate_proj", "model.layers.78.mlp.experts.117.gate_proj", "model.layers.78.mlp.experts.118.gate_proj", "model.layers.78.mlp.experts.119.gate_proj", "model.layers.78.mlp.experts.120.gate_proj", "model.layers.78.mlp.experts.121.gate_proj", "model.layers.78.mlp.experts.122.gate_proj", "model.layers.78.mlp.experts.123.gate_proj", "model.layers.78.mlp.experts.124.gate_proj", "model.layers.78.mlp.experts.125.gate_proj", "model.layers.78.mlp.experts.126.gate_proj", "model.layers.78.mlp.experts.127.gate_proj", "model.layers.78.mlp.experts.128.gate_proj", "model.layers.78.mlp.experts.129.gate_proj", "model.layers.78.mlp.experts.130.gate_proj", "model.layers.78.mlp.experts.131.gate_proj", "model.layers.78.mlp.experts.132.gate_proj", "model.layers.78.mlp.experts.133.gate_proj", "model.layers.78.mlp.experts.134.gate_proj", "model.layers.78.mlp.experts.135.gate_proj", "model.layers.78.mlp.experts.136.gate_proj", "model.layers.78.mlp.experts.137.gate_proj", "model.layers.78.mlp.experts.138.gate_proj", "model.layers.78.mlp.experts.139.gate_proj", "model.layers.78.mlp.experts.140.gate_proj", "model.layers.78.mlp.experts.141.gate_proj", "model.layers.78.mlp.experts.142.gate_proj", "model.layers.78.mlp.experts.143.gate_proj", "model.layers.78.mlp.experts.144.gate_proj", "model.layers.78.mlp.experts.145.gate_proj", "model.layers.78.mlp.experts.146.gate_proj", "model.layers.78.mlp.experts.147.gate_proj", "model.layers.78.mlp.experts.148.gate_proj", "model.layers.78.mlp.experts.149.gate_proj", "model.layers.78.mlp.experts.150.gate_proj", "model.layers.78.mlp.experts.151.gate_proj", "model.layers.78.mlp.experts.152.gate_proj", "model.layers.78.mlp.experts.153.gate_proj", "model.layers.78.mlp.experts.154.gate_proj", "model.layers.78.mlp.experts.155.gate_proj", "model.layers.78.mlp.experts.156.gate_proj", "model.layers.78.mlp.experts.157.gate_proj", "model.layers.78.mlp.experts.158.gate_proj", "model.layers.78.mlp.experts.159.gate_proj", "model.layers.78.mlp.experts.0.up_proj", "model.layers.78.mlp.experts.1.up_proj", "model.layers.78.mlp.experts.2.up_proj", "model.layers.78.mlp.experts.3.up_proj", "model.layers.78.mlp.experts.4.up_proj", "model.layers.78.mlp.experts.5.up_proj", "model.layers.78.mlp.experts.6.up_proj", "model.layers.78.mlp.experts.7.up_proj", "model.layers.78.mlp.experts.8.up_proj", "model.layers.78.mlp.experts.9.up_proj", "model.layers.78.mlp.experts.10.up_proj", "model.layers.78.mlp.experts.11.up_proj", "model.layers.78.mlp.experts.12.up_proj", "model.layers.78.mlp.experts.13.up_proj", "model.layers.78.mlp.experts.14.up_proj", "model.layers.78.mlp.experts.15.up_proj", "model.layers.78.mlp.experts.16.up_proj", "model.layers.78.mlp.experts.17.up_proj", "model.layers.78.mlp.experts.18.up_proj", "model.layers.78.mlp.experts.19.up_proj", "model.layers.78.mlp.experts.20.up_proj", "model.layers.78.mlp.experts.21.up_proj", "model.layers.78.mlp.experts.22.up_proj", "model.layers.78.mlp.experts.23.up_proj", "model.layers.78.mlp.experts.24.up_proj", "model.layers.78.mlp.experts.25.up_proj", "model.layers.78.mlp.experts.26.up_proj", "model.layers.78.mlp.experts.27.up_proj", "model.layers.78.mlp.experts.28.up_proj", "model.layers.78.mlp.experts.29.up_proj", "model.layers.78.mlp.experts.30.up_proj", "model.layers.78.mlp.experts.31.up_proj", "model.layers.78.mlp.experts.32.up_proj", "model.layers.78.mlp.experts.33.up_proj", "model.layers.78.mlp.experts.34.up_proj", "model.layers.78.mlp.experts.35.up_proj", "model.layers.78.mlp.experts.36.up_proj", "model.layers.78.mlp.experts.37.up_proj", "model.layers.78.mlp.experts.38.up_proj", "model.layers.78.mlp.experts.39.up_proj", "model.layers.78.mlp.experts.40.up_proj", "model.layers.78.mlp.experts.41.up_proj", "model.layers.78.mlp.experts.42.up_proj", "model.layers.78.mlp.experts.43.up_proj", "model.layers.78.mlp.experts.44.up_proj", "model.layers.78.mlp.experts.45.up_proj", "model.layers.78.mlp.experts.46.up_proj", "model.layers.78.mlp.experts.47.up_proj", "model.layers.78.mlp.experts.48.up_proj", "model.layers.78.mlp.experts.49.up_proj", "model.layers.78.mlp.experts.50.up_proj", "model.layers.78.mlp.experts.51.up_proj", "model.layers.78.mlp.experts.52.up_proj", "model.layers.78.mlp.experts.53.up_proj", "model.layers.78.mlp.experts.54.up_proj", "model.layers.78.mlp.experts.55.up_proj", "model.layers.78.mlp.experts.56.up_proj", "model.layers.78.mlp.experts.57.up_proj", "model.layers.78.mlp.experts.58.up_proj", "model.layers.78.mlp.experts.59.up_proj", "model.layers.78.mlp.experts.60.up_proj", "model.layers.78.mlp.experts.61.up_proj", "model.layers.78.mlp.experts.62.up_proj", "model.layers.78.mlp.experts.63.up_proj", "model.layers.78.mlp.experts.64.up_proj", "model.layers.78.mlp.experts.65.up_proj", "model.layers.78.mlp.experts.66.up_proj", "model.layers.78.mlp.experts.67.up_proj", "model.layers.78.mlp.experts.68.up_proj", "model.layers.78.mlp.experts.69.up_proj", "model.layers.78.mlp.experts.70.up_proj", "model.layers.78.mlp.experts.71.up_proj", "model.layers.78.mlp.experts.72.up_proj", "model.layers.78.mlp.experts.73.up_proj", "model.layers.78.mlp.experts.74.up_proj", "model.layers.78.mlp.experts.75.up_proj", "model.layers.78.mlp.experts.76.up_proj", "model.layers.78.mlp.experts.77.up_proj", "model.layers.78.mlp.experts.78.up_proj", "model.layers.78.mlp.experts.79.up_proj", "model.layers.78.mlp.experts.80.up_proj", "model.layers.78.mlp.experts.81.up_proj", "model.layers.78.mlp.experts.82.up_proj", "model.layers.78.mlp.experts.83.up_proj", "model.layers.78.mlp.experts.84.up_proj", "model.layers.78.mlp.experts.85.up_proj", "model.layers.78.mlp.experts.86.up_proj", "model.layers.78.mlp.experts.87.up_proj", "model.layers.78.mlp.experts.88.up_proj", "model.layers.78.mlp.experts.89.up_proj", "model.layers.78.mlp.experts.90.up_proj", "model.layers.78.mlp.experts.91.up_proj", "model.layers.78.mlp.experts.92.up_proj", "model.layers.78.mlp.experts.93.up_proj", "model.layers.78.mlp.experts.94.up_proj", "model.layers.78.mlp.experts.95.up_proj", "model.layers.78.mlp.experts.96.up_proj", "model.layers.78.mlp.experts.97.up_proj", "model.layers.78.mlp.experts.98.up_proj", "model.layers.78.mlp.experts.99.up_proj", "model.layers.78.mlp.experts.100.up_proj", "model.layers.78.mlp.experts.101.up_proj", "model.layers.78.mlp.experts.102.up_proj", "model.layers.78.mlp.experts.103.up_proj", "model.layers.78.mlp.experts.104.up_proj", "model.layers.78.mlp.experts.105.up_proj", "model.layers.78.mlp.experts.106.up_proj", "model.layers.78.mlp.experts.107.up_proj", "model.layers.78.mlp.experts.108.up_proj", "model.layers.78.mlp.experts.109.up_proj", "model.layers.78.mlp.experts.110.up_proj", "model.layers.78.mlp.experts.111.up_proj", "model.layers.78.mlp.experts.112.up_proj", "model.layers.78.mlp.experts.113.up_proj", "model.layers.78.mlp.experts.114.up_proj", "model.layers.78.mlp.experts.115.up_proj", "model.layers.78.mlp.experts.116.up_proj", "model.layers.78.mlp.experts.117.up_proj", "model.layers.78.mlp.experts.118.up_proj", "model.layers.78.mlp.experts.119.up_proj", "model.layers.78.mlp.experts.120.up_proj", "model.layers.78.mlp.experts.121.up_proj", "model.layers.78.mlp.experts.122.up_proj", "model.layers.78.mlp.experts.123.up_proj", "model.layers.78.mlp.experts.124.up_proj", "model.layers.78.mlp.experts.125.up_proj", "model.layers.78.mlp.experts.126.up_proj", "model.layers.78.mlp.experts.127.up_proj", "model.layers.78.mlp.experts.128.up_proj", "model.layers.78.mlp.experts.129.up_proj", "model.layers.78.mlp.experts.130.up_proj", "model.layers.78.mlp.experts.131.up_proj", "model.layers.78.mlp.experts.132.up_proj", "model.layers.78.mlp.experts.133.up_proj", "model.layers.78.mlp.experts.134.up_proj", "model.layers.78.mlp.experts.135.up_proj", "model.layers.78.mlp.experts.136.up_proj", "model.layers.78.mlp.experts.137.up_proj", "model.layers.78.mlp.experts.138.up_proj", "model.layers.78.mlp.experts.139.up_proj", "model.layers.78.mlp.experts.140.up_proj", "model.layers.78.mlp.experts.141.up_proj", "model.layers.78.mlp.experts.142.up_proj", "model.layers.78.mlp.experts.143.up_proj", "model.layers.78.mlp.experts.144.up_proj", "model.layers.78.mlp.experts.145.up_proj", "model.layers.78.mlp.experts.146.up_proj", "model.layers.78.mlp.experts.147.up_proj", "model.layers.78.mlp.experts.148.up_proj", "model.layers.78.mlp.experts.149.up_proj", "model.layers.78.mlp.experts.150.up_proj", "model.layers.78.mlp.experts.151.up_proj", "model.layers.78.mlp.experts.152.up_proj", "model.layers.78.mlp.experts.153.up_proj", "model.layers.78.mlp.experts.154.up_proj", "model.layers.78.mlp.experts.155.up_proj", "model.layers.78.mlp.experts.156.up_proj", "model.layers.78.mlp.experts.157.up_proj", "model.layers.78.mlp.experts.158.up_proj", "model.layers.78.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00011121965944767276, "dbits": 2516582400 }, { "dkld": -7.472019642591754e-05, "dbits": 5033164800 }, { "dkld": -2.4225935339933224e-05, "dbits": 7549747200 }, { "dkld": -7.638232782483656e-05, "dbits": 12582912000 } ] }, { "idx": 470, "layers": [ "model.layers.78.mlp.experts.0.down_proj", "model.layers.78.mlp.experts.1.down_proj", "model.layers.78.mlp.experts.2.down_proj", "model.layers.78.mlp.experts.3.down_proj", "model.layers.78.mlp.experts.4.down_proj", "model.layers.78.mlp.experts.5.down_proj", "model.layers.78.mlp.experts.6.down_proj", "model.layers.78.mlp.experts.7.down_proj", "model.layers.78.mlp.experts.8.down_proj", "model.layers.78.mlp.experts.9.down_proj", "model.layers.78.mlp.experts.10.down_proj", "model.layers.78.mlp.experts.11.down_proj", "model.layers.78.mlp.experts.12.down_proj", "model.layers.78.mlp.experts.13.down_proj", "model.layers.78.mlp.experts.14.down_proj", "model.layers.78.mlp.experts.15.down_proj", "model.layers.78.mlp.experts.16.down_proj", "model.layers.78.mlp.experts.17.down_proj", "model.layers.78.mlp.experts.18.down_proj", "model.layers.78.mlp.experts.19.down_proj", "model.layers.78.mlp.experts.20.down_proj", "model.layers.78.mlp.experts.21.down_proj", "model.layers.78.mlp.experts.22.down_proj", "model.layers.78.mlp.experts.23.down_proj", "model.layers.78.mlp.experts.24.down_proj", "model.layers.78.mlp.experts.25.down_proj", "model.layers.78.mlp.experts.26.down_proj", "model.layers.78.mlp.experts.27.down_proj", "model.layers.78.mlp.experts.28.down_proj", "model.layers.78.mlp.experts.29.down_proj", "model.layers.78.mlp.experts.30.down_proj", "model.layers.78.mlp.experts.31.down_proj", "model.layers.78.mlp.experts.32.down_proj", "model.layers.78.mlp.experts.33.down_proj", "model.layers.78.mlp.experts.34.down_proj", "model.layers.78.mlp.experts.35.down_proj", "model.layers.78.mlp.experts.36.down_proj", "model.layers.78.mlp.experts.37.down_proj", "model.layers.78.mlp.experts.38.down_proj", "model.layers.78.mlp.experts.39.down_proj", "model.layers.78.mlp.experts.40.down_proj", "model.layers.78.mlp.experts.41.down_proj", "model.layers.78.mlp.experts.42.down_proj", "model.layers.78.mlp.experts.43.down_proj", "model.layers.78.mlp.experts.44.down_proj", "model.layers.78.mlp.experts.45.down_proj", "model.layers.78.mlp.experts.46.down_proj", "model.layers.78.mlp.experts.47.down_proj", "model.layers.78.mlp.experts.48.down_proj", "model.layers.78.mlp.experts.49.down_proj", "model.layers.78.mlp.experts.50.down_proj", "model.layers.78.mlp.experts.51.down_proj", "model.layers.78.mlp.experts.52.down_proj", "model.layers.78.mlp.experts.53.down_proj", "model.layers.78.mlp.experts.54.down_proj", "model.layers.78.mlp.experts.55.down_proj", "model.layers.78.mlp.experts.56.down_proj", "model.layers.78.mlp.experts.57.down_proj", "model.layers.78.mlp.experts.58.down_proj", "model.layers.78.mlp.experts.59.down_proj", "model.layers.78.mlp.experts.60.down_proj", "model.layers.78.mlp.experts.61.down_proj", "model.layers.78.mlp.experts.62.down_proj", "model.layers.78.mlp.experts.63.down_proj", "model.layers.78.mlp.experts.64.down_proj", "model.layers.78.mlp.experts.65.down_proj", "model.layers.78.mlp.experts.66.down_proj", "model.layers.78.mlp.experts.67.down_proj", "model.layers.78.mlp.experts.68.down_proj", "model.layers.78.mlp.experts.69.down_proj", "model.layers.78.mlp.experts.70.down_proj", "model.layers.78.mlp.experts.71.down_proj", "model.layers.78.mlp.experts.72.down_proj", "model.layers.78.mlp.experts.73.down_proj", "model.layers.78.mlp.experts.74.down_proj", "model.layers.78.mlp.experts.75.down_proj", "model.layers.78.mlp.experts.76.down_proj", "model.layers.78.mlp.experts.77.down_proj", "model.layers.78.mlp.experts.78.down_proj", "model.layers.78.mlp.experts.79.down_proj", "model.layers.78.mlp.experts.80.down_proj", "model.layers.78.mlp.experts.81.down_proj", "model.layers.78.mlp.experts.82.down_proj", "model.layers.78.mlp.experts.83.down_proj", "model.layers.78.mlp.experts.84.down_proj", "model.layers.78.mlp.experts.85.down_proj", "model.layers.78.mlp.experts.86.down_proj", "model.layers.78.mlp.experts.87.down_proj", "model.layers.78.mlp.experts.88.down_proj", "model.layers.78.mlp.experts.89.down_proj", "model.layers.78.mlp.experts.90.down_proj", "model.layers.78.mlp.experts.91.down_proj", "model.layers.78.mlp.experts.92.down_proj", "model.layers.78.mlp.experts.93.down_proj", "model.layers.78.mlp.experts.94.down_proj", "model.layers.78.mlp.experts.95.down_proj", "model.layers.78.mlp.experts.96.down_proj", "model.layers.78.mlp.experts.97.down_proj", "model.layers.78.mlp.experts.98.down_proj", "model.layers.78.mlp.experts.99.down_proj", "model.layers.78.mlp.experts.100.down_proj", "model.layers.78.mlp.experts.101.down_proj", "model.layers.78.mlp.experts.102.down_proj", "model.layers.78.mlp.experts.103.down_proj", "model.layers.78.mlp.experts.104.down_proj", "model.layers.78.mlp.experts.105.down_proj", "model.layers.78.mlp.experts.106.down_proj", "model.layers.78.mlp.experts.107.down_proj", "model.layers.78.mlp.experts.108.down_proj", "model.layers.78.mlp.experts.109.down_proj", "model.layers.78.mlp.experts.110.down_proj", "model.layers.78.mlp.experts.111.down_proj", "model.layers.78.mlp.experts.112.down_proj", "model.layers.78.mlp.experts.113.down_proj", "model.layers.78.mlp.experts.114.down_proj", "model.layers.78.mlp.experts.115.down_proj", "model.layers.78.mlp.experts.116.down_proj", "model.layers.78.mlp.experts.117.down_proj", "model.layers.78.mlp.experts.118.down_proj", "model.layers.78.mlp.experts.119.down_proj", "model.layers.78.mlp.experts.120.down_proj", "model.layers.78.mlp.experts.121.down_proj", "model.layers.78.mlp.experts.122.down_proj", "model.layers.78.mlp.experts.123.down_proj", "model.layers.78.mlp.experts.124.down_proj", "model.layers.78.mlp.experts.125.down_proj", "model.layers.78.mlp.experts.126.down_proj", "model.layers.78.mlp.experts.127.down_proj", "model.layers.78.mlp.experts.128.down_proj", "model.layers.78.mlp.experts.129.down_proj", "model.layers.78.mlp.experts.130.down_proj", "model.layers.78.mlp.experts.131.down_proj", "model.layers.78.mlp.experts.132.down_proj", "model.layers.78.mlp.experts.133.down_proj", "model.layers.78.mlp.experts.134.down_proj", "model.layers.78.mlp.experts.135.down_proj", "model.layers.78.mlp.experts.136.down_proj", "model.layers.78.mlp.experts.137.down_proj", "model.layers.78.mlp.experts.138.down_proj", "model.layers.78.mlp.experts.139.down_proj", "model.layers.78.mlp.experts.140.down_proj", "model.layers.78.mlp.experts.141.down_proj", "model.layers.78.mlp.experts.142.down_proj", "model.layers.78.mlp.experts.143.down_proj", "model.layers.78.mlp.experts.144.down_proj", "model.layers.78.mlp.experts.145.down_proj", "model.layers.78.mlp.experts.146.down_proj", "model.layers.78.mlp.experts.147.down_proj", "model.layers.78.mlp.experts.148.down_proj", "model.layers.78.mlp.experts.149.down_proj", "model.layers.78.mlp.experts.150.down_proj", "model.layers.78.mlp.experts.151.down_proj", "model.layers.78.mlp.experts.152.down_proj", "model.layers.78.mlp.experts.153.down_proj", "model.layers.78.mlp.experts.154.down_proj", "model.layers.78.mlp.experts.155.down_proj", "model.layers.78.mlp.experts.156.down_proj", "model.layers.78.mlp.experts.157.down_proj", "model.layers.78.mlp.experts.158.down_proj", "model.layers.78.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -7.565654814244149e-05, "dbits": 1258291200 }, { "dkld": -0.00010496592149139561, "dbits": 2516582400 }, { "dkld": -0.00013125380501151362, "dbits": 3774873600 }, { "dkld": -0.00010989783331752379, "dbits": 6291456000 } ] }, { "idx": 471, "layers": [ "model.layers.79.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00026848074048757553, "dbits": 62914560 }, { "dkld": -0.000276188459247359, "dbits": 125829120 }, { "dkld": -0.0003246402367949597, "dbits": 188743680 }, { "dkld": -0.000321862567216169, "dbits": 314572800 } ] }, { "idx": 472, "layers": [ "model.layers.79.self_attn.k_proj", "model.layers.79.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0005183385685086361, "dbits": 10485760 }, { "dkld": -0.0009613214060664177, "dbits": 20971520 }, { "dkld": -0.0010623502545058783, "dbits": 31457280 }, { "dkld": -0.0010292579419911019, "dbits": 52428800 } ] }, { "idx": 473, "layers": [ "model.layers.79.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003696668893098942, "dbits": 62914560 }, { "dkld": -0.0003522401675581932, "dbits": 125829120 }, { "dkld": -0.00030205845832825817, "dbits": 188743680 }, { "dkld": -0.0003377574495971314, "dbits": 314572800 } ] }, { "idx": 474, "layers": [ "model.layers.79.mlp.shared_experts.gate_proj", "model.layers.79.mlp.shared_experts.up_proj", "model.layers.79.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0006721338257193593, "dbits": 23592960 }, { "dkld": -0.0007006955333054149, "dbits": 47185920 }, { "dkld": -0.0006860432215035073, "dbits": 70778880 }, { "dkld": -0.0007461961358785629, "dbits": 117964800 } ] }, { "idx": 475, "layers": [ "model.layers.79.mlp.experts.0.gate_proj", "model.layers.79.mlp.experts.1.gate_proj", "model.layers.79.mlp.experts.2.gate_proj", "model.layers.79.mlp.experts.3.gate_proj", "model.layers.79.mlp.experts.4.gate_proj", "model.layers.79.mlp.experts.5.gate_proj", "model.layers.79.mlp.experts.6.gate_proj", "model.layers.79.mlp.experts.7.gate_proj", "model.layers.79.mlp.experts.8.gate_proj", "model.layers.79.mlp.experts.9.gate_proj", "model.layers.79.mlp.experts.10.gate_proj", "model.layers.79.mlp.experts.11.gate_proj", "model.layers.79.mlp.experts.12.gate_proj", "model.layers.79.mlp.experts.13.gate_proj", "model.layers.79.mlp.experts.14.gate_proj", "model.layers.79.mlp.experts.15.gate_proj", "model.layers.79.mlp.experts.16.gate_proj", "model.layers.79.mlp.experts.17.gate_proj", "model.layers.79.mlp.experts.18.gate_proj", "model.layers.79.mlp.experts.19.gate_proj", "model.layers.79.mlp.experts.20.gate_proj", "model.layers.79.mlp.experts.21.gate_proj", "model.layers.79.mlp.experts.22.gate_proj", "model.layers.79.mlp.experts.23.gate_proj", "model.layers.79.mlp.experts.24.gate_proj", "model.layers.79.mlp.experts.25.gate_proj", "model.layers.79.mlp.experts.26.gate_proj", "model.layers.79.mlp.experts.27.gate_proj", "model.layers.79.mlp.experts.28.gate_proj", "model.layers.79.mlp.experts.29.gate_proj", "model.layers.79.mlp.experts.30.gate_proj", "model.layers.79.mlp.experts.31.gate_proj", "model.layers.79.mlp.experts.32.gate_proj", "model.layers.79.mlp.experts.33.gate_proj", "model.layers.79.mlp.experts.34.gate_proj", "model.layers.79.mlp.experts.35.gate_proj", "model.layers.79.mlp.experts.36.gate_proj", "model.layers.79.mlp.experts.37.gate_proj", "model.layers.79.mlp.experts.38.gate_proj", "model.layers.79.mlp.experts.39.gate_proj", "model.layers.79.mlp.experts.40.gate_proj", "model.layers.79.mlp.experts.41.gate_proj", "model.layers.79.mlp.experts.42.gate_proj", "model.layers.79.mlp.experts.43.gate_proj", "model.layers.79.mlp.experts.44.gate_proj", "model.layers.79.mlp.experts.45.gate_proj", "model.layers.79.mlp.experts.46.gate_proj", "model.layers.79.mlp.experts.47.gate_proj", "model.layers.79.mlp.experts.48.gate_proj", "model.layers.79.mlp.experts.49.gate_proj", "model.layers.79.mlp.experts.50.gate_proj", "model.layers.79.mlp.experts.51.gate_proj", "model.layers.79.mlp.experts.52.gate_proj", "model.layers.79.mlp.experts.53.gate_proj", "model.layers.79.mlp.experts.54.gate_proj", "model.layers.79.mlp.experts.55.gate_proj", "model.layers.79.mlp.experts.56.gate_proj", "model.layers.79.mlp.experts.57.gate_proj", "model.layers.79.mlp.experts.58.gate_proj", "model.layers.79.mlp.experts.59.gate_proj", "model.layers.79.mlp.experts.60.gate_proj", "model.layers.79.mlp.experts.61.gate_proj", "model.layers.79.mlp.experts.62.gate_proj", "model.layers.79.mlp.experts.63.gate_proj", "model.layers.79.mlp.experts.64.gate_proj", "model.layers.79.mlp.experts.65.gate_proj", "model.layers.79.mlp.experts.66.gate_proj", "model.layers.79.mlp.experts.67.gate_proj", "model.layers.79.mlp.experts.68.gate_proj", "model.layers.79.mlp.experts.69.gate_proj", "model.layers.79.mlp.experts.70.gate_proj", "model.layers.79.mlp.experts.71.gate_proj", "model.layers.79.mlp.experts.72.gate_proj", "model.layers.79.mlp.experts.73.gate_proj", "model.layers.79.mlp.experts.74.gate_proj", "model.layers.79.mlp.experts.75.gate_proj", "model.layers.79.mlp.experts.76.gate_proj", "model.layers.79.mlp.experts.77.gate_proj", "model.layers.79.mlp.experts.78.gate_proj", "model.layers.79.mlp.experts.79.gate_proj", "model.layers.79.mlp.experts.80.gate_proj", "model.layers.79.mlp.experts.81.gate_proj", "model.layers.79.mlp.experts.82.gate_proj", "model.layers.79.mlp.experts.83.gate_proj", "model.layers.79.mlp.experts.84.gate_proj", "model.layers.79.mlp.experts.85.gate_proj", "model.layers.79.mlp.experts.86.gate_proj", "model.layers.79.mlp.experts.87.gate_proj", "model.layers.79.mlp.experts.88.gate_proj", "model.layers.79.mlp.experts.89.gate_proj", "model.layers.79.mlp.experts.90.gate_proj", "model.layers.79.mlp.experts.91.gate_proj", "model.layers.79.mlp.experts.92.gate_proj", "model.layers.79.mlp.experts.93.gate_proj", "model.layers.79.mlp.experts.94.gate_proj", "model.layers.79.mlp.experts.95.gate_proj", "model.layers.79.mlp.experts.96.gate_proj", "model.layers.79.mlp.experts.97.gate_proj", "model.layers.79.mlp.experts.98.gate_proj", "model.layers.79.mlp.experts.99.gate_proj", "model.layers.79.mlp.experts.100.gate_proj", "model.layers.79.mlp.experts.101.gate_proj", "model.layers.79.mlp.experts.102.gate_proj", "model.layers.79.mlp.experts.103.gate_proj", "model.layers.79.mlp.experts.104.gate_proj", "model.layers.79.mlp.experts.105.gate_proj", "model.layers.79.mlp.experts.106.gate_proj", "model.layers.79.mlp.experts.107.gate_proj", "model.layers.79.mlp.experts.108.gate_proj", "model.layers.79.mlp.experts.109.gate_proj", "model.layers.79.mlp.experts.110.gate_proj", "model.layers.79.mlp.experts.111.gate_proj", "model.layers.79.mlp.experts.112.gate_proj", "model.layers.79.mlp.experts.113.gate_proj", "model.layers.79.mlp.experts.114.gate_proj", "model.layers.79.mlp.experts.115.gate_proj", "model.layers.79.mlp.experts.116.gate_proj", "model.layers.79.mlp.experts.117.gate_proj", "model.layers.79.mlp.experts.118.gate_proj", "model.layers.79.mlp.experts.119.gate_proj", "model.layers.79.mlp.experts.120.gate_proj", "model.layers.79.mlp.experts.121.gate_proj", "model.layers.79.mlp.experts.122.gate_proj", "model.layers.79.mlp.experts.123.gate_proj", "model.layers.79.mlp.experts.124.gate_proj", "model.layers.79.mlp.experts.125.gate_proj", "model.layers.79.mlp.experts.126.gate_proj", "model.layers.79.mlp.experts.127.gate_proj", "model.layers.79.mlp.experts.128.gate_proj", "model.layers.79.mlp.experts.129.gate_proj", "model.layers.79.mlp.experts.130.gate_proj", "model.layers.79.mlp.experts.131.gate_proj", "model.layers.79.mlp.experts.132.gate_proj", "model.layers.79.mlp.experts.133.gate_proj", "model.layers.79.mlp.experts.134.gate_proj", "model.layers.79.mlp.experts.135.gate_proj", "model.layers.79.mlp.experts.136.gate_proj", "model.layers.79.mlp.experts.137.gate_proj", "model.layers.79.mlp.experts.138.gate_proj", "model.layers.79.mlp.experts.139.gate_proj", "model.layers.79.mlp.experts.140.gate_proj", "model.layers.79.mlp.experts.141.gate_proj", "model.layers.79.mlp.experts.142.gate_proj", "model.layers.79.mlp.experts.143.gate_proj", "model.layers.79.mlp.experts.144.gate_proj", "model.layers.79.mlp.experts.145.gate_proj", "model.layers.79.mlp.experts.146.gate_proj", "model.layers.79.mlp.experts.147.gate_proj", "model.layers.79.mlp.experts.148.gate_proj", "model.layers.79.mlp.experts.149.gate_proj", "model.layers.79.mlp.experts.150.gate_proj", "model.layers.79.mlp.experts.151.gate_proj", "model.layers.79.mlp.experts.152.gate_proj", "model.layers.79.mlp.experts.153.gate_proj", "model.layers.79.mlp.experts.154.gate_proj", "model.layers.79.mlp.experts.155.gate_proj", "model.layers.79.mlp.experts.156.gate_proj", "model.layers.79.mlp.experts.157.gate_proj", "model.layers.79.mlp.experts.158.gate_proj", "model.layers.79.mlp.experts.159.gate_proj", "model.layers.79.mlp.experts.0.up_proj", "model.layers.79.mlp.experts.1.up_proj", "model.layers.79.mlp.experts.2.up_proj", "model.layers.79.mlp.experts.3.up_proj", "model.layers.79.mlp.experts.4.up_proj", "model.layers.79.mlp.experts.5.up_proj", "model.layers.79.mlp.experts.6.up_proj", "model.layers.79.mlp.experts.7.up_proj", "model.layers.79.mlp.experts.8.up_proj", "model.layers.79.mlp.experts.9.up_proj", "model.layers.79.mlp.experts.10.up_proj", "model.layers.79.mlp.experts.11.up_proj", "model.layers.79.mlp.experts.12.up_proj", "model.layers.79.mlp.experts.13.up_proj", "model.layers.79.mlp.experts.14.up_proj", "model.layers.79.mlp.experts.15.up_proj", "model.layers.79.mlp.experts.16.up_proj", "model.layers.79.mlp.experts.17.up_proj", "model.layers.79.mlp.experts.18.up_proj", "model.layers.79.mlp.experts.19.up_proj", "model.layers.79.mlp.experts.20.up_proj", "model.layers.79.mlp.experts.21.up_proj", "model.layers.79.mlp.experts.22.up_proj", "model.layers.79.mlp.experts.23.up_proj", "model.layers.79.mlp.experts.24.up_proj", "model.layers.79.mlp.experts.25.up_proj", "model.layers.79.mlp.experts.26.up_proj", "model.layers.79.mlp.experts.27.up_proj", "model.layers.79.mlp.experts.28.up_proj", "model.layers.79.mlp.experts.29.up_proj", "model.layers.79.mlp.experts.30.up_proj", "model.layers.79.mlp.experts.31.up_proj", "model.layers.79.mlp.experts.32.up_proj", "model.layers.79.mlp.experts.33.up_proj", "model.layers.79.mlp.experts.34.up_proj", "model.layers.79.mlp.experts.35.up_proj", "model.layers.79.mlp.experts.36.up_proj", "model.layers.79.mlp.experts.37.up_proj", "model.layers.79.mlp.experts.38.up_proj", "model.layers.79.mlp.experts.39.up_proj", "model.layers.79.mlp.experts.40.up_proj", "model.layers.79.mlp.experts.41.up_proj", "model.layers.79.mlp.experts.42.up_proj", "model.layers.79.mlp.experts.43.up_proj", "model.layers.79.mlp.experts.44.up_proj", "model.layers.79.mlp.experts.45.up_proj", "model.layers.79.mlp.experts.46.up_proj", "model.layers.79.mlp.experts.47.up_proj", "model.layers.79.mlp.experts.48.up_proj", "model.layers.79.mlp.experts.49.up_proj", "model.layers.79.mlp.experts.50.up_proj", "model.layers.79.mlp.experts.51.up_proj", "model.layers.79.mlp.experts.52.up_proj", "model.layers.79.mlp.experts.53.up_proj", "model.layers.79.mlp.experts.54.up_proj", "model.layers.79.mlp.experts.55.up_proj", "model.layers.79.mlp.experts.56.up_proj", "model.layers.79.mlp.experts.57.up_proj", "model.layers.79.mlp.experts.58.up_proj", "model.layers.79.mlp.experts.59.up_proj", "model.layers.79.mlp.experts.60.up_proj", "model.layers.79.mlp.experts.61.up_proj", "model.layers.79.mlp.experts.62.up_proj", "model.layers.79.mlp.experts.63.up_proj", "model.layers.79.mlp.experts.64.up_proj", "model.layers.79.mlp.experts.65.up_proj", "model.layers.79.mlp.experts.66.up_proj", "model.layers.79.mlp.experts.67.up_proj", "model.layers.79.mlp.experts.68.up_proj", "model.layers.79.mlp.experts.69.up_proj", "model.layers.79.mlp.experts.70.up_proj", "model.layers.79.mlp.experts.71.up_proj", "model.layers.79.mlp.experts.72.up_proj", "model.layers.79.mlp.experts.73.up_proj", "model.layers.79.mlp.experts.74.up_proj", "model.layers.79.mlp.experts.75.up_proj", "model.layers.79.mlp.experts.76.up_proj", "model.layers.79.mlp.experts.77.up_proj", "model.layers.79.mlp.experts.78.up_proj", "model.layers.79.mlp.experts.79.up_proj", "model.layers.79.mlp.experts.80.up_proj", "model.layers.79.mlp.experts.81.up_proj", "model.layers.79.mlp.experts.82.up_proj", "model.layers.79.mlp.experts.83.up_proj", "model.layers.79.mlp.experts.84.up_proj", "model.layers.79.mlp.experts.85.up_proj", "model.layers.79.mlp.experts.86.up_proj", "model.layers.79.mlp.experts.87.up_proj", "model.layers.79.mlp.experts.88.up_proj", "model.layers.79.mlp.experts.89.up_proj", "model.layers.79.mlp.experts.90.up_proj", "model.layers.79.mlp.experts.91.up_proj", "model.layers.79.mlp.experts.92.up_proj", "model.layers.79.mlp.experts.93.up_proj", "model.layers.79.mlp.experts.94.up_proj", "model.layers.79.mlp.experts.95.up_proj", "model.layers.79.mlp.experts.96.up_proj", "model.layers.79.mlp.experts.97.up_proj", "model.layers.79.mlp.experts.98.up_proj", "model.layers.79.mlp.experts.99.up_proj", "model.layers.79.mlp.experts.100.up_proj", "model.layers.79.mlp.experts.101.up_proj", "model.layers.79.mlp.experts.102.up_proj", "model.layers.79.mlp.experts.103.up_proj", "model.layers.79.mlp.experts.104.up_proj", "model.layers.79.mlp.experts.105.up_proj", "model.layers.79.mlp.experts.106.up_proj", "model.layers.79.mlp.experts.107.up_proj", "model.layers.79.mlp.experts.108.up_proj", "model.layers.79.mlp.experts.109.up_proj", "model.layers.79.mlp.experts.110.up_proj", "model.layers.79.mlp.experts.111.up_proj", "model.layers.79.mlp.experts.112.up_proj", "model.layers.79.mlp.experts.113.up_proj", "model.layers.79.mlp.experts.114.up_proj", "model.layers.79.mlp.experts.115.up_proj", "model.layers.79.mlp.experts.116.up_proj", "model.layers.79.mlp.experts.117.up_proj", "model.layers.79.mlp.experts.118.up_proj", "model.layers.79.mlp.experts.119.up_proj", "model.layers.79.mlp.experts.120.up_proj", "model.layers.79.mlp.experts.121.up_proj", "model.layers.79.mlp.experts.122.up_proj", "model.layers.79.mlp.experts.123.up_proj", "model.layers.79.mlp.experts.124.up_proj", "model.layers.79.mlp.experts.125.up_proj", "model.layers.79.mlp.experts.126.up_proj", "model.layers.79.mlp.experts.127.up_proj", "model.layers.79.mlp.experts.128.up_proj", "model.layers.79.mlp.experts.129.up_proj", "model.layers.79.mlp.experts.130.up_proj", "model.layers.79.mlp.experts.131.up_proj", "model.layers.79.mlp.experts.132.up_proj", "model.layers.79.mlp.experts.133.up_proj", "model.layers.79.mlp.experts.134.up_proj", "model.layers.79.mlp.experts.135.up_proj", "model.layers.79.mlp.experts.136.up_proj", "model.layers.79.mlp.experts.137.up_proj", "model.layers.79.mlp.experts.138.up_proj", "model.layers.79.mlp.experts.139.up_proj", "model.layers.79.mlp.experts.140.up_proj", "model.layers.79.mlp.experts.141.up_proj", "model.layers.79.mlp.experts.142.up_proj", "model.layers.79.mlp.experts.143.up_proj", "model.layers.79.mlp.experts.144.up_proj", "model.layers.79.mlp.experts.145.up_proj", "model.layers.79.mlp.experts.146.up_proj", "model.layers.79.mlp.experts.147.up_proj", "model.layers.79.mlp.experts.148.up_proj", "model.layers.79.mlp.experts.149.up_proj", "model.layers.79.mlp.experts.150.up_proj", "model.layers.79.mlp.experts.151.up_proj", "model.layers.79.mlp.experts.152.up_proj", "model.layers.79.mlp.experts.153.up_proj", "model.layers.79.mlp.experts.154.up_proj", "model.layers.79.mlp.experts.155.up_proj", "model.layers.79.mlp.experts.156.up_proj", "model.layers.79.mlp.experts.157.up_proj", "model.layers.79.mlp.experts.158.up_proj", "model.layers.79.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.000239477027207613, "dbits": 2516582400 }, { "dkld": -0.0004030163399875164, "dbits": 5033164800 }, { "dkld": -0.00033297184854746975, "dbits": 7549747200 }, { "dkld": -0.0003581544384360369, "dbits": 12582912000 } ] }, { "idx": 476, "layers": [ "model.layers.79.mlp.experts.0.down_proj", "model.layers.79.mlp.experts.1.down_proj", "model.layers.79.mlp.experts.2.down_proj", "model.layers.79.mlp.experts.3.down_proj", "model.layers.79.mlp.experts.4.down_proj", "model.layers.79.mlp.experts.5.down_proj", "model.layers.79.mlp.experts.6.down_proj", "model.layers.79.mlp.experts.7.down_proj", "model.layers.79.mlp.experts.8.down_proj", "model.layers.79.mlp.experts.9.down_proj", "model.layers.79.mlp.experts.10.down_proj", "model.layers.79.mlp.experts.11.down_proj", "model.layers.79.mlp.experts.12.down_proj", "model.layers.79.mlp.experts.13.down_proj", "model.layers.79.mlp.experts.14.down_proj", "model.layers.79.mlp.experts.15.down_proj", "model.layers.79.mlp.experts.16.down_proj", "model.layers.79.mlp.experts.17.down_proj", "model.layers.79.mlp.experts.18.down_proj", "model.layers.79.mlp.experts.19.down_proj", "model.layers.79.mlp.experts.20.down_proj", "model.layers.79.mlp.experts.21.down_proj", "model.layers.79.mlp.experts.22.down_proj", "model.layers.79.mlp.experts.23.down_proj", "model.layers.79.mlp.experts.24.down_proj", "model.layers.79.mlp.experts.25.down_proj", "model.layers.79.mlp.experts.26.down_proj", "model.layers.79.mlp.experts.27.down_proj", "model.layers.79.mlp.experts.28.down_proj", "model.layers.79.mlp.experts.29.down_proj", "model.layers.79.mlp.experts.30.down_proj", "model.layers.79.mlp.experts.31.down_proj", "model.layers.79.mlp.experts.32.down_proj", "model.layers.79.mlp.experts.33.down_proj", "model.layers.79.mlp.experts.34.down_proj", "model.layers.79.mlp.experts.35.down_proj", "model.layers.79.mlp.experts.36.down_proj", "model.layers.79.mlp.experts.37.down_proj", "model.layers.79.mlp.experts.38.down_proj", "model.layers.79.mlp.experts.39.down_proj", "model.layers.79.mlp.experts.40.down_proj", "model.layers.79.mlp.experts.41.down_proj", "model.layers.79.mlp.experts.42.down_proj", "model.layers.79.mlp.experts.43.down_proj", "model.layers.79.mlp.experts.44.down_proj", "model.layers.79.mlp.experts.45.down_proj", "model.layers.79.mlp.experts.46.down_proj", "model.layers.79.mlp.experts.47.down_proj", "model.layers.79.mlp.experts.48.down_proj", "model.layers.79.mlp.experts.49.down_proj", "model.layers.79.mlp.experts.50.down_proj", "model.layers.79.mlp.experts.51.down_proj", "model.layers.79.mlp.experts.52.down_proj", "model.layers.79.mlp.experts.53.down_proj", "model.layers.79.mlp.experts.54.down_proj", "model.layers.79.mlp.experts.55.down_proj", "model.layers.79.mlp.experts.56.down_proj", "model.layers.79.mlp.experts.57.down_proj", "model.layers.79.mlp.experts.58.down_proj", "model.layers.79.mlp.experts.59.down_proj", "model.layers.79.mlp.experts.60.down_proj", "model.layers.79.mlp.experts.61.down_proj", "model.layers.79.mlp.experts.62.down_proj", "model.layers.79.mlp.experts.63.down_proj", "model.layers.79.mlp.experts.64.down_proj", "model.layers.79.mlp.experts.65.down_proj", "model.layers.79.mlp.experts.66.down_proj", "model.layers.79.mlp.experts.67.down_proj", "model.layers.79.mlp.experts.68.down_proj", "model.layers.79.mlp.experts.69.down_proj", "model.layers.79.mlp.experts.70.down_proj", "model.layers.79.mlp.experts.71.down_proj", "model.layers.79.mlp.experts.72.down_proj", "model.layers.79.mlp.experts.73.down_proj", "model.layers.79.mlp.experts.74.down_proj", "model.layers.79.mlp.experts.75.down_proj", "model.layers.79.mlp.experts.76.down_proj", "model.layers.79.mlp.experts.77.down_proj", "model.layers.79.mlp.experts.78.down_proj", "model.layers.79.mlp.experts.79.down_proj", "model.layers.79.mlp.experts.80.down_proj", "model.layers.79.mlp.experts.81.down_proj", "model.layers.79.mlp.experts.82.down_proj", "model.layers.79.mlp.experts.83.down_proj", "model.layers.79.mlp.experts.84.down_proj", "model.layers.79.mlp.experts.85.down_proj", "model.layers.79.mlp.experts.86.down_proj", "model.layers.79.mlp.experts.87.down_proj", "model.layers.79.mlp.experts.88.down_proj", "model.layers.79.mlp.experts.89.down_proj", "model.layers.79.mlp.experts.90.down_proj", "model.layers.79.mlp.experts.91.down_proj", "model.layers.79.mlp.experts.92.down_proj", "model.layers.79.mlp.experts.93.down_proj", "model.layers.79.mlp.experts.94.down_proj", "model.layers.79.mlp.experts.95.down_proj", "model.layers.79.mlp.experts.96.down_proj", "model.layers.79.mlp.experts.97.down_proj", "model.layers.79.mlp.experts.98.down_proj", "model.layers.79.mlp.experts.99.down_proj", "model.layers.79.mlp.experts.100.down_proj", "model.layers.79.mlp.experts.101.down_proj", "model.layers.79.mlp.experts.102.down_proj", "model.layers.79.mlp.experts.103.down_proj", "model.layers.79.mlp.experts.104.down_proj", "model.layers.79.mlp.experts.105.down_proj", "model.layers.79.mlp.experts.106.down_proj", "model.layers.79.mlp.experts.107.down_proj", "model.layers.79.mlp.experts.108.down_proj", "model.layers.79.mlp.experts.109.down_proj", "model.layers.79.mlp.experts.110.down_proj", "model.layers.79.mlp.experts.111.down_proj", "model.layers.79.mlp.experts.112.down_proj", "model.layers.79.mlp.experts.113.down_proj", "model.layers.79.mlp.experts.114.down_proj", "model.layers.79.mlp.experts.115.down_proj", "model.layers.79.mlp.experts.116.down_proj", "model.layers.79.mlp.experts.117.down_proj", "model.layers.79.mlp.experts.118.down_proj", "model.layers.79.mlp.experts.119.down_proj", "model.layers.79.mlp.experts.120.down_proj", "model.layers.79.mlp.experts.121.down_proj", "model.layers.79.mlp.experts.122.down_proj", "model.layers.79.mlp.experts.123.down_proj", "model.layers.79.mlp.experts.124.down_proj", "model.layers.79.mlp.experts.125.down_proj", "model.layers.79.mlp.experts.126.down_proj", "model.layers.79.mlp.experts.127.down_proj", "model.layers.79.mlp.experts.128.down_proj", "model.layers.79.mlp.experts.129.down_proj", "model.layers.79.mlp.experts.130.down_proj", "model.layers.79.mlp.experts.131.down_proj", "model.layers.79.mlp.experts.132.down_proj", "model.layers.79.mlp.experts.133.down_proj", "model.layers.79.mlp.experts.134.down_proj", "model.layers.79.mlp.experts.135.down_proj", "model.layers.79.mlp.experts.136.down_proj", "model.layers.79.mlp.experts.137.down_proj", "model.layers.79.mlp.experts.138.down_proj", "model.layers.79.mlp.experts.139.down_proj", "model.layers.79.mlp.experts.140.down_proj", "model.layers.79.mlp.experts.141.down_proj", "model.layers.79.mlp.experts.142.down_proj", "model.layers.79.mlp.experts.143.down_proj", "model.layers.79.mlp.experts.144.down_proj", "model.layers.79.mlp.experts.145.down_proj", "model.layers.79.mlp.experts.146.down_proj", "model.layers.79.mlp.experts.147.down_proj", "model.layers.79.mlp.experts.148.down_proj", "model.layers.79.mlp.experts.149.down_proj", "model.layers.79.mlp.experts.150.down_proj", "model.layers.79.mlp.experts.151.down_proj", "model.layers.79.mlp.experts.152.down_proj", "model.layers.79.mlp.experts.153.down_proj", "model.layers.79.mlp.experts.154.down_proj", "model.layers.79.mlp.experts.155.down_proj", "model.layers.79.mlp.experts.156.down_proj", "model.layers.79.mlp.experts.157.down_proj", "model.layers.79.mlp.experts.158.down_proj", "model.layers.79.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00014120824635029117, "dbits": 1258291200 }, { "dkld": -9.668767452240823e-05, "dbits": 2516582400 }, { "dkld": -9.852405637503225e-05, "dbits": 3774873600 }, { "dkld": -8.989498019218445e-05, "dbits": 6291456000 } ] }, { "idx": 477, "layers": [ "model.layers.80.self_attn.q_proj" ], "candidates": [ { "dkld": 0.0008333431556820786, "dbits": 62914560 }, { "dkld": 0.0013711316511034882, "dbits": 125829120 }, { "dkld": 0.0010630950331687872, "dbits": 188743680 }, { "dkld": 0.0008273586630821173, "dbits": 314572800 } ] }, { "idx": 478, "layers": [ "model.layers.80.self_attn.k_proj", "model.layers.80.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0013597102835774533, "dbits": 10485760 }, { "dkld": -0.0010667313821613872, "dbits": 20971520 }, { "dkld": -0.0010932916775345802, "dbits": 31457280 }, { "dkld": -0.001180700212717059, "dbits": 52428800 } ] }, { "idx": 479, "layers": [ "model.layers.80.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0004542200826108539, "dbits": 62914560 }, { "dkld": -0.0003838788717985264, "dbits": 125829120 }, { "dkld": -0.0003289486281573828, "dbits": 188743680 }, { "dkld": -0.0003789925947785405, "dbits": 314572800 } ] }, { "idx": 480, "layers": [ "model.layers.80.mlp.shared_experts.gate_proj", "model.layers.80.mlp.shared_experts.up_proj", "model.layers.80.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00019700154662133373, "dbits": 23592960 }, { "dkld": -0.0004557191394269494, "dbits": 47185920 }, { "dkld": -0.000475996918976318, "dbits": 70778880 }, { "dkld": -0.0004392322152853012, "dbits": 117964800 } ] }, { "idx": 481, "layers": [ "model.layers.80.mlp.experts.0.gate_proj", "model.layers.80.mlp.experts.1.gate_proj", "model.layers.80.mlp.experts.2.gate_proj", "model.layers.80.mlp.experts.3.gate_proj", "model.layers.80.mlp.experts.4.gate_proj", "model.layers.80.mlp.experts.5.gate_proj", "model.layers.80.mlp.experts.6.gate_proj", "model.layers.80.mlp.experts.7.gate_proj", "model.layers.80.mlp.experts.8.gate_proj", "model.layers.80.mlp.experts.9.gate_proj", "model.layers.80.mlp.experts.10.gate_proj", "model.layers.80.mlp.experts.11.gate_proj", "model.layers.80.mlp.experts.12.gate_proj", "model.layers.80.mlp.experts.13.gate_proj", "model.layers.80.mlp.experts.14.gate_proj", "model.layers.80.mlp.experts.15.gate_proj", "model.layers.80.mlp.experts.16.gate_proj", "model.layers.80.mlp.experts.17.gate_proj", "model.layers.80.mlp.experts.18.gate_proj", "model.layers.80.mlp.experts.19.gate_proj", "model.layers.80.mlp.experts.20.gate_proj", "model.layers.80.mlp.experts.21.gate_proj", "model.layers.80.mlp.experts.22.gate_proj", "model.layers.80.mlp.experts.23.gate_proj", "model.layers.80.mlp.experts.24.gate_proj", "model.layers.80.mlp.experts.25.gate_proj", "model.layers.80.mlp.experts.26.gate_proj", "model.layers.80.mlp.experts.27.gate_proj", "model.layers.80.mlp.experts.28.gate_proj", "model.layers.80.mlp.experts.29.gate_proj", "model.layers.80.mlp.experts.30.gate_proj", "model.layers.80.mlp.experts.31.gate_proj", "model.layers.80.mlp.experts.32.gate_proj", "model.layers.80.mlp.experts.33.gate_proj", "model.layers.80.mlp.experts.34.gate_proj", "model.layers.80.mlp.experts.35.gate_proj", "model.layers.80.mlp.experts.36.gate_proj", "model.layers.80.mlp.experts.37.gate_proj", "model.layers.80.mlp.experts.38.gate_proj", "model.layers.80.mlp.experts.39.gate_proj", "model.layers.80.mlp.experts.40.gate_proj", "model.layers.80.mlp.experts.41.gate_proj", "model.layers.80.mlp.experts.42.gate_proj", "model.layers.80.mlp.experts.43.gate_proj", "model.layers.80.mlp.experts.44.gate_proj", "model.layers.80.mlp.experts.45.gate_proj", "model.layers.80.mlp.experts.46.gate_proj", "model.layers.80.mlp.experts.47.gate_proj", "model.layers.80.mlp.experts.48.gate_proj", "model.layers.80.mlp.experts.49.gate_proj", "model.layers.80.mlp.experts.50.gate_proj", "model.layers.80.mlp.experts.51.gate_proj", "model.layers.80.mlp.experts.52.gate_proj", "model.layers.80.mlp.experts.53.gate_proj", "model.layers.80.mlp.experts.54.gate_proj", "model.layers.80.mlp.experts.55.gate_proj", "model.layers.80.mlp.experts.56.gate_proj", "model.layers.80.mlp.experts.57.gate_proj", "model.layers.80.mlp.experts.58.gate_proj", "model.layers.80.mlp.experts.59.gate_proj", "model.layers.80.mlp.experts.60.gate_proj", "model.layers.80.mlp.experts.61.gate_proj", "model.layers.80.mlp.experts.62.gate_proj", "model.layers.80.mlp.experts.63.gate_proj", "model.layers.80.mlp.experts.64.gate_proj", "model.layers.80.mlp.experts.65.gate_proj", "model.layers.80.mlp.experts.66.gate_proj", "model.layers.80.mlp.experts.67.gate_proj", "model.layers.80.mlp.experts.68.gate_proj", "model.layers.80.mlp.experts.69.gate_proj", "model.layers.80.mlp.experts.70.gate_proj", "model.layers.80.mlp.experts.71.gate_proj", "model.layers.80.mlp.experts.72.gate_proj", "model.layers.80.mlp.experts.73.gate_proj", "model.layers.80.mlp.experts.74.gate_proj", "model.layers.80.mlp.experts.75.gate_proj", "model.layers.80.mlp.experts.76.gate_proj", "model.layers.80.mlp.experts.77.gate_proj", "model.layers.80.mlp.experts.78.gate_proj", "model.layers.80.mlp.experts.79.gate_proj", "model.layers.80.mlp.experts.80.gate_proj", "model.layers.80.mlp.experts.81.gate_proj", "model.layers.80.mlp.experts.82.gate_proj", "model.layers.80.mlp.experts.83.gate_proj", "model.layers.80.mlp.experts.84.gate_proj", "model.layers.80.mlp.experts.85.gate_proj", "model.layers.80.mlp.experts.86.gate_proj", "model.layers.80.mlp.experts.87.gate_proj", "model.layers.80.mlp.experts.88.gate_proj", "model.layers.80.mlp.experts.89.gate_proj", "model.layers.80.mlp.experts.90.gate_proj", "model.layers.80.mlp.experts.91.gate_proj", "model.layers.80.mlp.experts.92.gate_proj", "model.layers.80.mlp.experts.93.gate_proj", "model.layers.80.mlp.experts.94.gate_proj", "model.layers.80.mlp.experts.95.gate_proj", "model.layers.80.mlp.experts.96.gate_proj", "model.layers.80.mlp.experts.97.gate_proj", "model.layers.80.mlp.experts.98.gate_proj", "model.layers.80.mlp.experts.99.gate_proj", "model.layers.80.mlp.experts.100.gate_proj", "model.layers.80.mlp.experts.101.gate_proj", "model.layers.80.mlp.experts.102.gate_proj", "model.layers.80.mlp.experts.103.gate_proj", "model.layers.80.mlp.experts.104.gate_proj", "model.layers.80.mlp.experts.105.gate_proj", "model.layers.80.mlp.experts.106.gate_proj", "model.layers.80.mlp.experts.107.gate_proj", "model.layers.80.mlp.experts.108.gate_proj", "model.layers.80.mlp.experts.109.gate_proj", "model.layers.80.mlp.experts.110.gate_proj", "model.layers.80.mlp.experts.111.gate_proj", "model.layers.80.mlp.experts.112.gate_proj", "model.layers.80.mlp.experts.113.gate_proj", "model.layers.80.mlp.experts.114.gate_proj", "model.layers.80.mlp.experts.115.gate_proj", "model.layers.80.mlp.experts.116.gate_proj", "model.layers.80.mlp.experts.117.gate_proj", "model.layers.80.mlp.experts.118.gate_proj", "model.layers.80.mlp.experts.119.gate_proj", "model.layers.80.mlp.experts.120.gate_proj", "model.layers.80.mlp.experts.121.gate_proj", "model.layers.80.mlp.experts.122.gate_proj", "model.layers.80.mlp.experts.123.gate_proj", "model.layers.80.mlp.experts.124.gate_proj", "model.layers.80.mlp.experts.125.gate_proj", "model.layers.80.mlp.experts.126.gate_proj", "model.layers.80.mlp.experts.127.gate_proj", "model.layers.80.mlp.experts.128.gate_proj", "model.layers.80.mlp.experts.129.gate_proj", "model.layers.80.mlp.experts.130.gate_proj", "model.layers.80.mlp.experts.131.gate_proj", "model.layers.80.mlp.experts.132.gate_proj", "model.layers.80.mlp.experts.133.gate_proj", "model.layers.80.mlp.experts.134.gate_proj", "model.layers.80.mlp.experts.135.gate_proj", "model.layers.80.mlp.experts.136.gate_proj", "model.layers.80.mlp.experts.137.gate_proj", "model.layers.80.mlp.experts.138.gate_proj", "model.layers.80.mlp.experts.139.gate_proj", "model.layers.80.mlp.experts.140.gate_proj", "model.layers.80.mlp.experts.141.gate_proj", "model.layers.80.mlp.experts.142.gate_proj", "model.layers.80.mlp.experts.143.gate_proj", "model.layers.80.mlp.experts.144.gate_proj", "model.layers.80.mlp.experts.145.gate_proj", "model.layers.80.mlp.experts.146.gate_proj", "model.layers.80.mlp.experts.147.gate_proj", "model.layers.80.mlp.experts.148.gate_proj", "model.layers.80.mlp.experts.149.gate_proj", "model.layers.80.mlp.experts.150.gate_proj", "model.layers.80.mlp.experts.151.gate_proj", "model.layers.80.mlp.experts.152.gate_proj", "model.layers.80.mlp.experts.153.gate_proj", "model.layers.80.mlp.experts.154.gate_proj", "model.layers.80.mlp.experts.155.gate_proj", "model.layers.80.mlp.experts.156.gate_proj", "model.layers.80.mlp.experts.157.gate_proj", "model.layers.80.mlp.experts.158.gate_proj", "model.layers.80.mlp.experts.159.gate_proj", "model.layers.80.mlp.experts.0.up_proj", "model.layers.80.mlp.experts.1.up_proj", "model.layers.80.mlp.experts.2.up_proj", "model.layers.80.mlp.experts.3.up_proj", "model.layers.80.mlp.experts.4.up_proj", "model.layers.80.mlp.experts.5.up_proj", "model.layers.80.mlp.experts.6.up_proj", "model.layers.80.mlp.experts.7.up_proj", "model.layers.80.mlp.experts.8.up_proj", "model.layers.80.mlp.experts.9.up_proj", "model.layers.80.mlp.experts.10.up_proj", "model.layers.80.mlp.experts.11.up_proj", "model.layers.80.mlp.experts.12.up_proj", "model.layers.80.mlp.experts.13.up_proj", "model.layers.80.mlp.experts.14.up_proj", "model.layers.80.mlp.experts.15.up_proj", "model.layers.80.mlp.experts.16.up_proj", "model.layers.80.mlp.experts.17.up_proj", "model.layers.80.mlp.experts.18.up_proj", "model.layers.80.mlp.experts.19.up_proj", "model.layers.80.mlp.experts.20.up_proj", "model.layers.80.mlp.experts.21.up_proj", "model.layers.80.mlp.experts.22.up_proj", "model.layers.80.mlp.experts.23.up_proj", "model.layers.80.mlp.experts.24.up_proj", "model.layers.80.mlp.experts.25.up_proj", "model.layers.80.mlp.experts.26.up_proj", "model.layers.80.mlp.experts.27.up_proj", "model.layers.80.mlp.experts.28.up_proj", "model.layers.80.mlp.experts.29.up_proj", "model.layers.80.mlp.experts.30.up_proj", "model.layers.80.mlp.experts.31.up_proj", "model.layers.80.mlp.experts.32.up_proj", "model.layers.80.mlp.experts.33.up_proj", "model.layers.80.mlp.experts.34.up_proj", "model.layers.80.mlp.experts.35.up_proj", "model.layers.80.mlp.experts.36.up_proj", "model.layers.80.mlp.experts.37.up_proj", "model.layers.80.mlp.experts.38.up_proj", "model.layers.80.mlp.experts.39.up_proj", "model.layers.80.mlp.experts.40.up_proj", "model.layers.80.mlp.experts.41.up_proj", "model.layers.80.mlp.experts.42.up_proj", "model.layers.80.mlp.experts.43.up_proj", "model.layers.80.mlp.experts.44.up_proj", "model.layers.80.mlp.experts.45.up_proj", "model.layers.80.mlp.experts.46.up_proj", "model.layers.80.mlp.experts.47.up_proj", "model.layers.80.mlp.experts.48.up_proj", "model.layers.80.mlp.experts.49.up_proj", "model.layers.80.mlp.experts.50.up_proj", "model.layers.80.mlp.experts.51.up_proj", "model.layers.80.mlp.experts.52.up_proj", "model.layers.80.mlp.experts.53.up_proj", "model.layers.80.mlp.experts.54.up_proj", "model.layers.80.mlp.experts.55.up_proj", "model.layers.80.mlp.experts.56.up_proj", "model.layers.80.mlp.experts.57.up_proj", "model.layers.80.mlp.experts.58.up_proj", "model.layers.80.mlp.experts.59.up_proj", "model.layers.80.mlp.experts.60.up_proj", "model.layers.80.mlp.experts.61.up_proj", "model.layers.80.mlp.experts.62.up_proj", "model.layers.80.mlp.experts.63.up_proj", "model.layers.80.mlp.experts.64.up_proj", "model.layers.80.mlp.experts.65.up_proj", "model.layers.80.mlp.experts.66.up_proj", "model.layers.80.mlp.experts.67.up_proj", "model.layers.80.mlp.experts.68.up_proj", "model.layers.80.mlp.experts.69.up_proj", "model.layers.80.mlp.experts.70.up_proj", "model.layers.80.mlp.experts.71.up_proj", "model.layers.80.mlp.experts.72.up_proj", "model.layers.80.mlp.experts.73.up_proj", "model.layers.80.mlp.experts.74.up_proj", "model.layers.80.mlp.experts.75.up_proj", "model.layers.80.mlp.experts.76.up_proj", "model.layers.80.mlp.experts.77.up_proj", "model.layers.80.mlp.experts.78.up_proj", "model.layers.80.mlp.experts.79.up_proj", "model.layers.80.mlp.experts.80.up_proj", "model.layers.80.mlp.experts.81.up_proj", "model.layers.80.mlp.experts.82.up_proj", "model.layers.80.mlp.experts.83.up_proj", "model.layers.80.mlp.experts.84.up_proj", "model.layers.80.mlp.experts.85.up_proj", "model.layers.80.mlp.experts.86.up_proj", "model.layers.80.mlp.experts.87.up_proj", "model.layers.80.mlp.experts.88.up_proj", "model.layers.80.mlp.experts.89.up_proj", "model.layers.80.mlp.experts.90.up_proj", "model.layers.80.mlp.experts.91.up_proj", "model.layers.80.mlp.experts.92.up_proj", "model.layers.80.mlp.experts.93.up_proj", "model.layers.80.mlp.experts.94.up_proj", "model.layers.80.mlp.experts.95.up_proj", "model.layers.80.mlp.experts.96.up_proj", "model.layers.80.mlp.experts.97.up_proj", "model.layers.80.mlp.experts.98.up_proj", "model.layers.80.mlp.experts.99.up_proj", "model.layers.80.mlp.experts.100.up_proj", "model.layers.80.mlp.experts.101.up_proj", "model.layers.80.mlp.experts.102.up_proj", "model.layers.80.mlp.experts.103.up_proj", "model.layers.80.mlp.experts.104.up_proj", "model.layers.80.mlp.experts.105.up_proj", "model.layers.80.mlp.experts.106.up_proj", "model.layers.80.mlp.experts.107.up_proj", "model.layers.80.mlp.experts.108.up_proj", "model.layers.80.mlp.experts.109.up_proj", "model.layers.80.mlp.experts.110.up_proj", "model.layers.80.mlp.experts.111.up_proj", "model.layers.80.mlp.experts.112.up_proj", "model.layers.80.mlp.experts.113.up_proj", "model.layers.80.mlp.experts.114.up_proj", "model.layers.80.mlp.experts.115.up_proj", "model.layers.80.mlp.experts.116.up_proj", "model.layers.80.mlp.experts.117.up_proj", "model.layers.80.mlp.experts.118.up_proj", "model.layers.80.mlp.experts.119.up_proj", "model.layers.80.mlp.experts.120.up_proj", "model.layers.80.mlp.experts.121.up_proj", "model.layers.80.mlp.experts.122.up_proj", "model.layers.80.mlp.experts.123.up_proj", "model.layers.80.mlp.experts.124.up_proj", "model.layers.80.mlp.experts.125.up_proj", "model.layers.80.mlp.experts.126.up_proj", "model.layers.80.mlp.experts.127.up_proj", "model.layers.80.mlp.experts.128.up_proj", "model.layers.80.mlp.experts.129.up_proj", "model.layers.80.mlp.experts.130.up_proj", "model.layers.80.mlp.experts.131.up_proj", "model.layers.80.mlp.experts.132.up_proj", "model.layers.80.mlp.experts.133.up_proj", "model.layers.80.mlp.experts.134.up_proj", "model.layers.80.mlp.experts.135.up_proj", "model.layers.80.mlp.experts.136.up_proj", "model.layers.80.mlp.experts.137.up_proj", "model.layers.80.mlp.experts.138.up_proj", "model.layers.80.mlp.experts.139.up_proj", "model.layers.80.mlp.experts.140.up_proj", "model.layers.80.mlp.experts.141.up_proj", "model.layers.80.mlp.experts.142.up_proj", "model.layers.80.mlp.experts.143.up_proj", "model.layers.80.mlp.experts.144.up_proj", "model.layers.80.mlp.experts.145.up_proj", "model.layers.80.mlp.experts.146.up_proj", "model.layers.80.mlp.experts.147.up_proj", "model.layers.80.mlp.experts.148.up_proj", "model.layers.80.mlp.experts.149.up_proj", "model.layers.80.mlp.experts.150.up_proj", "model.layers.80.mlp.experts.151.up_proj", "model.layers.80.mlp.experts.152.up_proj", "model.layers.80.mlp.experts.153.up_proj", "model.layers.80.mlp.experts.154.up_proj", "model.layers.80.mlp.experts.155.up_proj", "model.layers.80.mlp.experts.156.up_proj", "model.layers.80.mlp.experts.157.up_proj", "model.layers.80.mlp.experts.158.up_proj", "model.layers.80.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0001028042286634473, "dbits": 2516582400 }, { "dkld": -7.842537015677331e-05, "dbits": 5033164800 }, { "dkld": -9.278375655413784e-05, "dbits": 7549747200 }, { "dkld": -9.133964776993075e-05, "dbits": 12582912000 } ] }, { "idx": 482, "layers": [ "model.layers.80.mlp.experts.0.down_proj", "model.layers.80.mlp.experts.1.down_proj", "model.layers.80.mlp.experts.2.down_proj", "model.layers.80.mlp.experts.3.down_proj", "model.layers.80.mlp.experts.4.down_proj", "model.layers.80.mlp.experts.5.down_proj", "model.layers.80.mlp.experts.6.down_proj", "model.layers.80.mlp.experts.7.down_proj", "model.layers.80.mlp.experts.8.down_proj", "model.layers.80.mlp.experts.9.down_proj", "model.layers.80.mlp.experts.10.down_proj", "model.layers.80.mlp.experts.11.down_proj", "model.layers.80.mlp.experts.12.down_proj", "model.layers.80.mlp.experts.13.down_proj", "model.layers.80.mlp.experts.14.down_proj", "model.layers.80.mlp.experts.15.down_proj", "model.layers.80.mlp.experts.16.down_proj", "model.layers.80.mlp.experts.17.down_proj", "model.layers.80.mlp.experts.18.down_proj", "model.layers.80.mlp.experts.19.down_proj", "model.layers.80.mlp.experts.20.down_proj", "model.layers.80.mlp.experts.21.down_proj", "model.layers.80.mlp.experts.22.down_proj", "model.layers.80.mlp.experts.23.down_proj", "model.layers.80.mlp.experts.24.down_proj", "model.layers.80.mlp.experts.25.down_proj", "model.layers.80.mlp.experts.26.down_proj", "model.layers.80.mlp.experts.27.down_proj", "model.layers.80.mlp.experts.28.down_proj", "model.layers.80.mlp.experts.29.down_proj", "model.layers.80.mlp.experts.30.down_proj", "model.layers.80.mlp.experts.31.down_proj", "model.layers.80.mlp.experts.32.down_proj", "model.layers.80.mlp.experts.33.down_proj", "model.layers.80.mlp.experts.34.down_proj", "model.layers.80.mlp.experts.35.down_proj", "model.layers.80.mlp.experts.36.down_proj", "model.layers.80.mlp.experts.37.down_proj", "model.layers.80.mlp.experts.38.down_proj", "model.layers.80.mlp.experts.39.down_proj", "model.layers.80.mlp.experts.40.down_proj", "model.layers.80.mlp.experts.41.down_proj", "model.layers.80.mlp.experts.42.down_proj", "model.layers.80.mlp.experts.43.down_proj", "model.layers.80.mlp.experts.44.down_proj", "model.layers.80.mlp.experts.45.down_proj", "model.layers.80.mlp.experts.46.down_proj", "model.layers.80.mlp.experts.47.down_proj", "model.layers.80.mlp.experts.48.down_proj", "model.layers.80.mlp.experts.49.down_proj", "model.layers.80.mlp.experts.50.down_proj", "model.layers.80.mlp.experts.51.down_proj", "model.layers.80.mlp.experts.52.down_proj", "model.layers.80.mlp.experts.53.down_proj", "model.layers.80.mlp.experts.54.down_proj", "model.layers.80.mlp.experts.55.down_proj", "model.layers.80.mlp.experts.56.down_proj", "model.layers.80.mlp.experts.57.down_proj", "model.layers.80.mlp.experts.58.down_proj", "model.layers.80.mlp.experts.59.down_proj", "model.layers.80.mlp.experts.60.down_proj", "model.layers.80.mlp.experts.61.down_proj", "model.layers.80.mlp.experts.62.down_proj", "model.layers.80.mlp.experts.63.down_proj", "model.layers.80.mlp.experts.64.down_proj", "model.layers.80.mlp.experts.65.down_proj", "model.layers.80.mlp.experts.66.down_proj", "model.layers.80.mlp.experts.67.down_proj", "model.layers.80.mlp.experts.68.down_proj", "model.layers.80.mlp.experts.69.down_proj", "model.layers.80.mlp.experts.70.down_proj", "model.layers.80.mlp.experts.71.down_proj", "model.layers.80.mlp.experts.72.down_proj", "model.layers.80.mlp.experts.73.down_proj", "model.layers.80.mlp.experts.74.down_proj", "model.layers.80.mlp.experts.75.down_proj", "model.layers.80.mlp.experts.76.down_proj", "model.layers.80.mlp.experts.77.down_proj", "model.layers.80.mlp.experts.78.down_proj", "model.layers.80.mlp.experts.79.down_proj", "model.layers.80.mlp.experts.80.down_proj", "model.layers.80.mlp.experts.81.down_proj", "model.layers.80.mlp.experts.82.down_proj", "model.layers.80.mlp.experts.83.down_proj", "model.layers.80.mlp.experts.84.down_proj", "model.layers.80.mlp.experts.85.down_proj", "model.layers.80.mlp.experts.86.down_proj", "model.layers.80.mlp.experts.87.down_proj", "model.layers.80.mlp.experts.88.down_proj", "model.layers.80.mlp.experts.89.down_proj", "model.layers.80.mlp.experts.90.down_proj", "model.layers.80.mlp.experts.91.down_proj", "model.layers.80.mlp.experts.92.down_proj", "model.layers.80.mlp.experts.93.down_proj", "model.layers.80.mlp.experts.94.down_proj", "model.layers.80.mlp.experts.95.down_proj", "model.layers.80.mlp.experts.96.down_proj", "model.layers.80.mlp.experts.97.down_proj", "model.layers.80.mlp.experts.98.down_proj", "model.layers.80.mlp.experts.99.down_proj", "model.layers.80.mlp.experts.100.down_proj", "model.layers.80.mlp.experts.101.down_proj", "model.layers.80.mlp.experts.102.down_proj", "model.layers.80.mlp.experts.103.down_proj", "model.layers.80.mlp.experts.104.down_proj", "model.layers.80.mlp.experts.105.down_proj", "model.layers.80.mlp.experts.106.down_proj", "model.layers.80.mlp.experts.107.down_proj", "model.layers.80.mlp.experts.108.down_proj", "model.layers.80.mlp.experts.109.down_proj", "model.layers.80.mlp.experts.110.down_proj", "model.layers.80.mlp.experts.111.down_proj", "model.layers.80.mlp.experts.112.down_proj", "model.layers.80.mlp.experts.113.down_proj", "model.layers.80.mlp.experts.114.down_proj", "model.layers.80.mlp.experts.115.down_proj", "model.layers.80.mlp.experts.116.down_proj", "model.layers.80.mlp.experts.117.down_proj", "model.layers.80.mlp.experts.118.down_proj", "model.layers.80.mlp.experts.119.down_proj", "model.layers.80.mlp.experts.120.down_proj", "model.layers.80.mlp.experts.121.down_proj", "model.layers.80.mlp.experts.122.down_proj", "model.layers.80.mlp.experts.123.down_proj", "model.layers.80.mlp.experts.124.down_proj", "model.layers.80.mlp.experts.125.down_proj", "model.layers.80.mlp.experts.126.down_proj", "model.layers.80.mlp.experts.127.down_proj", "model.layers.80.mlp.experts.128.down_proj", "model.layers.80.mlp.experts.129.down_proj", "model.layers.80.mlp.experts.130.down_proj", "model.layers.80.mlp.experts.131.down_proj", "model.layers.80.mlp.experts.132.down_proj", "model.layers.80.mlp.experts.133.down_proj", "model.layers.80.mlp.experts.134.down_proj", "model.layers.80.mlp.experts.135.down_proj", "model.layers.80.mlp.experts.136.down_proj", "model.layers.80.mlp.experts.137.down_proj", "model.layers.80.mlp.experts.138.down_proj", "model.layers.80.mlp.experts.139.down_proj", "model.layers.80.mlp.experts.140.down_proj", "model.layers.80.mlp.experts.141.down_proj", "model.layers.80.mlp.experts.142.down_proj", "model.layers.80.mlp.experts.143.down_proj", "model.layers.80.mlp.experts.144.down_proj", "model.layers.80.mlp.experts.145.down_proj", "model.layers.80.mlp.experts.146.down_proj", "model.layers.80.mlp.experts.147.down_proj", "model.layers.80.mlp.experts.148.down_proj", "model.layers.80.mlp.experts.149.down_proj", "model.layers.80.mlp.experts.150.down_proj", "model.layers.80.mlp.experts.151.down_proj", "model.layers.80.mlp.experts.152.down_proj", "model.layers.80.mlp.experts.153.down_proj", "model.layers.80.mlp.experts.154.down_proj", "model.layers.80.mlp.experts.155.down_proj", "model.layers.80.mlp.experts.156.down_proj", "model.layers.80.mlp.experts.157.down_proj", "model.layers.80.mlp.experts.158.down_proj", "model.layers.80.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00014993883669377206, "dbits": 1258291200 }, { "dkld": -0.00012743081897498287, "dbits": 2516582400 }, { "dkld": -0.0001209394074976472, "dbits": 3774873600 }, { "dkld": -0.00015547797083854953, "dbits": 6291456000 } ] }, { "idx": 483, "layers": [ "model.layers.81.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0002701452001929311, "dbits": 62914560 }, { "dkld": -9.228987619280815e-05, "dbits": 125829120 }, { "dkld": -9.109126403927803e-05, "dbits": 188743680 }, { "dkld": -0.00015569720417261956, "dbits": 314572800 } ] }, { "idx": 484, "layers": [ "model.layers.81.self_attn.k_proj", "model.layers.81.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0009698407724499786, "dbits": 10485760 }, { "dkld": -0.0010417157784104458, "dbits": 20971520 }, { "dkld": -0.0009599914774298668, "dbits": 31457280 }, { "dkld": -0.0009571466594934519, "dbits": 52428800 } ] }, { "idx": 485, "layers": [ "model.layers.81.self_attn.o_proj" ], "candidates": [ { "dkld": -0.000483322981745013, "dbits": 62914560 }, { "dkld": -0.0004746553488075844, "dbits": 125829120 }, { "dkld": -0.00043863831087948, "dbits": 188743680 }, { "dkld": -0.00047108931466938175, "dbits": 314572800 } ] }, { "idx": 486, "layers": [ "model.layers.81.mlp.shared_experts.gate_proj", "model.layers.81.mlp.shared_experts.up_proj", "model.layers.81.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00020104320719838142, "dbits": 23592960 }, { "dkld": -0.00036753118038178323, "dbits": 47185920 }, { "dkld": -0.00039987955242395956, "dbits": 70778880 }, { "dkld": -0.00041093248873949606, "dbits": 117964800 } ] }, { "idx": 487, "layers": [ "model.layers.81.mlp.experts.0.gate_proj", "model.layers.81.mlp.experts.1.gate_proj", "model.layers.81.mlp.experts.2.gate_proj", "model.layers.81.mlp.experts.3.gate_proj", "model.layers.81.mlp.experts.4.gate_proj", "model.layers.81.mlp.experts.5.gate_proj", "model.layers.81.mlp.experts.6.gate_proj", "model.layers.81.mlp.experts.7.gate_proj", "model.layers.81.mlp.experts.8.gate_proj", "model.layers.81.mlp.experts.9.gate_proj", "model.layers.81.mlp.experts.10.gate_proj", "model.layers.81.mlp.experts.11.gate_proj", "model.layers.81.mlp.experts.12.gate_proj", "model.layers.81.mlp.experts.13.gate_proj", "model.layers.81.mlp.experts.14.gate_proj", "model.layers.81.mlp.experts.15.gate_proj", "model.layers.81.mlp.experts.16.gate_proj", "model.layers.81.mlp.experts.17.gate_proj", "model.layers.81.mlp.experts.18.gate_proj", "model.layers.81.mlp.experts.19.gate_proj", "model.layers.81.mlp.experts.20.gate_proj", "model.layers.81.mlp.experts.21.gate_proj", "model.layers.81.mlp.experts.22.gate_proj", "model.layers.81.mlp.experts.23.gate_proj", "model.layers.81.mlp.experts.24.gate_proj", "model.layers.81.mlp.experts.25.gate_proj", "model.layers.81.mlp.experts.26.gate_proj", "model.layers.81.mlp.experts.27.gate_proj", "model.layers.81.mlp.experts.28.gate_proj", "model.layers.81.mlp.experts.29.gate_proj", "model.layers.81.mlp.experts.30.gate_proj", "model.layers.81.mlp.experts.31.gate_proj", "model.layers.81.mlp.experts.32.gate_proj", "model.layers.81.mlp.experts.33.gate_proj", "model.layers.81.mlp.experts.34.gate_proj", "model.layers.81.mlp.experts.35.gate_proj", "model.layers.81.mlp.experts.36.gate_proj", "model.layers.81.mlp.experts.37.gate_proj", "model.layers.81.mlp.experts.38.gate_proj", "model.layers.81.mlp.experts.39.gate_proj", "model.layers.81.mlp.experts.40.gate_proj", "model.layers.81.mlp.experts.41.gate_proj", "model.layers.81.mlp.experts.42.gate_proj", "model.layers.81.mlp.experts.43.gate_proj", "model.layers.81.mlp.experts.44.gate_proj", "model.layers.81.mlp.experts.45.gate_proj", "model.layers.81.mlp.experts.46.gate_proj", "model.layers.81.mlp.experts.47.gate_proj", "model.layers.81.mlp.experts.48.gate_proj", "model.layers.81.mlp.experts.49.gate_proj", "model.layers.81.mlp.experts.50.gate_proj", "model.layers.81.mlp.experts.51.gate_proj", "model.layers.81.mlp.experts.52.gate_proj", "model.layers.81.mlp.experts.53.gate_proj", "model.layers.81.mlp.experts.54.gate_proj", "model.layers.81.mlp.experts.55.gate_proj", "model.layers.81.mlp.experts.56.gate_proj", "model.layers.81.mlp.experts.57.gate_proj", "model.layers.81.mlp.experts.58.gate_proj", "model.layers.81.mlp.experts.59.gate_proj", "model.layers.81.mlp.experts.60.gate_proj", "model.layers.81.mlp.experts.61.gate_proj", "model.layers.81.mlp.experts.62.gate_proj", "model.layers.81.mlp.experts.63.gate_proj", "model.layers.81.mlp.experts.64.gate_proj", "model.layers.81.mlp.experts.65.gate_proj", "model.layers.81.mlp.experts.66.gate_proj", "model.layers.81.mlp.experts.67.gate_proj", "model.layers.81.mlp.experts.68.gate_proj", "model.layers.81.mlp.experts.69.gate_proj", "model.layers.81.mlp.experts.70.gate_proj", "model.layers.81.mlp.experts.71.gate_proj", "model.layers.81.mlp.experts.72.gate_proj", "model.layers.81.mlp.experts.73.gate_proj", "model.layers.81.mlp.experts.74.gate_proj", "model.layers.81.mlp.experts.75.gate_proj", "model.layers.81.mlp.experts.76.gate_proj", "model.layers.81.mlp.experts.77.gate_proj", "model.layers.81.mlp.experts.78.gate_proj", "model.layers.81.mlp.experts.79.gate_proj", "model.layers.81.mlp.experts.80.gate_proj", "model.layers.81.mlp.experts.81.gate_proj", "model.layers.81.mlp.experts.82.gate_proj", "model.layers.81.mlp.experts.83.gate_proj", "model.layers.81.mlp.experts.84.gate_proj", "model.layers.81.mlp.experts.85.gate_proj", "model.layers.81.mlp.experts.86.gate_proj", "model.layers.81.mlp.experts.87.gate_proj", "model.layers.81.mlp.experts.88.gate_proj", "model.layers.81.mlp.experts.89.gate_proj", "model.layers.81.mlp.experts.90.gate_proj", "model.layers.81.mlp.experts.91.gate_proj", "model.layers.81.mlp.experts.92.gate_proj", "model.layers.81.mlp.experts.93.gate_proj", "model.layers.81.mlp.experts.94.gate_proj", "model.layers.81.mlp.experts.95.gate_proj", "model.layers.81.mlp.experts.96.gate_proj", "model.layers.81.mlp.experts.97.gate_proj", "model.layers.81.mlp.experts.98.gate_proj", "model.layers.81.mlp.experts.99.gate_proj", "model.layers.81.mlp.experts.100.gate_proj", "model.layers.81.mlp.experts.101.gate_proj", "model.layers.81.mlp.experts.102.gate_proj", "model.layers.81.mlp.experts.103.gate_proj", "model.layers.81.mlp.experts.104.gate_proj", "model.layers.81.mlp.experts.105.gate_proj", "model.layers.81.mlp.experts.106.gate_proj", "model.layers.81.mlp.experts.107.gate_proj", "model.layers.81.mlp.experts.108.gate_proj", "model.layers.81.mlp.experts.109.gate_proj", "model.layers.81.mlp.experts.110.gate_proj", "model.layers.81.mlp.experts.111.gate_proj", "model.layers.81.mlp.experts.112.gate_proj", "model.layers.81.mlp.experts.113.gate_proj", "model.layers.81.mlp.experts.114.gate_proj", "model.layers.81.mlp.experts.115.gate_proj", "model.layers.81.mlp.experts.116.gate_proj", "model.layers.81.mlp.experts.117.gate_proj", "model.layers.81.mlp.experts.118.gate_proj", "model.layers.81.mlp.experts.119.gate_proj", "model.layers.81.mlp.experts.120.gate_proj", "model.layers.81.mlp.experts.121.gate_proj", "model.layers.81.mlp.experts.122.gate_proj", "model.layers.81.mlp.experts.123.gate_proj", "model.layers.81.mlp.experts.124.gate_proj", "model.layers.81.mlp.experts.125.gate_proj", "model.layers.81.mlp.experts.126.gate_proj", "model.layers.81.mlp.experts.127.gate_proj", "model.layers.81.mlp.experts.128.gate_proj", "model.layers.81.mlp.experts.129.gate_proj", "model.layers.81.mlp.experts.130.gate_proj", "model.layers.81.mlp.experts.131.gate_proj", "model.layers.81.mlp.experts.132.gate_proj", "model.layers.81.mlp.experts.133.gate_proj", "model.layers.81.mlp.experts.134.gate_proj", "model.layers.81.mlp.experts.135.gate_proj", "model.layers.81.mlp.experts.136.gate_proj", "model.layers.81.mlp.experts.137.gate_proj", "model.layers.81.mlp.experts.138.gate_proj", "model.layers.81.mlp.experts.139.gate_proj", "model.layers.81.mlp.experts.140.gate_proj", "model.layers.81.mlp.experts.141.gate_proj", "model.layers.81.mlp.experts.142.gate_proj", "model.layers.81.mlp.experts.143.gate_proj", "model.layers.81.mlp.experts.144.gate_proj", "model.layers.81.mlp.experts.145.gate_proj", "model.layers.81.mlp.experts.146.gate_proj", "model.layers.81.mlp.experts.147.gate_proj", "model.layers.81.mlp.experts.148.gate_proj", "model.layers.81.mlp.experts.149.gate_proj", "model.layers.81.mlp.experts.150.gate_proj", "model.layers.81.mlp.experts.151.gate_proj", "model.layers.81.mlp.experts.152.gate_proj", "model.layers.81.mlp.experts.153.gate_proj", "model.layers.81.mlp.experts.154.gate_proj", "model.layers.81.mlp.experts.155.gate_proj", "model.layers.81.mlp.experts.156.gate_proj", "model.layers.81.mlp.experts.157.gate_proj", "model.layers.81.mlp.experts.158.gate_proj", "model.layers.81.mlp.experts.159.gate_proj", "model.layers.81.mlp.experts.0.up_proj", "model.layers.81.mlp.experts.1.up_proj", "model.layers.81.mlp.experts.2.up_proj", "model.layers.81.mlp.experts.3.up_proj", "model.layers.81.mlp.experts.4.up_proj", "model.layers.81.mlp.experts.5.up_proj", "model.layers.81.mlp.experts.6.up_proj", "model.layers.81.mlp.experts.7.up_proj", "model.layers.81.mlp.experts.8.up_proj", "model.layers.81.mlp.experts.9.up_proj", "model.layers.81.mlp.experts.10.up_proj", "model.layers.81.mlp.experts.11.up_proj", "model.layers.81.mlp.experts.12.up_proj", "model.layers.81.mlp.experts.13.up_proj", "model.layers.81.mlp.experts.14.up_proj", "model.layers.81.mlp.experts.15.up_proj", "model.layers.81.mlp.experts.16.up_proj", "model.layers.81.mlp.experts.17.up_proj", "model.layers.81.mlp.experts.18.up_proj", "model.layers.81.mlp.experts.19.up_proj", "model.layers.81.mlp.experts.20.up_proj", "model.layers.81.mlp.experts.21.up_proj", "model.layers.81.mlp.experts.22.up_proj", "model.layers.81.mlp.experts.23.up_proj", "model.layers.81.mlp.experts.24.up_proj", "model.layers.81.mlp.experts.25.up_proj", "model.layers.81.mlp.experts.26.up_proj", "model.layers.81.mlp.experts.27.up_proj", "model.layers.81.mlp.experts.28.up_proj", "model.layers.81.mlp.experts.29.up_proj", "model.layers.81.mlp.experts.30.up_proj", "model.layers.81.mlp.experts.31.up_proj", "model.layers.81.mlp.experts.32.up_proj", "model.layers.81.mlp.experts.33.up_proj", "model.layers.81.mlp.experts.34.up_proj", "model.layers.81.mlp.experts.35.up_proj", "model.layers.81.mlp.experts.36.up_proj", "model.layers.81.mlp.experts.37.up_proj", "model.layers.81.mlp.experts.38.up_proj", "model.layers.81.mlp.experts.39.up_proj", "model.layers.81.mlp.experts.40.up_proj", "model.layers.81.mlp.experts.41.up_proj", "model.layers.81.mlp.experts.42.up_proj", "model.layers.81.mlp.experts.43.up_proj", "model.layers.81.mlp.experts.44.up_proj", "model.layers.81.mlp.experts.45.up_proj", "model.layers.81.mlp.experts.46.up_proj", "model.layers.81.mlp.experts.47.up_proj", "model.layers.81.mlp.experts.48.up_proj", "model.layers.81.mlp.experts.49.up_proj", "model.layers.81.mlp.experts.50.up_proj", "model.layers.81.mlp.experts.51.up_proj", "model.layers.81.mlp.experts.52.up_proj", "model.layers.81.mlp.experts.53.up_proj", "model.layers.81.mlp.experts.54.up_proj", "model.layers.81.mlp.experts.55.up_proj", "model.layers.81.mlp.experts.56.up_proj", "model.layers.81.mlp.experts.57.up_proj", "model.layers.81.mlp.experts.58.up_proj", "model.layers.81.mlp.experts.59.up_proj", "model.layers.81.mlp.experts.60.up_proj", "model.layers.81.mlp.experts.61.up_proj", "model.layers.81.mlp.experts.62.up_proj", "model.layers.81.mlp.experts.63.up_proj", "model.layers.81.mlp.experts.64.up_proj", "model.layers.81.mlp.experts.65.up_proj", "model.layers.81.mlp.experts.66.up_proj", "model.layers.81.mlp.experts.67.up_proj", "model.layers.81.mlp.experts.68.up_proj", "model.layers.81.mlp.experts.69.up_proj", "model.layers.81.mlp.experts.70.up_proj", "model.layers.81.mlp.experts.71.up_proj", "model.layers.81.mlp.experts.72.up_proj", "model.layers.81.mlp.experts.73.up_proj", "model.layers.81.mlp.experts.74.up_proj", "model.layers.81.mlp.experts.75.up_proj", "model.layers.81.mlp.experts.76.up_proj", "model.layers.81.mlp.experts.77.up_proj", "model.layers.81.mlp.experts.78.up_proj", "model.layers.81.mlp.experts.79.up_proj", "model.layers.81.mlp.experts.80.up_proj", "model.layers.81.mlp.experts.81.up_proj", "model.layers.81.mlp.experts.82.up_proj", "model.layers.81.mlp.experts.83.up_proj", "model.layers.81.mlp.experts.84.up_proj", "model.layers.81.mlp.experts.85.up_proj", "model.layers.81.mlp.experts.86.up_proj", "model.layers.81.mlp.experts.87.up_proj", "model.layers.81.mlp.experts.88.up_proj", "model.layers.81.mlp.experts.89.up_proj", "model.layers.81.mlp.experts.90.up_proj", "model.layers.81.mlp.experts.91.up_proj", "model.layers.81.mlp.experts.92.up_proj", "model.layers.81.mlp.experts.93.up_proj", "model.layers.81.mlp.experts.94.up_proj", "model.layers.81.mlp.experts.95.up_proj", "model.layers.81.mlp.experts.96.up_proj", "model.layers.81.mlp.experts.97.up_proj", "model.layers.81.mlp.experts.98.up_proj", "model.layers.81.mlp.experts.99.up_proj", "model.layers.81.mlp.experts.100.up_proj", "model.layers.81.mlp.experts.101.up_proj", "model.layers.81.mlp.experts.102.up_proj", "model.layers.81.mlp.experts.103.up_proj", "model.layers.81.mlp.experts.104.up_proj", "model.layers.81.mlp.experts.105.up_proj", "model.layers.81.mlp.experts.106.up_proj", "model.layers.81.mlp.experts.107.up_proj", "model.layers.81.mlp.experts.108.up_proj", "model.layers.81.mlp.experts.109.up_proj", "model.layers.81.mlp.experts.110.up_proj", "model.layers.81.mlp.experts.111.up_proj", "model.layers.81.mlp.experts.112.up_proj", "model.layers.81.mlp.experts.113.up_proj", "model.layers.81.mlp.experts.114.up_proj", "model.layers.81.mlp.experts.115.up_proj", "model.layers.81.mlp.experts.116.up_proj", "model.layers.81.mlp.experts.117.up_proj", "model.layers.81.mlp.experts.118.up_proj", "model.layers.81.mlp.experts.119.up_proj", "model.layers.81.mlp.experts.120.up_proj", "model.layers.81.mlp.experts.121.up_proj", "model.layers.81.mlp.experts.122.up_proj", "model.layers.81.mlp.experts.123.up_proj", "model.layers.81.mlp.experts.124.up_proj", "model.layers.81.mlp.experts.125.up_proj", "model.layers.81.mlp.experts.126.up_proj", "model.layers.81.mlp.experts.127.up_proj", "model.layers.81.mlp.experts.128.up_proj", "model.layers.81.mlp.experts.129.up_proj", "model.layers.81.mlp.experts.130.up_proj", "model.layers.81.mlp.experts.131.up_proj", "model.layers.81.mlp.experts.132.up_proj", "model.layers.81.mlp.experts.133.up_proj", "model.layers.81.mlp.experts.134.up_proj", "model.layers.81.mlp.experts.135.up_proj", "model.layers.81.mlp.experts.136.up_proj", "model.layers.81.mlp.experts.137.up_proj", "model.layers.81.mlp.experts.138.up_proj", "model.layers.81.mlp.experts.139.up_proj", "model.layers.81.mlp.experts.140.up_proj", "model.layers.81.mlp.experts.141.up_proj", "model.layers.81.mlp.experts.142.up_proj", "model.layers.81.mlp.experts.143.up_proj", "model.layers.81.mlp.experts.144.up_proj", "model.layers.81.mlp.experts.145.up_proj", "model.layers.81.mlp.experts.146.up_proj", "model.layers.81.mlp.experts.147.up_proj", "model.layers.81.mlp.experts.148.up_proj", "model.layers.81.mlp.experts.149.up_proj", "model.layers.81.mlp.experts.150.up_proj", "model.layers.81.mlp.experts.151.up_proj", "model.layers.81.mlp.experts.152.up_proj", "model.layers.81.mlp.experts.153.up_proj", "model.layers.81.mlp.experts.154.up_proj", "model.layers.81.mlp.experts.155.up_proj", "model.layers.81.mlp.experts.156.up_proj", "model.layers.81.mlp.experts.157.up_proj", "model.layers.81.mlp.experts.158.up_proj", "model.layers.81.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00015011262148618976, "dbits": 2516582400 }, { "dkld": -0.0001595075242221411, "dbits": 5033164800 }, { "dkld": -0.00015765018761158822, "dbits": 7549747200 }, { "dkld": -0.00014431346207857687, "dbits": 12582912000 } ] }, { "idx": 488, "layers": [ "model.layers.81.mlp.experts.0.down_proj", "model.layers.81.mlp.experts.1.down_proj", "model.layers.81.mlp.experts.2.down_proj", "model.layers.81.mlp.experts.3.down_proj", "model.layers.81.mlp.experts.4.down_proj", "model.layers.81.mlp.experts.5.down_proj", "model.layers.81.mlp.experts.6.down_proj", "model.layers.81.mlp.experts.7.down_proj", "model.layers.81.mlp.experts.8.down_proj", "model.layers.81.mlp.experts.9.down_proj", "model.layers.81.mlp.experts.10.down_proj", "model.layers.81.mlp.experts.11.down_proj", "model.layers.81.mlp.experts.12.down_proj", "model.layers.81.mlp.experts.13.down_proj", "model.layers.81.mlp.experts.14.down_proj", "model.layers.81.mlp.experts.15.down_proj", "model.layers.81.mlp.experts.16.down_proj", "model.layers.81.mlp.experts.17.down_proj", "model.layers.81.mlp.experts.18.down_proj", "model.layers.81.mlp.experts.19.down_proj", "model.layers.81.mlp.experts.20.down_proj", "model.layers.81.mlp.experts.21.down_proj", "model.layers.81.mlp.experts.22.down_proj", "model.layers.81.mlp.experts.23.down_proj", "model.layers.81.mlp.experts.24.down_proj", "model.layers.81.mlp.experts.25.down_proj", "model.layers.81.mlp.experts.26.down_proj", "model.layers.81.mlp.experts.27.down_proj", "model.layers.81.mlp.experts.28.down_proj", "model.layers.81.mlp.experts.29.down_proj", "model.layers.81.mlp.experts.30.down_proj", "model.layers.81.mlp.experts.31.down_proj", "model.layers.81.mlp.experts.32.down_proj", "model.layers.81.mlp.experts.33.down_proj", "model.layers.81.mlp.experts.34.down_proj", "model.layers.81.mlp.experts.35.down_proj", "model.layers.81.mlp.experts.36.down_proj", "model.layers.81.mlp.experts.37.down_proj", "model.layers.81.mlp.experts.38.down_proj", "model.layers.81.mlp.experts.39.down_proj", "model.layers.81.mlp.experts.40.down_proj", "model.layers.81.mlp.experts.41.down_proj", "model.layers.81.mlp.experts.42.down_proj", "model.layers.81.mlp.experts.43.down_proj", "model.layers.81.mlp.experts.44.down_proj", "model.layers.81.mlp.experts.45.down_proj", "model.layers.81.mlp.experts.46.down_proj", "model.layers.81.mlp.experts.47.down_proj", "model.layers.81.mlp.experts.48.down_proj", "model.layers.81.mlp.experts.49.down_proj", "model.layers.81.mlp.experts.50.down_proj", "model.layers.81.mlp.experts.51.down_proj", "model.layers.81.mlp.experts.52.down_proj", "model.layers.81.mlp.experts.53.down_proj", "model.layers.81.mlp.experts.54.down_proj", "model.layers.81.mlp.experts.55.down_proj", "model.layers.81.mlp.experts.56.down_proj", "model.layers.81.mlp.experts.57.down_proj", "model.layers.81.mlp.experts.58.down_proj", "model.layers.81.mlp.experts.59.down_proj", "model.layers.81.mlp.experts.60.down_proj", "model.layers.81.mlp.experts.61.down_proj", "model.layers.81.mlp.experts.62.down_proj", "model.layers.81.mlp.experts.63.down_proj", "model.layers.81.mlp.experts.64.down_proj", "model.layers.81.mlp.experts.65.down_proj", "model.layers.81.mlp.experts.66.down_proj", "model.layers.81.mlp.experts.67.down_proj", "model.layers.81.mlp.experts.68.down_proj", "model.layers.81.mlp.experts.69.down_proj", "model.layers.81.mlp.experts.70.down_proj", "model.layers.81.mlp.experts.71.down_proj", "model.layers.81.mlp.experts.72.down_proj", "model.layers.81.mlp.experts.73.down_proj", "model.layers.81.mlp.experts.74.down_proj", "model.layers.81.mlp.experts.75.down_proj", "model.layers.81.mlp.experts.76.down_proj", "model.layers.81.mlp.experts.77.down_proj", "model.layers.81.mlp.experts.78.down_proj", "model.layers.81.mlp.experts.79.down_proj", "model.layers.81.mlp.experts.80.down_proj", "model.layers.81.mlp.experts.81.down_proj", "model.layers.81.mlp.experts.82.down_proj", "model.layers.81.mlp.experts.83.down_proj", "model.layers.81.mlp.experts.84.down_proj", "model.layers.81.mlp.experts.85.down_proj", "model.layers.81.mlp.experts.86.down_proj", "model.layers.81.mlp.experts.87.down_proj", "model.layers.81.mlp.experts.88.down_proj", "model.layers.81.mlp.experts.89.down_proj", "model.layers.81.mlp.experts.90.down_proj", "model.layers.81.mlp.experts.91.down_proj", "model.layers.81.mlp.experts.92.down_proj", "model.layers.81.mlp.experts.93.down_proj", "model.layers.81.mlp.experts.94.down_proj", "model.layers.81.mlp.experts.95.down_proj", "model.layers.81.mlp.experts.96.down_proj", "model.layers.81.mlp.experts.97.down_proj", "model.layers.81.mlp.experts.98.down_proj", "model.layers.81.mlp.experts.99.down_proj", "model.layers.81.mlp.experts.100.down_proj", "model.layers.81.mlp.experts.101.down_proj", "model.layers.81.mlp.experts.102.down_proj", "model.layers.81.mlp.experts.103.down_proj", "model.layers.81.mlp.experts.104.down_proj", "model.layers.81.mlp.experts.105.down_proj", "model.layers.81.mlp.experts.106.down_proj", "model.layers.81.mlp.experts.107.down_proj", "model.layers.81.mlp.experts.108.down_proj", "model.layers.81.mlp.experts.109.down_proj", "model.layers.81.mlp.experts.110.down_proj", "model.layers.81.mlp.experts.111.down_proj", "model.layers.81.mlp.experts.112.down_proj", "model.layers.81.mlp.experts.113.down_proj", "model.layers.81.mlp.experts.114.down_proj", "model.layers.81.mlp.experts.115.down_proj", "model.layers.81.mlp.experts.116.down_proj", "model.layers.81.mlp.experts.117.down_proj", "model.layers.81.mlp.experts.118.down_proj", "model.layers.81.mlp.experts.119.down_proj", "model.layers.81.mlp.experts.120.down_proj", "model.layers.81.mlp.experts.121.down_proj", "model.layers.81.mlp.experts.122.down_proj", "model.layers.81.mlp.experts.123.down_proj", "model.layers.81.mlp.experts.124.down_proj", "model.layers.81.mlp.experts.125.down_proj", "model.layers.81.mlp.experts.126.down_proj", "model.layers.81.mlp.experts.127.down_proj", "model.layers.81.mlp.experts.128.down_proj", "model.layers.81.mlp.experts.129.down_proj", "model.layers.81.mlp.experts.130.down_proj", "model.layers.81.mlp.experts.131.down_proj", "model.layers.81.mlp.experts.132.down_proj", "model.layers.81.mlp.experts.133.down_proj", "model.layers.81.mlp.experts.134.down_proj", "model.layers.81.mlp.experts.135.down_proj", "model.layers.81.mlp.experts.136.down_proj", "model.layers.81.mlp.experts.137.down_proj", "model.layers.81.mlp.experts.138.down_proj", "model.layers.81.mlp.experts.139.down_proj", "model.layers.81.mlp.experts.140.down_proj", "model.layers.81.mlp.experts.141.down_proj", "model.layers.81.mlp.experts.142.down_proj", "model.layers.81.mlp.experts.143.down_proj", "model.layers.81.mlp.experts.144.down_proj", "model.layers.81.mlp.experts.145.down_proj", "model.layers.81.mlp.experts.146.down_proj", "model.layers.81.mlp.experts.147.down_proj", "model.layers.81.mlp.experts.148.down_proj", "model.layers.81.mlp.experts.149.down_proj", "model.layers.81.mlp.experts.150.down_proj", "model.layers.81.mlp.experts.151.down_proj", "model.layers.81.mlp.experts.152.down_proj", "model.layers.81.mlp.experts.153.down_proj", "model.layers.81.mlp.experts.154.down_proj", "model.layers.81.mlp.experts.155.down_proj", "model.layers.81.mlp.experts.156.down_proj", "model.layers.81.mlp.experts.157.down_proj", "model.layers.81.mlp.experts.158.down_proj", "model.layers.81.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -5.323849618435461e-05, "dbits": 1258291200 }, { "dkld": -0.00010273754596711038, "dbits": 2516582400 }, { "dkld": -9.730691090226173e-05, "dbits": 3774873600 }, { "dkld": -0.00011564698070287704, "dbits": 6291456000 } ] }, { "idx": 489, "layers": [ "model.layers.82.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00013682544231415072, "dbits": 62914560 }, { "dkld": -0.00027533611282706816, "dbits": 125829120 }, { "dkld": -0.0001959764398634517, "dbits": 188743680 }, { "dkld": -0.0002121670171618545, "dbits": 314572800 } ] }, { "idx": 490, "layers": [ "model.layers.82.self_attn.k_proj", "model.layers.82.self_attn.v_proj" ], "candidates": [ { "dkld": -9.257243946195204e-05, "dbits": 10485760 }, { "dkld": -0.00016780253499747155, "dbits": 20971520 }, { "dkld": -0.00016148574650287628, "dbits": 31457280 }, { "dkld": -0.0001931065693497741, "dbits": 52428800 } ] }, { "idx": 491, "layers": [ "model.layers.82.self_attn.o_proj" ], "candidates": [ { "dkld": -8.017010986807738e-06, "dbits": 62914560 }, { "dkld": -0.0002345322631299579, "dbits": 125829120 }, { "dkld": -0.0001959876157343471, "dbits": 188743680 }, { "dkld": -0.00020585274323821345, "dbits": 314572800 } ] }, { "idx": 492, "layers": [ "model.layers.82.mlp.shared_experts.gate_proj", "model.layers.82.mlp.shared_experts.up_proj", "model.layers.82.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.00021654851734637104, "dbits": 23592960 }, { "dkld": 0.0002875948324799482, "dbits": 47185920 }, { "dkld": 0.0002126904204487745, "dbits": 70778880 }, { "dkld": 0.00013259649276732566, "dbits": 117964800 } ] }, { "idx": 493, "layers": [ "model.layers.82.mlp.experts.0.gate_proj", "model.layers.82.mlp.experts.1.gate_proj", "model.layers.82.mlp.experts.2.gate_proj", "model.layers.82.mlp.experts.3.gate_proj", "model.layers.82.mlp.experts.4.gate_proj", "model.layers.82.mlp.experts.5.gate_proj", "model.layers.82.mlp.experts.6.gate_proj", "model.layers.82.mlp.experts.7.gate_proj", "model.layers.82.mlp.experts.8.gate_proj", "model.layers.82.mlp.experts.9.gate_proj", "model.layers.82.mlp.experts.10.gate_proj", "model.layers.82.mlp.experts.11.gate_proj", "model.layers.82.mlp.experts.12.gate_proj", "model.layers.82.mlp.experts.13.gate_proj", "model.layers.82.mlp.experts.14.gate_proj", "model.layers.82.mlp.experts.15.gate_proj", "model.layers.82.mlp.experts.16.gate_proj", "model.layers.82.mlp.experts.17.gate_proj", "model.layers.82.mlp.experts.18.gate_proj", "model.layers.82.mlp.experts.19.gate_proj", "model.layers.82.mlp.experts.20.gate_proj", "model.layers.82.mlp.experts.21.gate_proj", "model.layers.82.mlp.experts.22.gate_proj", "model.layers.82.mlp.experts.23.gate_proj", "model.layers.82.mlp.experts.24.gate_proj", "model.layers.82.mlp.experts.25.gate_proj", "model.layers.82.mlp.experts.26.gate_proj", "model.layers.82.mlp.experts.27.gate_proj", "model.layers.82.mlp.experts.28.gate_proj", "model.layers.82.mlp.experts.29.gate_proj", "model.layers.82.mlp.experts.30.gate_proj", "model.layers.82.mlp.experts.31.gate_proj", "model.layers.82.mlp.experts.32.gate_proj", "model.layers.82.mlp.experts.33.gate_proj", "model.layers.82.mlp.experts.34.gate_proj", "model.layers.82.mlp.experts.35.gate_proj", "model.layers.82.mlp.experts.36.gate_proj", "model.layers.82.mlp.experts.37.gate_proj", "model.layers.82.mlp.experts.38.gate_proj", "model.layers.82.mlp.experts.39.gate_proj", "model.layers.82.mlp.experts.40.gate_proj", "model.layers.82.mlp.experts.41.gate_proj", "model.layers.82.mlp.experts.42.gate_proj", "model.layers.82.mlp.experts.43.gate_proj", "model.layers.82.mlp.experts.44.gate_proj", "model.layers.82.mlp.experts.45.gate_proj", "model.layers.82.mlp.experts.46.gate_proj", "model.layers.82.mlp.experts.47.gate_proj", "model.layers.82.mlp.experts.48.gate_proj", "model.layers.82.mlp.experts.49.gate_proj", "model.layers.82.mlp.experts.50.gate_proj", "model.layers.82.mlp.experts.51.gate_proj", "model.layers.82.mlp.experts.52.gate_proj", "model.layers.82.mlp.experts.53.gate_proj", "model.layers.82.mlp.experts.54.gate_proj", "model.layers.82.mlp.experts.55.gate_proj", "model.layers.82.mlp.experts.56.gate_proj", "model.layers.82.mlp.experts.57.gate_proj", "model.layers.82.mlp.experts.58.gate_proj", "model.layers.82.mlp.experts.59.gate_proj", "model.layers.82.mlp.experts.60.gate_proj", "model.layers.82.mlp.experts.61.gate_proj", "model.layers.82.mlp.experts.62.gate_proj", "model.layers.82.mlp.experts.63.gate_proj", "model.layers.82.mlp.experts.64.gate_proj", "model.layers.82.mlp.experts.65.gate_proj", "model.layers.82.mlp.experts.66.gate_proj", "model.layers.82.mlp.experts.67.gate_proj", "model.layers.82.mlp.experts.68.gate_proj", "model.layers.82.mlp.experts.69.gate_proj", "model.layers.82.mlp.experts.70.gate_proj", "model.layers.82.mlp.experts.71.gate_proj", "model.layers.82.mlp.experts.72.gate_proj", "model.layers.82.mlp.experts.73.gate_proj", "model.layers.82.mlp.experts.74.gate_proj", "model.layers.82.mlp.experts.75.gate_proj", "model.layers.82.mlp.experts.76.gate_proj", "model.layers.82.mlp.experts.77.gate_proj", "model.layers.82.mlp.experts.78.gate_proj", "model.layers.82.mlp.experts.79.gate_proj", "model.layers.82.mlp.experts.80.gate_proj", "model.layers.82.mlp.experts.81.gate_proj", "model.layers.82.mlp.experts.82.gate_proj", "model.layers.82.mlp.experts.83.gate_proj", "model.layers.82.mlp.experts.84.gate_proj", "model.layers.82.mlp.experts.85.gate_proj", "model.layers.82.mlp.experts.86.gate_proj", "model.layers.82.mlp.experts.87.gate_proj", "model.layers.82.mlp.experts.88.gate_proj", "model.layers.82.mlp.experts.89.gate_proj", "model.layers.82.mlp.experts.90.gate_proj", "model.layers.82.mlp.experts.91.gate_proj", "model.layers.82.mlp.experts.92.gate_proj", "model.layers.82.mlp.experts.93.gate_proj", "model.layers.82.mlp.experts.94.gate_proj", "model.layers.82.mlp.experts.95.gate_proj", "model.layers.82.mlp.experts.96.gate_proj", "model.layers.82.mlp.experts.97.gate_proj", "model.layers.82.mlp.experts.98.gate_proj", "model.layers.82.mlp.experts.99.gate_proj", "model.layers.82.mlp.experts.100.gate_proj", "model.layers.82.mlp.experts.101.gate_proj", "model.layers.82.mlp.experts.102.gate_proj", "model.layers.82.mlp.experts.103.gate_proj", "model.layers.82.mlp.experts.104.gate_proj", "model.layers.82.mlp.experts.105.gate_proj", "model.layers.82.mlp.experts.106.gate_proj", "model.layers.82.mlp.experts.107.gate_proj", "model.layers.82.mlp.experts.108.gate_proj", "model.layers.82.mlp.experts.109.gate_proj", "model.layers.82.mlp.experts.110.gate_proj", "model.layers.82.mlp.experts.111.gate_proj", "model.layers.82.mlp.experts.112.gate_proj", "model.layers.82.mlp.experts.113.gate_proj", "model.layers.82.mlp.experts.114.gate_proj", "model.layers.82.mlp.experts.115.gate_proj", "model.layers.82.mlp.experts.116.gate_proj", "model.layers.82.mlp.experts.117.gate_proj", "model.layers.82.mlp.experts.118.gate_proj", "model.layers.82.mlp.experts.119.gate_proj", "model.layers.82.mlp.experts.120.gate_proj", "model.layers.82.mlp.experts.121.gate_proj", "model.layers.82.mlp.experts.122.gate_proj", "model.layers.82.mlp.experts.123.gate_proj", "model.layers.82.mlp.experts.124.gate_proj", "model.layers.82.mlp.experts.125.gate_proj", "model.layers.82.mlp.experts.126.gate_proj", "model.layers.82.mlp.experts.127.gate_proj", "model.layers.82.mlp.experts.128.gate_proj", "model.layers.82.mlp.experts.129.gate_proj", "model.layers.82.mlp.experts.130.gate_proj", "model.layers.82.mlp.experts.131.gate_proj", "model.layers.82.mlp.experts.132.gate_proj", "model.layers.82.mlp.experts.133.gate_proj", "model.layers.82.mlp.experts.134.gate_proj", "model.layers.82.mlp.experts.135.gate_proj", "model.layers.82.mlp.experts.136.gate_proj", "model.layers.82.mlp.experts.137.gate_proj", "model.layers.82.mlp.experts.138.gate_proj", "model.layers.82.mlp.experts.139.gate_proj", "model.layers.82.mlp.experts.140.gate_proj", "model.layers.82.mlp.experts.141.gate_proj", "model.layers.82.mlp.experts.142.gate_proj", "model.layers.82.mlp.experts.143.gate_proj", "model.layers.82.mlp.experts.144.gate_proj", "model.layers.82.mlp.experts.145.gate_proj", "model.layers.82.mlp.experts.146.gate_proj", "model.layers.82.mlp.experts.147.gate_proj", "model.layers.82.mlp.experts.148.gate_proj", "model.layers.82.mlp.experts.149.gate_proj", "model.layers.82.mlp.experts.150.gate_proj", "model.layers.82.mlp.experts.151.gate_proj", "model.layers.82.mlp.experts.152.gate_proj", "model.layers.82.mlp.experts.153.gate_proj", "model.layers.82.mlp.experts.154.gate_proj", "model.layers.82.mlp.experts.155.gate_proj", "model.layers.82.mlp.experts.156.gate_proj", "model.layers.82.mlp.experts.157.gate_proj", "model.layers.82.mlp.experts.158.gate_proj", "model.layers.82.mlp.experts.159.gate_proj", "model.layers.82.mlp.experts.0.up_proj", "model.layers.82.mlp.experts.1.up_proj", "model.layers.82.mlp.experts.2.up_proj", "model.layers.82.mlp.experts.3.up_proj", "model.layers.82.mlp.experts.4.up_proj", "model.layers.82.mlp.experts.5.up_proj", "model.layers.82.mlp.experts.6.up_proj", "model.layers.82.mlp.experts.7.up_proj", "model.layers.82.mlp.experts.8.up_proj", "model.layers.82.mlp.experts.9.up_proj", "model.layers.82.mlp.experts.10.up_proj", "model.layers.82.mlp.experts.11.up_proj", "model.layers.82.mlp.experts.12.up_proj", "model.layers.82.mlp.experts.13.up_proj", "model.layers.82.mlp.experts.14.up_proj", "model.layers.82.mlp.experts.15.up_proj", "model.layers.82.mlp.experts.16.up_proj", "model.layers.82.mlp.experts.17.up_proj", "model.layers.82.mlp.experts.18.up_proj", "model.layers.82.mlp.experts.19.up_proj", "model.layers.82.mlp.experts.20.up_proj", "model.layers.82.mlp.experts.21.up_proj", "model.layers.82.mlp.experts.22.up_proj", "model.layers.82.mlp.experts.23.up_proj", "model.layers.82.mlp.experts.24.up_proj", "model.layers.82.mlp.experts.25.up_proj", "model.layers.82.mlp.experts.26.up_proj", "model.layers.82.mlp.experts.27.up_proj", "model.layers.82.mlp.experts.28.up_proj", "model.layers.82.mlp.experts.29.up_proj", "model.layers.82.mlp.experts.30.up_proj", "model.layers.82.mlp.experts.31.up_proj", "model.layers.82.mlp.experts.32.up_proj", "model.layers.82.mlp.experts.33.up_proj", "model.layers.82.mlp.experts.34.up_proj", "model.layers.82.mlp.experts.35.up_proj", "model.layers.82.mlp.experts.36.up_proj", "model.layers.82.mlp.experts.37.up_proj", "model.layers.82.mlp.experts.38.up_proj", "model.layers.82.mlp.experts.39.up_proj", "model.layers.82.mlp.experts.40.up_proj", "model.layers.82.mlp.experts.41.up_proj", "model.layers.82.mlp.experts.42.up_proj", "model.layers.82.mlp.experts.43.up_proj", "model.layers.82.mlp.experts.44.up_proj", "model.layers.82.mlp.experts.45.up_proj", "model.layers.82.mlp.experts.46.up_proj", "model.layers.82.mlp.experts.47.up_proj", "model.layers.82.mlp.experts.48.up_proj", "model.layers.82.mlp.experts.49.up_proj", "model.layers.82.mlp.experts.50.up_proj", "model.layers.82.mlp.experts.51.up_proj", "model.layers.82.mlp.experts.52.up_proj", "model.layers.82.mlp.experts.53.up_proj", "model.layers.82.mlp.experts.54.up_proj", "model.layers.82.mlp.experts.55.up_proj", "model.layers.82.mlp.experts.56.up_proj", "model.layers.82.mlp.experts.57.up_proj", "model.layers.82.mlp.experts.58.up_proj", "model.layers.82.mlp.experts.59.up_proj", "model.layers.82.mlp.experts.60.up_proj", "model.layers.82.mlp.experts.61.up_proj", "model.layers.82.mlp.experts.62.up_proj", "model.layers.82.mlp.experts.63.up_proj", "model.layers.82.mlp.experts.64.up_proj", "model.layers.82.mlp.experts.65.up_proj", "model.layers.82.mlp.experts.66.up_proj", "model.layers.82.mlp.experts.67.up_proj", "model.layers.82.mlp.experts.68.up_proj", "model.layers.82.mlp.experts.69.up_proj", "model.layers.82.mlp.experts.70.up_proj", "model.layers.82.mlp.experts.71.up_proj", "model.layers.82.mlp.experts.72.up_proj", "model.layers.82.mlp.experts.73.up_proj", "model.layers.82.mlp.experts.74.up_proj", "model.layers.82.mlp.experts.75.up_proj", "model.layers.82.mlp.experts.76.up_proj", "model.layers.82.mlp.experts.77.up_proj", "model.layers.82.mlp.experts.78.up_proj", "model.layers.82.mlp.experts.79.up_proj", "model.layers.82.mlp.experts.80.up_proj", "model.layers.82.mlp.experts.81.up_proj", "model.layers.82.mlp.experts.82.up_proj", "model.layers.82.mlp.experts.83.up_proj", "model.layers.82.mlp.experts.84.up_proj", "model.layers.82.mlp.experts.85.up_proj", "model.layers.82.mlp.experts.86.up_proj", "model.layers.82.mlp.experts.87.up_proj", "model.layers.82.mlp.experts.88.up_proj", "model.layers.82.mlp.experts.89.up_proj", "model.layers.82.mlp.experts.90.up_proj", "model.layers.82.mlp.experts.91.up_proj", "model.layers.82.mlp.experts.92.up_proj", "model.layers.82.mlp.experts.93.up_proj", "model.layers.82.mlp.experts.94.up_proj", "model.layers.82.mlp.experts.95.up_proj", "model.layers.82.mlp.experts.96.up_proj", "model.layers.82.mlp.experts.97.up_proj", "model.layers.82.mlp.experts.98.up_proj", "model.layers.82.mlp.experts.99.up_proj", "model.layers.82.mlp.experts.100.up_proj", "model.layers.82.mlp.experts.101.up_proj", "model.layers.82.mlp.experts.102.up_proj", "model.layers.82.mlp.experts.103.up_proj", "model.layers.82.mlp.experts.104.up_proj", "model.layers.82.mlp.experts.105.up_proj", "model.layers.82.mlp.experts.106.up_proj", "model.layers.82.mlp.experts.107.up_proj", "model.layers.82.mlp.experts.108.up_proj", "model.layers.82.mlp.experts.109.up_proj", "model.layers.82.mlp.experts.110.up_proj", "model.layers.82.mlp.experts.111.up_proj", "model.layers.82.mlp.experts.112.up_proj", "model.layers.82.mlp.experts.113.up_proj", "model.layers.82.mlp.experts.114.up_proj", "model.layers.82.mlp.experts.115.up_proj", "model.layers.82.mlp.experts.116.up_proj", "model.layers.82.mlp.experts.117.up_proj", "model.layers.82.mlp.experts.118.up_proj", "model.layers.82.mlp.experts.119.up_proj", "model.layers.82.mlp.experts.120.up_proj", "model.layers.82.mlp.experts.121.up_proj", "model.layers.82.mlp.experts.122.up_proj", "model.layers.82.mlp.experts.123.up_proj", "model.layers.82.mlp.experts.124.up_proj", "model.layers.82.mlp.experts.125.up_proj", "model.layers.82.mlp.experts.126.up_proj", "model.layers.82.mlp.experts.127.up_proj", "model.layers.82.mlp.experts.128.up_proj", "model.layers.82.mlp.experts.129.up_proj", "model.layers.82.mlp.experts.130.up_proj", "model.layers.82.mlp.experts.131.up_proj", "model.layers.82.mlp.experts.132.up_proj", "model.layers.82.mlp.experts.133.up_proj", "model.layers.82.mlp.experts.134.up_proj", "model.layers.82.mlp.experts.135.up_proj", "model.layers.82.mlp.experts.136.up_proj", "model.layers.82.mlp.experts.137.up_proj", "model.layers.82.mlp.experts.138.up_proj", "model.layers.82.mlp.experts.139.up_proj", "model.layers.82.mlp.experts.140.up_proj", "model.layers.82.mlp.experts.141.up_proj", "model.layers.82.mlp.experts.142.up_proj", "model.layers.82.mlp.experts.143.up_proj", "model.layers.82.mlp.experts.144.up_proj", "model.layers.82.mlp.experts.145.up_proj", "model.layers.82.mlp.experts.146.up_proj", "model.layers.82.mlp.experts.147.up_proj", "model.layers.82.mlp.experts.148.up_proj", "model.layers.82.mlp.experts.149.up_proj", "model.layers.82.mlp.experts.150.up_proj", "model.layers.82.mlp.experts.151.up_proj", "model.layers.82.mlp.experts.152.up_proj", "model.layers.82.mlp.experts.153.up_proj", "model.layers.82.mlp.experts.154.up_proj", "model.layers.82.mlp.experts.155.up_proj", "model.layers.82.mlp.experts.156.up_proj", "model.layers.82.mlp.experts.157.up_proj", "model.layers.82.mlp.experts.158.up_proj", "model.layers.82.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.0002772534266114318, "dbits": 2516582400 }, { "dkld": -0.0002911992371082417, "dbits": 5033164800 }, { "dkld": -0.00029375376179814616, "dbits": 7549747200 }, { "dkld": -0.00028186347335577844, "dbits": 12582912000 } ] }, { "idx": 494, "layers": [ "model.layers.82.mlp.experts.0.down_proj", "model.layers.82.mlp.experts.1.down_proj", "model.layers.82.mlp.experts.2.down_proj", "model.layers.82.mlp.experts.3.down_proj", "model.layers.82.mlp.experts.4.down_proj", "model.layers.82.mlp.experts.5.down_proj", "model.layers.82.mlp.experts.6.down_proj", "model.layers.82.mlp.experts.7.down_proj", "model.layers.82.mlp.experts.8.down_proj", "model.layers.82.mlp.experts.9.down_proj", "model.layers.82.mlp.experts.10.down_proj", "model.layers.82.mlp.experts.11.down_proj", "model.layers.82.mlp.experts.12.down_proj", "model.layers.82.mlp.experts.13.down_proj", "model.layers.82.mlp.experts.14.down_proj", "model.layers.82.mlp.experts.15.down_proj", "model.layers.82.mlp.experts.16.down_proj", "model.layers.82.mlp.experts.17.down_proj", "model.layers.82.mlp.experts.18.down_proj", "model.layers.82.mlp.experts.19.down_proj", "model.layers.82.mlp.experts.20.down_proj", "model.layers.82.mlp.experts.21.down_proj", "model.layers.82.mlp.experts.22.down_proj", "model.layers.82.mlp.experts.23.down_proj", "model.layers.82.mlp.experts.24.down_proj", "model.layers.82.mlp.experts.25.down_proj", "model.layers.82.mlp.experts.26.down_proj", "model.layers.82.mlp.experts.27.down_proj", "model.layers.82.mlp.experts.28.down_proj", "model.layers.82.mlp.experts.29.down_proj", "model.layers.82.mlp.experts.30.down_proj", "model.layers.82.mlp.experts.31.down_proj", "model.layers.82.mlp.experts.32.down_proj", "model.layers.82.mlp.experts.33.down_proj", "model.layers.82.mlp.experts.34.down_proj", "model.layers.82.mlp.experts.35.down_proj", "model.layers.82.mlp.experts.36.down_proj", "model.layers.82.mlp.experts.37.down_proj", "model.layers.82.mlp.experts.38.down_proj", "model.layers.82.mlp.experts.39.down_proj", "model.layers.82.mlp.experts.40.down_proj", "model.layers.82.mlp.experts.41.down_proj", "model.layers.82.mlp.experts.42.down_proj", "model.layers.82.mlp.experts.43.down_proj", "model.layers.82.mlp.experts.44.down_proj", "model.layers.82.mlp.experts.45.down_proj", "model.layers.82.mlp.experts.46.down_proj", "model.layers.82.mlp.experts.47.down_proj", "model.layers.82.mlp.experts.48.down_proj", "model.layers.82.mlp.experts.49.down_proj", "model.layers.82.mlp.experts.50.down_proj", "model.layers.82.mlp.experts.51.down_proj", "model.layers.82.mlp.experts.52.down_proj", "model.layers.82.mlp.experts.53.down_proj", "model.layers.82.mlp.experts.54.down_proj", "model.layers.82.mlp.experts.55.down_proj", "model.layers.82.mlp.experts.56.down_proj", "model.layers.82.mlp.experts.57.down_proj", "model.layers.82.mlp.experts.58.down_proj", "model.layers.82.mlp.experts.59.down_proj", "model.layers.82.mlp.experts.60.down_proj", "model.layers.82.mlp.experts.61.down_proj", "model.layers.82.mlp.experts.62.down_proj", "model.layers.82.mlp.experts.63.down_proj", "model.layers.82.mlp.experts.64.down_proj", "model.layers.82.mlp.experts.65.down_proj", "model.layers.82.mlp.experts.66.down_proj", "model.layers.82.mlp.experts.67.down_proj", "model.layers.82.mlp.experts.68.down_proj", "model.layers.82.mlp.experts.69.down_proj", "model.layers.82.mlp.experts.70.down_proj", "model.layers.82.mlp.experts.71.down_proj", "model.layers.82.mlp.experts.72.down_proj", "model.layers.82.mlp.experts.73.down_proj", "model.layers.82.mlp.experts.74.down_proj", "model.layers.82.mlp.experts.75.down_proj", "model.layers.82.mlp.experts.76.down_proj", "model.layers.82.mlp.experts.77.down_proj", "model.layers.82.mlp.experts.78.down_proj", "model.layers.82.mlp.experts.79.down_proj", "model.layers.82.mlp.experts.80.down_proj", "model.layers.82.mlp.experts.81.down_proj", "model.layers.82.mlp.experts.82.down_proj", "model.layers.82.mlp.experts.83.down_proj", "model.layers.82.mlp.experts.84.down_proj", "model.layers.82.mlp.experts.85.down_proj", "model.layers.82.mlp.experts.86.down_proj", "model.layers.82.mlp.experts.87.down_proj", "model.layers.82.mlp.experts.88.down_proj", "model.layers.82.mlp.experts.89.down_proj", "model.layers.82.mlp.experts.90.down_proj", "model.layers.82.mlp.experts.91.down_proj", "model.layers.82.mlp.experts.92.down_proj", "model.layers.82.mlp.experts.93.down_proj", "model.layers.82.mlp.experts.94.down_proj", "model.layers.82.mlp.experts.95.down_proj", "model.layers.82.mlp.experts.96.down_proj", "model.layers.82.mlp.experts.97.down_proj", "model.layers.82.mlp.experts.98.down_proj", "model.layers.82.mlp.experts.99.down_proj", "model.layers.82.mlp.experts.100.down_proj", "model.layers.82.mlp.experts.101.down_proj", "model.layers.82.mlp.experts.102.down_proj", "model.layers.82.mlp.experts.103.down_proj", "model.layers.82.mlp.experts.104.down_proj", "model.layers.82.mlp.experts.105.down_proj", "model.layers.82.mlp.experts.106.down_proj", "model.layers.82.mlp.experts.107.down_proj", "model.layers.82.mlp.experts.108.down_proj", "model.layers.82.mlp.experts.109.down_proj", "model.layers.82.mlp.experts.110.down_proj", "model.layers.82.mlp.experts.111.down_proj", "model.layers.82.mlp.experts.112.down_proj", "model.layers.82.mlp.experts.113.down_proj", "model.layers.82.mlp.experts.114.down_proj", "model.layers.82.mlp.experts.115.down_proj", "model.layers.82.mlp.experts.116.down_proj", "model.layers.82.mlp.experts.117.down_proj", "model.layers.82.mlp.experts.118.down_proj", "model.layers.82.mlp.experts.119.down_proj", "model.layers.82.mlp.experts.120.down_proj", "model.layers.82.mlp.experts.121.down_proj", "model.layers.82.mlp.experts.122.down_proj", "model.layers.82.mlp.experts.123.down_proj", "model.layers.82.mlp.experts.124.down_proj", "model.layers.82.mlp.experts.125.down_proj", "model.layers.82.mlp.experts.126.down_proj", "model.layers.82.mlp.experts.127.down_proj", "model.layers.82.mlp.experts.128.down_proj", "model.layers.82.mlp.experts.129.down_proj", "model.layers.82.mlp.experts.130.down_proj", "model.layers.82.mlp.experts.131.down_proj", "model.layers.82.mlp.experts.132.down_proj", "model.layers.82.mlp.experts.133.down_proj", "model.layers.82.mlp.experts.134.down_proj", "model.layers.82.mlp.experts.135.down_proj", "model.layers.82.mlp.experts.136.down_proj", "model.layers.82.mlp.experts.137.down_proj", "model.layers.82.mlp.experts.138.down_proj", "model.layers.82.mlp.experts.139.down_proj", "model.layers.82.mlp.experts.140.down_proj", "model.layers.82.mlp.experts.141.down_proj", "model.layers.82.mlp.experts.142.down_proj", "model.layers.82.mlp.experts.143.down_proj", "model.layers.82.mlp.experts.144.down_proj", "model.layers.82.mlp.experts.145.down_proj", "model.layers.82.mlp.experts.146.down_proj", "model.layers.82.mlp.experts.147.down_proj", "model.layers.82.mlp.experts.148.down_proj", "model.layers.82.mlp.experts.149.down_proj", "model.layers.82.mlp.experts.150.down_proj", "model.layers.82.mlp.experts.151.down_proj", "model.layers.82.mlp.experts.152.down_proj", "model.layers.82.mlp.experts.153.down_proj", "model.layers.82.mlp.experts.154.down_proj", "model.layers.82.mlp.experts.155.down_proj", "model.layers.82.mlp.experts.156.down_proj", "model.layers.82.mlp.experts.157.down_proj", "model.layers.82.mlp.experts.158.down_proj", "model.layers.82.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -0.00015501538291573524, "dbits": 1258291200 }, { "dkld": -0.00018354654312134622, "dbits": 2516582400 }, { "dkld": -0.0001697193831205479, "dbits": 3774873600 }, { "dkld": -0.00016715331003069878, "dbits": 6291456000 } ] }, { "idx": 495, "layers": [ "model.layers.83.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00018685348331928808, "dbits": 62914560 }, { "dkld": -0.00016017872840166925, "dbits": 125829120 }, { "dkld": -0.0002615317702293479, "dbits": 188743680 }, { "dkld": -0.0002464449964463822, "dbits": 314572800 } ] }, { "idx": 496, "layers": [ "model.layers.83.self_attn.k_proj", "model.layers.83.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0006761783733963966, "dbits": 10485760 }, { "dkld": -0.0009316165000200355, "dbits": 20971520 }, { "dkld": -0.0008338637650013025, "dbits": 31457280 }, { "dkld": -0.0007885630242526614, "dbits": 52428800 } ] }, { "idx": 497, "layers": [ "model.layers.83.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00011791577562689226, "dbits": 62914560 }, { "dkld": -4.897695034743188e-05, "dbits": 125829120 }, { "dkld": 5.0045643001789264e-05, "dbits": 188743680 }, { "dkld": 5.7475641369736374e-06, "dbits": 314572800 } ] }, { "idx": 498, "layers": [ "model.layers.83.mlp.shared_experts.gate_proj", "model.layers.83.mlp.shared_experts.up_proj", "model.layers.83.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00015290416777134774, "dbits": 23592960 }, { "dkld": -0.00023240251466633, "dbits": 47185920 }, { "dkld": -0.00010349471122027276, "dbits": 70778880 }, { "dkld": -0.00018764464184642393, "dbits": 117964800 } ] }, { "idx": 499, "layers": [ "model.layers.83.mlp.experts.0.gate_proj", "model.layers.83.mlp.experts.1.gate_proj", "model.layers.83.mlp.experts.2.gate_proj", "model.layers.83.mlp.experts.3.gate_proj", "model.layers.83.mlp.experts.4.gate_proj", "model.layers.83.mlp.experts.5.gate_proj", "model.layers.83.mlp.experts.6.gate_proj", "model.layers.83.mlp.experts.7.gate_proj", "model.layers.83.mlp.experts.8.gate_proj", "model.layers.83.mlp.experts.9.gate_proj", "model.layers.83.mlp.experts.10.gate_proj", "model.layers.83.mlp.experts.11.gate_proj", "model.layers.83.mlp.experts.12.gate_proj", "model.layers.83.mlp.experts.13.gate_proj", "model.layers.83.mlp.experts.14.gate_proj", "model.layers.83.mlp.experts.15.gate_proj", "model.layers.83.mlp.experts.16.gate_proj", "model.layers.83.mlp.experts.17.gate_proj", "model.layers.83.mlp.experts.18.gate_proj", "model.layers.83.mlp.experts.19.gate_proj", "model.layers.83.mlp.experts.20.gate_proj", "model.layers.83.mlp.experts.21.gate_proj", "model.layers.83.mlp.experts.22.gate_proj", "model.layers.83.mlp.experts.23.gate_proj", "model.layers.83.mlp.experts.24.gate_proj", "model.layers.83.mlp.experts.25.gate_proj", "model.layers.83.mlp.experts.26.gate_proj", "model.layers.83.mlp.experts.27.gate_proj", "model.layers.83.mlp.experts.28.gate_proj", "model.layers.83.mlp.experts.29.gate_proj", "model.layers.83.mlp.experts.30.gate_proj", "model.layers.83.mlp.experts.31.gate_proj", "model.layers.83.mlp.experts.32.gate_proj", "model.layers.83.mlp.experts.33.gate_proj", "model.layers.83.mlp.experts.34.gate_proj", "model.layers.83.mlp.experts.35.gate_proj", "model.layers.83.mlp.experts.36.gate_proj", "model.layers.83.mlp.experts.37.gate_proj", "model.layers.83.mlp.experts.38.gate_proj", "model.layers.83.mlp.experts.39.gate_proj", "model.layers.83.mlp.experts.40.gate_proj", "model.layers.83.mlp.experts.41.gate_proj", "model.layers.83.mlp.experts.42.gate_proj", "model.layers.83.mlp.experts.43.gate_proj", "model.layers.83.mlp.experts.44.gate_proj", "model.layers.83.mlp.experts.45.gate_proj", "model.layers.83.mlp.experts.46.gate_proj", "model.layers.83.mlp.experts.47.gate_proj", "model.layers.83.mlp.experts.48.gate_proj", "model.layers.83.mlp.experts.49.gate_proj", "model.layers.83.mlp.experts.50.gate_proj", "model.layers.83.mlp.experts.51.gate_proj", "model.layers.83.mlp.experts.52.gate_proj", "model.layers.83.mlp.experts.53.gate_proj", "model.layers.83.mlp.experts.54.gate_proj", "model.layers.83.mlp.experts.55.gate_proj", "model.layers.83.mlp.experts.56.gate_proj", "model.layers.83.mlp.experts.57.gate_proj", "model.layers.83.mlp.experts.58.gate_proj", "model.layers.83.mlp.experts.59.gate_proj", "model.layers.83.mlp.experts.60.gate_proj", "model.layers.83.mlp.experts.61.gate_proj", "model.layers.83.mlp.experts.62.gate_proj", "model.layers.83.mlp.experts.63.gate_proj", "model.layers.83.mlp.experts.64.gate_proj", "model.layers.83.mlp.experts.65.gate_proj", "model.layers.83.mlp.experts.66.gate_proj", "model.layers.83.mlp.experts.67.gate_proj", "model.layers.83.mlp.experts.68.gate_proj", "model.layers.83.mlp.experts.69.gate_proj", "model.layers.83.mlp.experts.70.gate_proj", "model.layers.83.mlp.experts.71.gate_proj", "model.layers.83.mlp.experts.72.gate_proj", "model.layers.83.mlp.experts.73.gate_proj", "model.layers.83.mlp.experts.74.gate_proj", "model.layers.83.mlp.experts.75.gate_proj", "model.layers.83.mlp.experts.76.gate_proj", "model.layers.83.mlp.experts.77.gate_proj", "model.layers.83.mlp.experts.78.gate_proj", "model.layers.83.mlp.experts.79.gate_proj", "model.layers.83.mlp.experts.80.gate_proj", "model.layers.83.mlp.experts.81.gate_proj", "model.layers.83.mlp.experts.82.gate_proj", "model.layers.83.mlp.experts.83.gate_proj", "model.layers.83.mlp.experts.84.gate_proj", "model.layers.83.mlp.experts.85.gate_proj", "model.layers.83.mlp.experts.86.gate_proj", "model.layers.83.mlp.experts.87.gate_proj", "model.layers.83.mlp.experts.88.gate_proj", "model.layers.83.mlp.experts.89.gate_proj", "model.layers.83.mlp.experts.90.gate_proj", "model.layers.83.mlp.experts.91.gate_proj", "model.layers.83.mlp.experts.92.gate_proj", "model.layers.83.mlp.experts.93.gate_proj", "model.layers.83.mlp.experts.94.gate_proj", "model.layers.83.mlp.experts.95.gate_proj", "model.layers.83.mlp.experts.96.gate_proj", "model.layers.83.mlp.experts.97.gate_proj", "model.layers.83.mlp.experts.98.gate_proj", "model.layers.83.mlp.experts.99.gate_proj", "model.layers.83.mlp.experts.100.gate_proj", "model.layers.83.mlp.experts.101.gate_proj", "model.layers.83.mlp.experts.102.gate_proj", "model.layers.83.mlp.experts.103.gate_proj", "model.layers.83.mlp.experts.104.gate_proj", "model.layers.83.mlp.experts.105.gate_proj", "model.layers.83.mlp.experts.106.gate_proj", "model.layers.83.mlp.experts.107.gate_proj", "model.layers.83.mlp.experts.108.gate_proj", "model.layers.83.mlp.experts.109.gate_proj", "model.layers.83.mlp.experts.110.gate_proj", "model.layers.83.mlp.experts.111.gate_proj", "model.layers.83.mlp.experts.112.gate_proj", "model.layers.83.mlp.experts.113.gate_proj", "model.layers.83.mlp.experts.114.gate_proj", "model.layers.83.mlp.experts.115.gate_proj", "model.layers.83.mlp.experts.116.gate_proj", "model.layers.83.mlp.experts.117.gate_proj", "model.layers.83.mlp.experts.118.gate_proj", "model.layers.83.mlp.experts.119.gate_proj", "model.layers.83.mlp.experts.120.gate_proj", "model.layers.83.mlp.experts.121.gate_proj", "model.layers.83.mlp.experts.122.gate_proj", "model.layers.83.mlp.experts.123.gate_proj", "model.layers.83.mlp.experts.124.gate_proj", "model.layers.83.mlp.experts.125.gate_proj", "model.layers.83.mlp.experts.126.gate_proj", "model.layers.83.mlp.experts.127.gate_proj", "model.layers.83.mlp.experts.128.gate_proj", "model.layers.83.mlp.experts.129.gate_proj", "model.layers.83.mlp.experts.130.gate_proj", "model.layers.83.mlp.experts.131.gate_proj", "model.layers.83.mlp.experts.132.gate_proj", "model.layers.83.mlp.experts.133.gate_proj", "model.layers.83.mlp.experts.134.gate_proj", "model.layers.83.mlp.experts.135.gate_proj", "model.layers.83.mlp.experts.136.gate_proj", "model.layers.83.mlp.experts.137.gate_proj", "model.layers.83.mlp.experts.138.gate_proj", "model.layers.83.mlp.experts.139.gate_proj", "model.layers.83.mlp.experts.140.gate_proj", "model.layers.83.mlp.experts.141.gate_proj", "model.layers.83.mlp.experts.142.gate_proj", "model.layers.83.mlp.experts.143.gate_proj", "model.layers.83.mlp.experts.144.gate_proj", "model.layers.83.mlp.experts.145.gate_proj", "model.layers.83.mlp.experts.146.gate_proj", "model.layers.83.mlp.experts.147.gate_proj", "model.layers.83.mlp.experts.148.gate_proj", "model.layers.83.mlp.experts.149.gate_proj", "model.layers.83.mlp.experts.150.gate_proj", "model.layers.83.mlp.experts.151.gate_proj", "model.layers.83.mlp.experts.152.gate_proj", "model.layers.83.mlp.experts.153.gate_proj", "model.layers.83.mlp.experts.154.gate_proj", "model.layers.83.mlp.experts.155.gate_proj", "model.layers.83.mlp.experts.156.gate_proj", "model.layers.83.mlp.experts.157.gate_proj", "model.layers.83.mlp.experts.158.gate_proj", "model.layers.83.mlp.experts.159.gate_proj", "model.layers.83.mlp.experts.0.up_proj", "model.layers.83.mlp.experts.1.up_proj", "model.layers.83.mlp.experts.2.up_proj", "model.layers.83.mlp.experts.3.up_proj", "model.layers.83.mlp.experts.4.up_proj", "model.layers.83.mlp.experts.5.up_proj", "model.layers.83.mlp.experts.6.up_proj", "model.layers.83.mlp.experts.7.up_proj", "model.layers.83.mlp.experts.8.up_proj", "model.layers.83.mlp.experts.9.up_proj", "model.layers.83.mlp.experts.10.up_proj", "model.layers.83.mlp.experts.11.up_proj", "model.layers.83.mlp.experts.12.up_proj", "model.layers.83.mlp.experts.13.up_proj", "model.layers.83.mlp.experts.14.up_proj", "model.layers.83.mlp.experts.15.up_proj", "model.layers.83.mlp.experts.16.up_proj", "model.layers.83.mlp.experts.17.up_proj", "model.layers.83.mlp.experts.18.up_proj", "model.layers.83.mlp.experts.19.up_proj", "model.layers.83.mlp.experts.20.up_proj", "model.layers.83.mlp.experts.21.up_proj", "model.layers.83.mlp.experts.22.up_proj", "model.layers.83.mlp.experts.23.up_proj", "model.layers.83.mlp.experts.24.up_proj", "model.layers.83.mlp.experts.25.up_proj", "model.layers.83.mlp.experts.26.up_proj", "model.layers.83.mlp.experts.27.up_proj", "model.layers.83.mlp.experts.28.up_proj", "model.layers.83.mlp.experts.29.up_proj", "model.layers.83.mlp.experts.30.up_proj", "model.layers.83.mlp.experts.31.up_proj", "model.layers.83.mlp.experts.32.up_proj", "model.layers.83.mlp.experts.33.up_proj", "model.layers.83.mlp.experts.34.up_proj", "model.layers.83.mlp.experts.35.up_proj", "model.layers.83.mlp.experts.36.up_proj", "model.layers.83.mlp.experts.37.up_proj", "model.layers.83.mlp.experts.38.up_proj", "model.layers.83.mlp.experts.39.up_proj", "model.layers.83.mlp.experts.40.up_proj", "model.layers.83.mlp.experts.41.up_proj", "model.layers.83.mlp.experts.42.up_proj", "model.layers.83.mlp.experts.43.up_proj", "model.layers.83.mlp.experts.44.up_proj", "model.layers.83.mlp.experts.45.up_proj", "model.layers.83.mlp.experts.46.up_proj", "model.layers.83.mlp.experts.47.up_proj", "model.layers.83.mlp.experts.48.up_proj", "model.layers.83.mlp.experts.49.up_proj", "model.layers.83.mlp.experts.50.up_proj", "model.layers.83.mlp.experts.51.up_proj", "model.layers.83.mlp.experts.52.up_proj", "model.layers.83.mlp.experts.53.up_proj", "model.layers.83.mlp.experts.54.up_proj", "model.layers.83.mlp.experts.55.up_proj", "model.layers.83.mlp.experts.56.up_proj", "model.layers.83.mlp.experts.57.up_proj", "model.layers.83.mlp.experts.58.up_proj", "model.layers.83.mlp.experts.59.up_proj", "model.layers.83.mlp.experts.60.up_proj", "model.layers.83.mlp.experts.61.up_proj", "model.layers.83.mlp.experts.62.up_proj", "model.layers.83.mlp.experts.63.up_proj", "model.layers.83.mlp.experts.64.up_proj", "model.layers.83.mlp.experts.65.up_proj", "model.layers.83.mlp.experts.66.up_proj", "model.layers.83.mlp.experts.67.up_proj", "model.layers.83.mlp.experts.68.up_proj", "model.layers.83.mlp.experts.69.up_proj", "model.layers.83.mlp.experts.70.up_proj", "model.layers.83.mlp.experts.71.up_proj", "model.layers.83.mlp.experts.72.up_proj", "model.layers.83.mlp.experts.73.up_proj", "model.layers.83.mlp.experts.74.up_proj", "model.layers.83.mlp.experts.75.up_proj", "model.layers.83.mlp.experts.76.up_proj", "model.layers.83.mlp.experts.77.up_proj", "model.layers.83.mlp.experts.78.up_proj", "model.layers.83.mlp.experts.79.up_proj", "model.layers.83.mlp.experts.80.up_proj", "model.layers.83.mlp.experts.81.up_proj", "model.layers.83.mlp.experts.82.up_proj", "model.layers.83.mlp.experts.83.up_proj", "model.layers.83.mlp.experts.84.up_proj", "model.layers.83.mlp.experts.85.up_proj", "model.layers.83.mlp.experts.86.up_proj", "model.layers.83.mlp.experts.87.up_proj", "model.layers.83.mlp.experts.88.up_proj", "model.layers.83.mlp.experts.89.up_proj", "model.layers.83.mlp.experts.90.up_proj", "model.layers.83.mlp.experts.91.up_proj", "model.layers.83.mlp.experts.92.up_proj", "model.layers.83.mlp.experts.93.up_proj", "model.layers.83.mlp.experts.94.up_proj", "model.layers.83.mlp.experts.95.up_proj", "model.layers.83.mlp.experts.96.up_proj", "model.layers.83.mlp.experts.97.up_proj", "model.layers.83.mlp.experts.98.up_proj", "model.layers.83.mlp.experts.99.up_proj", "model.layers.83.mlp.experts.100.up_proj", "model.layers.83.mlp.experts.101.up_proj", "model.layers.83.mlp.experts.102.up_proj", "model.layers.83.mlp.experts.103.up_proj", "model.layers.83.mlp.experts.104.up_proj", "model.layers.83.mlp.experts.105.up_proj", "model.layers.83.mlp.experts.106.up_proj", "model.layers.83.mlp.experts.107.up_proj", "model.layers.83.mlp.experts.108.up_proj", "model.layers.83.mlp.experts.109.up_proj", "model.layers.83.mlp.experts.110.up_proj", "model.layers.83.mlp.experts.111.up_proj", "model.layers.83.mlp.experts.112.up_proj", "model.layers.83.mlp.experts.113.up_proj", "model.layers.83.mlp.experts.114.up_proj", "model.layers.83.mlp.experts.115.up_proj", "model.layers.83.mlp.experts.116.up_proj", "model.layers.83.mlp.experts.117.up_proj", "model.layers.83.mlp.experts.118.up_proj", "model.layers.83.mlp.experts.119.up_proj", "model.layers.83.mlp.experts.120.up_proj", "model.layers.83.mlp.experts.121.up_proj", "model.layers.83.mlp.experts.122.up_proj", "model.layers.83.mlp.experts.123.up_proj", "model.layers.83.mlp.experts.124.up_proj", "model.layers.83.mlp.experts.125.up_proj", "model.layers.83.mlp.experts.126.up_proj", "model.layers.83.mlp.experts.127.up_proj", "model.layers.83.mlp.experts.128.up_proj", "model.layers.83.mlp.experts.129.up_proj", "model.layers.83.mlp.experts.130.up_proj", "model.layers.83.mlp.experts.131.up_proj", "model.layers.83.mlp.experts.132.up_proj", "model.layers.83.mlp.experts.133.up_proj", "model.layers.83.mlp.experts.134.up_proj", "model.layers.83.mlp.experts.135.up_proj", "model.layers.83.mlp.experts.136.up_proj", "model.layers.83.mlp.experts.137.up_proj", "model.layers.83.mlp.experts.138.up_proj", "model.layers.83.mlp.experts.139.up_proj", "model.layers.83.mlp.experts.140.up_proj", "model.layers.83.mlp.experts.141.up_proj", "model.layers.83.mlp.experts.142.up_proj", "model.layers.83.mlp.experts.143.up_proj", "model.layers.83.mlp.experts.144.up_proj", "model.layers.83.mlp.experts.145.up_proj", "model.layers.83.mlp.experts.146.up_proj", "model.layers.83.mlp.experts.147.up_proj", "model.layers.83.mlp.experts.148.up_proj", "model.layers.83.mlp.experts.149.up_proj", "model.layers.83.mlp.experts.150.up_proj", "model.layers.83.mlp.experts.151.up_proj", "model.layers.83.mlp.experts.152.up_proj", "model.layers.83.mlp.experts.153.up_proj", "model.layers.83.mlp.experts.154.up_proj", "model.layers.83.mlp.experts.155.up_proj", "model.layers.83.mlp.experts.156.up_proj", "model.layers.83.mlp.experts.157.up_proj", "model.layers.83.mlp.experts.158.up_proj", "model.layers.83.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00020489096641540527, "dbits": 2516582400 }, { "dkld": -0.00018798438832164088, "dbits": 5033164800 }, { "dkld": -0.00023943819105626263, "dbits": 7549747200 }, { "dkld": -0.0002290815114975031, "dbits": 12582912000 } ] }, { "idx": 500, "layers": [ "model.layers.83.mlp.experts.0.down_proj", "model.layers.83.mlp.experts.1.down_proj", "model.layers.83.mlp.experts.2.down_proj", "model.layers.83.mlp.experts.3.down_proj", "model.layers.83.mlp.experts.4.down_proj", "model.layers.83.mlp.experts.5.down_proj", "model.layers.83.mlp.experts.6.down_proj", "model.layers.83.mlp.experts.7.down_proj", "model.layers.83.mlp.experts.8.down_proj", "model.layers.83.mlp.experts.9.down_proj", "model.layers.83.mlp.experts.10.down_proj", "model.layers.83.mlp.experts.11.down_proj", "model.layers.83.mlp.experts.12.down_proj", "model.layers.83.mlp.experts.13.down_proj", "model.layers.83.mlp.experts.14.down_proj", "model.layers.83.mlp.experts.15.down_proj", "model.layers.83.mlp.experts.16.down_proj", "model.layers.83.mlp.experts.17.down_proj", "model.layers.83.mlp.experts.18.down_proj", "model.layers.83.mlp.experts.19.down_proj", "model.layers.83.mlp.experts.20.down_proj", "model.layers.83.mlp.experts.21.down_proj", "model.layers.83.mlp.experts.22.down_proj", "model.layers.83.mlp.experts.23.down_proj", "model.layers.83.mlp.experts.24.down_proj", "model.layers.83.mlp.experts.25.down_proj", "model.layers.83.mlp.experts.26.down_proj", "model.layers.83.mlp.experts.27.down_proj", "model.layers.83.mlp.experts.28.down_proj", "model.layers.83.mlp.experts.29.down_proj", "model.layers.83.mlp.experts.30.down_proj", "model.layers.83.mlp.experts.31.down_proj", "model.layers.83.mlp.experts.32.down_proj", "model.layers.83.mlp.experts.33.down_proj", "model.layers.83.mlp.experts.34.down_proj", "model.layers.83.mlp.experts.35.down_proj", "model.layers.83.mlp.experts.36.down_proj", "model.layers.83.mlp.experts.37.down_proj", "model.layers.83.mlp.experts.38.down_proj", "model.layers.83.mlp.experts.39.down_proj", "model.layers.83.mlp.experts.40.down_proj", "model.layers.83.mlp.experts.41.down_proj", "model.layers.83.mlp.experts.42.down_proj", "model.layers.83.mlp.experts.43.down_proj", "model.layers.83.mlp.experts.44.down_proj", "model.layers.83.mlp.experts.45.down_proj", "model.layers.83.mlp.experts.46.down_proj", "model.layers.83.mlp.experts.47.down_proj", "model.layers.83.mlp.experts.48.down_proj", "model.layers.83.mlp.experts.49.down_proj", "model.layers.83.mlp.experts.50.down_proj", "model.layers.83.mlp.experts.51.down_proj", "model.layers.83.mlp.experts.52.down_proj", "model.layers.83.mlp.experts.53.down_proj", "model.layers.83.mlp.experts.54.down_proj", "model.layers.83.mlp.experts.55.down_proj", "model.layers.83.mlp.experts.56.down_proj", "model.layers.83.mlp.experts.57.down_proj", "model.layers.83.mlp.experts.58.down_proj", "model.layers.83.mlp.experts.59.down_proj", "model.layers.83.mlp.experts.60.down_proj", "model.layers.83.mlp.experts.61.down_proj", "model.layers.83.mlp.experts.62.down_proj", "model.layers.83.mlp.experts.63.down_proj", "model.layers.83.mlp.experts.64.down_proj", "model.layers.83.mlp.experts.65.down_proj", "model.layers.83.mlp.experts.66.down_proj", "model.layers.83.mlp.experts.67.down_proj", "model.layers.83.mlp.experts.68.down_proj", "model.layers.83.mlp.experts.69.down_proj", "model.layers.83.mlp.experts.70.down_proj", "model.layers.83.mlp.experts.71.down_proj", "model.layers.83.mlp.experts.72.down_proj", "model.layers.83.mlp.experts.73.down_proj", "model.layers.83.mlp.experts.74.down_proj", "model.layers.83.mlp.experts.75.down_proj", "model.layers.83.mlp.experts.76.down_proj", "model.layers.83.mlp.experts.77.down_proj", "model.layers.83.mlp.experts.78.down_proj", "model.layers.83.mlp.experts.79.down_proj", "model.layers.83.mlp.experts.80.down_proj", "model.layers.83.mlp.experts.81.down_proj", "model.layers.83.mlp.experts.82.down_proj", "model.layers.83.mlp.experts.83.down_proj", "model.layers.83.mlp.experts.84.down_proj", "model.layers.83.mlp.experts.85.down_proj", "model.layers.83.mlp.experts.86.down_proj", "model.layers.83.mlp.experts.87.down_proj", "model.layers.83.mlp.experts.88.down_proj", "model.layers.83.mlp.experts.89.down_proj", "model.layers.83.mlp.experts.90.down_proj", "model.layers.83.mlp.experts.91.down_proj", "model.layers.83.mlp.experts.92.down_proj", "model.layers.83.mlp.experts.93.down_proj", "model.layers.83.mlp.experts.94.down_proj", "model.layers.83.mlp.experts.95.down_proj", "model.layers.83.mlp.experts.96.down_proj", "model.layers.83.mlp.experts.97.down_proj", "model.layers.83.mlp.experts.98.down_proj", "model.layers.83.mlp.experts.99.down_proj", "model.layers.83.mlp.experts.100.down_proj", "model.layers.83.mlp.experts.101.down_proj", "model.layers.83.mlp.experts.102.down_proj", "model.layers.83.mlp.experts.103.down_proj", "model.layers.83.mlp.experts.104.down_proj", "model.layers.83.mlp.experts.105.down_proj", "model.layers.83.mlp.experts.106.down_proj", "model.layers.83.mlp.experts.107.down_proj", "model.layers.83.mlp.experts.108.down_proj", "model.layers.83.mlp.experts.109.down_proj", "model.layers.83.mlp.experts.110.down_proj", "model.layers.83.mlp.experts.111.down_proj", "model.layers.83.mlp.experts.112.down_proj", "model.layers.83.mlp.experts.113.down_proj", "model.layers.83.mlp.experts.114.down_proj", "model.layers.83.mlp.experts.115.down_proj", "model.layers.83.mlp.experts.116.down_proj", "model.layers.83.mlp.experts.117.down_proj", "model.layers.83.mlp.experts.118.down_proj", "model.layers.83.mlp.experts.119.down_proj", "model.layers.83.mlp.experts.120.down_proj", "model.layers.83.mlp.experts.121.down_proj", "model.layers.83.mlp.experts.122.down_proj", "model.layers.83.mlp.experts.123.down_proj", "model.layers.83.mlp.experts.124.down_proj", "model.layers.83.mlp.experts.125.down_proj", "model.layers.83.mlp.experts.126.down_proj", "model.layers.83.mlp.experts.127.down_proj", "model.layers.83.mlp.experts.128.down_proj", "model.layers.83.mlp.experts.129.down_proj", "model.layers.83.mlp.experts.130.down_proj", "model.layers.83.mlp.experts.131.down_proj", "model.layers.83.mlp.experts.132.down_proj", "model.layers.83.mlp.experts.133.down_proj", "model.layers.83.mlp.experts.134.down_proj", "model.layers.83.mlp.experts.135.down_proj", "model.layers.83.mlp.experts.136.down_proj", "model.layers.83.mlp.experts.137.down_proj", "model.layers.83.mlp.experts.138.down_proj", "model.layers.83.mlp.experts.139.down_proj", "model.layers.83.mlp.experts.140.down_proj", "model.layers.83.mlp.experts.141.down_proj", "model.layers.83.mlp.experts.142.down_proj", "model.layers.83.mlp.experts.143.down_proj", "model.layers.83.mlp.experts.144.down_proj", "model.layers.83.mlp.experts.145.down_proj", "model.layers.83.mlp.experts.146.down_proj", "model.layers.83.mlp.experts.147.down_proj", "model.layers.83.mlp.experts.148.down_proj", "model.layers.83.mlp.experts.149.down_proj", "model.layers.83.mlp.experts.150.down_proj", "model.layers.83.mlp.experts.151.down_proj", "model.layers.83.mlp.experts.152.down_proj", "model.layers.83.mlp.experts.153.down_proj", "model.layers.83.mlp.experts.154.down_proj", "model.layers.83.mlp.experts.155.down_proj", "model.layers.83.mlp.experts.156.down_proj", "model.layers.83.mlp.experts.157.down_proj", "model.layers.83.mlp.experts.158.down_proj", "model.layers.83.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -6.367312744259834e-05, "dbits": 1258291200 }, { "dkld": -0.00012175459414720535, "dbits": 2516582400 }, { "dkld": -0.00014319140464068292, "dbits": 3774873600 }, { "dkld": -0.00012857960537077107, "dbits": 6291456000 } ] }, { "idx": 501, "layers": [ "model.layers.84.self_attn.q_proj" ], "candidates": [ { "dkld": 5.2102841436862946e-05, "dbits": 62914560 }, { "dkld": -0.00013569705188275216, "dbits": 125829120 }, { "dkld": -0.0002622959204018199, "dbits": 188743680 }, { "dkld": -0.00014494732022285461, "dbits": 314572800 } ] }, { "idx": 502, "layers": [ "model.layers.84.self_attn.k_proj", "model.layers.84.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0003614991903305026, "dbits": 10485760 }, { "dkld": -3.773644566536227e-05, "dbits": 20971520 }, { "dkld": 0.00014516916126011692, "dbits": 31457280 }, { "dkld": 0.00020771119743584476, "dbits": 52428800 } ] }, { "idx": 503, "layers": [ "model.layers.84.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0002981549128890093, "dbits": 62914560 }, { "dkld": -7.703136652709164e-05, "dbits": 125829120 }, { "dkld": -0.00010654367506504614, "dbits": 188743680 }, { "dkld": -5.030576139689047e-05, "dbits": 314572800 } ] }, { "idx": 504, "layers": [ "model.layers.84.mlp.shared_experts.gate_proj", "model.layers.84.mlp.shared_experts.up_proj", "model.layers.84.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.000596580654382714, "dbits": 23592960 }, { "dkld": -0.0003297368064522771, "dbits": 47185920 }, { "dkld": -0.0002907596528530232, "dbits": 70778880 }, { "dkld": -0.0003739383071661023, "dbits": 117964800 } ] }, { "idx": 505, "layers": [ "model.layers.84.mlp.experts.0.gate_proj", "model.layers.84.mlp.experts.1.gate_proj", "model.layers.84.mlp.experts.2.gate_proj", "model.layers.84.mlp.experts.3.gate_proj", "model.layers.84.mlp.experts.4.gate_proj", "model.layers.84.mlp.experts.5.gate_proj", "model.layers.84.mlp.experts.6.gate_proj", "model.layers.84.mlp.experts.7.gate_proj", "model.layers.84.mlp.experts.8.gate_proj", "model.layers.84.mlp.experts.9.gate_proj", "model.layers.84.mlp.experts.10.gate_proj", "model.layers.84.mlp.experts.11.gate_proj", "model.layers.84.mlp.experts.12.gate_proj", "model.layers.84.mlp.experts.13.gate_proj", "model.layers.84.mlp.experts.14.gate_proj", "model.layers.84.mlp.experts.15.gate_proj", "model.layers.84.mlp.experts.16.gate_proj", "model.layers.84.mlp.experts.17.gate_proj", "model.layers.84.mlp.experts.18.gate_proj", "model.layers.84.mlp.experts.19.gate_proj", "model.layers.84.mlp.experts.20.gate_proj", "model.layers.84.mlp.experts.21.gate_proj", "model.layers.84.mlp.experts.22.gate_proj", "model.layers.84.mlp.experts.23.gate_proj", "model.layers.84.mlp.experts.24.gate_proj", "model.layers.84.mlp.experts.25.gate_proj", "model.layers.84.mlp.experts.26.gate_proj", "model.layers.84.mlp.experts.27.gate_proj", "model.layers.84.mlp.experts.28.gate_proj", "model.layers.84.mlp.experts.29.gate_proj", "model.layers.84.mlp.experts.30.gate_proj", "model.layers.84.mlp.experts.31.gate_proj", "model.layers.84.mlp.experts.32.gate_proj", "model.layers.84.mlp.experts.33.gate_proj", "model.layers.84.mlp.experts.34.gate_proj", "model.layers.84.mlp.experts.35.gate_proj", "model.layers.84.mlp.experts.36.gate_proj", "model.layers.84.mlp.experts.37.gate_proj", "model.layers.84.mlp.experts.38.gate_proj", "model.layers.84.mlp.experts.39.gate_proj", "model.layers.84.mlp.experts.40.gate_proj", "model.layers.84.mlp.experts.41.gate_proj", "model.layers.84.mlp.experts.42.gate_proj", "model.layers.84.mlp.experts.43.gate_proj", "model.layers.84.mlp.experts.44.gate_proj", "model.layers.84.mlp.experts.45.gate_proj", "model.layers.84.mlp.experts.46.gate_proj", "model.layers.84.mlp.experts.47.gate_proj", "model.layers.84.mlp.experts.48.gate_proj", "model.layers.84.mlp.experts.49.gate_proj", "model.layers.84.mlp.experts.50.gate_proj", "model.layers.84.mlp.experts.51.gate_proj", "model.layers.84.mlp.experts.52.gate_proj", "model.layers.84.mlp.experts.53.gate_proj", "model.layers.84.mlp.experts.54.gate_proj", "model.layers.84.mlp.experts.55.gate_proj", "model.layers.84.mlp.experts.56.gate_proj", "model.layers.84.mlp.experts.57.gate_proj", "model.layers.84.mlp.experts.58.gate_proj", "model.layers.84.mlp.experts.59.gate_proj", "model.layers.84.mlp.experts.60.gate_proj", "model.layers.84.mlp.experts.61.gate_proj", "model.layers.84.mlp.experts.62.gate_proj", "model.layers.84.mlp.experts.63.gate_proj", "model.layers.84.mlp.experts.64.gate_proj", "model.layers.84.mlp.experts.65.gate_proj", "model.layers.84.mlp.experts.66.gate_proj", "model.layers.84.mlp.experts.67.gate_proj", "model.layers.84.mlp.experts.68.gate_proj", "model.layers.84.mlp.experts.69.gate_proj", "model.layers.84.mlp.experts.70.gate_proj", "model.layers.84.mlp.experts.71.gate_proj", "model.layers.84.mlp.experts.72.gate_proj", "model.layers.84.mlp.experts.73.gate_proj", "model.layers.84.mlp.experts.74.gate_proj", "model.layers.84.mlp.experts.75.gate_proj", "model.layers.84.mlp.experts.76.gate_proj", "model.layers.84.mlp.experts.77.gate_proj", "model.layers.84.mlp.experts.78.gate_proj", "model.layers.84.mlp.experts.79.gate_proj", "model.layers.84.mlp.experts.80.gate_proj", "model.layers.84.mlp.experts.81.gate_proj", "model.layers.84.mlp.experts.82.gate_proj", "model.layers.84.mlp.experts.83.gate_proj", "model.layers.84.mlp.experts.84.gate_proj", "model.layers.84.mlp.experts.85.gate_proj", "model.layers.84.mlp.experts.86.gate_proj", "model.layers.84.mlp.experts.87.gate_proj", "model.layers.84.mlp.experts.88.gate_proj", "model.layers.84.mlp.experts.89.gate_proj", "model.layers.84.mlp.experts.90.gate_proj", "model.layers.84.mlp.experts.91.gate_proj", "model.layers.84.mlp.experts.92.gate_proj", "model.layers.84.mlp.experts.93.gate_proj", "model.layers.84.mlp.experts.94.gate_proj", "model.layers.84.mlp.experts.95.gate_proj", "model.layers.84.mlp.experts.96.gate_proj", "model.layers.84.mlp.experts.97.gate_proj", "model.layers.84.mlp.experts.98.gate_proj", "model.layers.84.mlp.experts.99.gate_proj", "model.layers.84.mlp.experts.100.gate_proj", "model.layers.84.mlp.experts.101.gate_proj", "model.layers.84.mlp.experts.102.gate_proj", "model.layers.84.mlp.experts.103.gate_proj", "model.layers.84.mlp.experts.104.gate_proj", "model.layers.84.mlp.experts.105.gate_proj", "model.layers.84.mlp.experts.106.gate_proj", "model.layers.84.mlp.experts.107.gate_proj", "model.layers.84.mlp.experts.108.gate_proj", "model.layers.84.mlp.experts.109.gate_proj", "model.layers.84.mlp.experts.110.gate_proj", "model.layers.84.mlp.experts.111.gate_proj", "model.layers.84.mlp.experts.112.gate_proj", "model.layers.84.mlp.experts.113.gate_proj", "model.layers.84.mlp.experts.114.gate_proj", "model.layers.84.mlp.experts.115.gate_proj", "model.layers.84.mlp.experts.116.gate_proj", "model.layers.84.mlp.experts.117.gate_proj", "model.layers.84.mlp.experts.118.gate_proj", "model.layers.84.mlp.experts.119.gate_proj", "model.layers.84.mlp.experts.120.gate_proj", "model.layers.84.mlp.experts.121.gate_proj", "model.layers.84.mlp.experts.122.gate_proj", "model.layers.84.mlp.experts.123.gate_proj", "model.layers.84.mlp.experts.124.gate_proj", "model.layers.84.mlp.experts.125.gate_proj", "model.layers.84.mlp.experts.126.gate_proj", "model.layers.84.mlp.experts.127.gate_proj", "model.layers.84.mlp.experts.128.gate_proj", "model.layers.84.mlp.experts.129.gate_proj", "model.layers.84.mlp.experts.130.gate_proj", "model.layers.84.mlp.experts.131.gate_proj", "model.layers.84.mlp.experts.132.gate_proj", "model.layers.84.mlp.experts.133.gate_proj", "model.layers.84.mlp.experts.134.gate_proj", "model.layers.84.mlp.experts.135.gate_proj", "model.layers.84.mlp.experts.136.gate_proj", "model.layers.84.mlp.experts.137.gate_proj", "model.layers.84.mlp.experts.138.gate_proj", "model.layers.84.mlp.experts.139.gate_proj", "model.layers.84.mlp.experts.140.gate_proj", "model.layers.84.mlp.experts.141.gate_proj", "model.layers.84.mlp.experts.142.gate_proj", "model.layers.84.mlp.experts.143.gate_proj", "model.layers.84.mlp.experts.144.gate_proj", "model.layers.84.mlp.experts.145.gate_proj", "model.layers.84.mlp.experts.146.gate_proj", "model.layers.84.mlp.experts.147.gate_proj", "model.layers.84.mlp.experts.148.gate_proj", "model.layers.84.mlp.experts.149.gate_proj", "model.layers.84.mlp.experts.150.gate_proj", "model.layers.84.mlp.experts.151.gate_proj", "model.layers.84.mlp.experts.152.gate_proj", "model.layers.84.mlp.experts.153.gate_proj", "model.layers.84.mlp.experts.154.gate_proj", "model.layers.84.mlp.experts.155.gate_proj", "model.layers.84.mlp.experts.156.gate_proj", "model.layers.84.mlp.experts.157.gate_proj", "model.layers.84.mlp.experts.158.gate_proj", "model.layers.84.mlp.experts.159.gate_proj", "model.layers.84.mlp.experts.0.up_proj", "model.layers.84.mlp.experts.1.up_proj", "model.layers.84.mlp.experts.2.up_proj", "model.layers.84.mlp.experts.3.up_proj", "model.layers.84.mlp.experts.4.up_proj", "model.layers.84.mlp.experts.5.up_proj", "model.layers.84.mlp.experts.6.up_proj", "model.layers.84.mlp.experts.7.up_proj", "model.layers.84.mlp.experts.8.up_proj", "model.layers.84.mlp.experts.9.up_proj", "model.layers.84.mlp.experts.10.up_proj", "model.layers.84.mlp.experts.11.up_proj", "model.layers.84.mlp.experts.12.up_proj", "model.layers.84.mlp.experts.13.up_proj", "model.layers.84.mlp.experts.14.up_proj", "model.layers.84.mlp.experts.15.up_proj", "model.layers.84.mlp.experts.16.up_proj", "model.layers.84.mlp.experts.17.up_proj", "model.layers.84.mlp.experts.18.up_proj", "model.layers.84.mlp.experts.19.up_proj", "model.layers.84.mlp.experts.20.up_proj", "model.layers.84.mlp.experts.21.up_proj", "model.layers.84.mlp.experts.22.up_proj", "model.layers.84.mlp.experts.23.up_proj", "model.layers.84.mlp.experts.24.up_proj", "model.layers.84.mlp.experts.25.up_proj", "model.layers.84.mlp.experts.26.up_proj", "model.layers.84.mlp.experts.27.up_proj", "model.layers.84.mlp.experts.28.up_proj", "model.layers.84.mlp.experts.29.up_proj", "model.layers.84.mlp.experts.30.up_proj", "model.layers.84.mlp.experts.31.up_proj", "model.layers.84.mlp.experts.32.up_proj", "model.layers.84.mlp.experts.33.up_proj", "model.layers.84.mlp.experts.34.up_proj", "model.layers.84.mlp.experts.35.up_proj", "model.layers.84.mlp.experts.36.up_proj", "model.layers.84.mlp.experts.37.up_proj", "model.layers.84.mlp.experts.38.up_proj", "model.layers.84.mlp.experts.39.up_proj", "model.layers.84.mlp.experts.40.up_proj", "model.layers.84.mlp.experts.41.up_proj", "model.layers.84.mlp.experts.42.up_proj", "model.layers.84.mlp.experts.43.up_proj", "model.layers.84.mlp.experts.44.up_proj", "model.layers.84.mlp.experts.45.up_proj", "model.layers.84.mlp.experts.46.up_proj", "model.layers.84.mlp.experts.47.up_proj", "model.layers.84.mlp.experts.48.up_proj", "model.layers.84.mlp.experts.49.up_proj", "model.layers.84.mlp.experts.50.up_proj", "model.layers.84.mlp.experts.51.up_proj", "model.layers.84.mlp.experts.52.up_proj", "model.layers.84.mlp.experts.53.up_proj", "model.layers.84.mlp.experts.54.up_proj", "model.layers.84.mlp.experts.55.up_proj", "model.layers.84.mlp.experts.56.up_proj", "model.layers.84.mlp.experts.57.up_proj", "model.layers.84.mlp.experts.58.up_proj", "model.layers.84.mlp.experts.59.up_proj", "model.layers.84.mlp.experts.60.up_proj", "model.layers.84.mlp.experts.61.up_proj", "model.layers.84.mlp.experts.62.up_proj", "model.layers.84.mlp.experts.63.up_proj", "model.layers.84.mlp.experts.64.up_proj", "model.layers.84.mlp.experts.65.up_proj", "model.layers.84.mlp.experts.66.up_proj", "model.layers.84.mlp.experts.67.up_proj", "model.layers.84.mlp.experts.68.up_proj", "model.layers.84.mlp.experts.69.up_proj", "model.layers.84.mlp.experts.70.up_proj", "model.layers.84.mlp.experts.71.up_proj", "model.layers.84.mlp.experts.72.up_proj", "model.layers.84.mlp.experts.73.up_proj", "model.layers.84.mlp.experts.74.up_proj", "model.layers.84.mlp.experts.75.up_proj", "model.layers.84.mlp.experts.76.up_proj", "model.layers.84.mlp.experts.77.up_proj", "model.layers.84.mlp.experts.78.up_proj", "model.layers.84.mlp.experts.79.up_proj", "model.layers.84.mlp.experts.80.up_proj", "model.layers.84.mlp.experts.81.up_proj", "model.layers.84.mlp.experts.82.up_proj", "model.layers.84.mlp.experts.83.up_proj", "model.layers.84.mlp.experts.84.up_proj", "model.layers.84.mlp.experts.85.up_proj", "model.layers.84.mlp.experts.86.up_proj", "model.layers.84.mlp.experts.87.up_proj", "model.layers.84.mlp.experts.88.up_proj", "model.layers.84.mlp.experts.89.up_proj", "model.layers.84.mlp.experts.90.up_proj", "model.layers.84.mlp.experts.91.up_proj", "model.layers.84.mlp.experts.92.up_proj", "model.layers.84.mlp.experts.93.up_proj", "model.layers.84.mlp.experts.94.up_proj", "model.layers.84.mlp.experts.95.up_proj", "model.layers.84.mlp.experts.96.up_proj", "model.layers.84.mlp.experts.97.up_proj", "model.layers.84.mlp.experts.98.up_proj", "model.layers.84.mlp.experts.99.up_proj", "model.layers.84.mlp.experts.100.up_proj", "model.layers.84.mlp.experts.101.up_proj", "model.layers.84.mlp.experts.102.up_proj", "model.layers.84.mlp.experts.103.up_proj", "model.layers.84.mlp.experts.104.up_proj", "model.layers.84.mlp.experts.105.up_proj", "model.layers.84.mlp.experts.106.up_proj", "model.layers.84.mlp.experts.107.up_proj", "model.layers.84.mlp.experts.108.up_proj", "model.layers.84.mlp.experts.109.up_proj", "model.layers.84.mlp.experts.110.up_proj", "model.layers.84.mlp.experts.111.up_proj", "model.layers.84.mlp.experts.112.up_proj", "model.layers.84.mlp.experts.113.up_proj", "model.layers.84.mlp.experts.114.up_proj", "model.layers.84.mlp.experts.115.up_proj", "model.layers.84.mlp.experts.116.up_proj", "model.layers.84.mlp.experts.117.up_proj", "model.layers.84.mlp.experts.118.up_proj", "model.layers.84.mlp.experts.119.up_proj", "model.layers.84.mlp.experts.120.up_proj", "model.layers.84.mlp.experts.121.up_proj", "model.layers.84.mlp.experts.122.up_proj", "model.layers.84.mlp.experts.123.up_proj", "model.layers.84.mlp.experts.124.up_proj", "model.layers.84.mlp.experts.125.up_proj", "model.layers.84.mlp.experts.126.up_proj", "model.layers.84.mlp.experts.127.up_proj", "model.layers.84.mlp.experts.128.up_proj", "model.layers.84.mlp.experts.129.up_proj", "model.layers.84.mlp.experts.130.up_proj", "model.layers.84.mlp.experts.131.up_proj", "model.layers.84.mlp.experts.132.up_proj", "model.layers.84.mlp.experts.133.up_proj", "model.layers.84.mlp.experts.134.up_proj", "model.layers.84.mlp.experts.135.up_proj", "model.layers.84.mlp.experts.136.up_proj", "model.layers.84.mlp.experts.137.up_proj", "model.layers.84.mlp.experts.138.up_proj", "model.layers.84.mlp.experts.139.up_proj", "model.layers.84.mlp.experts.140.up_proj", "model.layers.84.mlp.experts.141.up_proj", "model.layers.84.mlp.experts.142.up_proj", "model.layers.84.mlp.experts.143.up_proj", "model.layers.84.mlp.experts.144.up_proj", "model.layers.84.mlp.experts.145.up_proj", "model.layers.84.mlp.experts.146.up_proj", "model.layers.84.mlp.experts.147.up_proj", "model.layers.84.mlp.experts.148.up_proj", "model.layers.84.mlp.experts.149.up_proj", "model.layers.84.mlp.experts.150.up_proj", "model.layers.84.mlp.experts.151.up_proj", "model.layers.84.mlp.experts.152.up_proj", "model.layers.84.mlp.experts.153.up_proj", "model.layers.84.mlp.experts.154.up_proj", "model.layers.84.mlp.experts.155.up_proj", "model.layers.84.mlp.experts.156.up_proj", "model.layers.84.mlp.experts.157.up_proj", "model.layers.84.mlp.experts.158.up_proj", "model.layers.84.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00016990825533867715, "dbits": 2516582400 }, { "dkld": -0.0001757466234266758, "dbits": 5033164800 }, { "dkld": -0.00013933852314949868, "dbits": 7549747200 }, { "dkld": -0.0001311083324253559, "dbits": 12582912000 } ] }, { "idx": 506, "layers": [ "model.layers.84.mlp.experts.0.down_proj", "model.layers.84.mlp.experts.1.down_proj", "model.layers.84.mlp.experts.2.down_proj", "model.layers.84.mlp.experts.3.down_proj", "model.layers.84.mlp.experts.4.down_proj", "model.layers.84.mlp.experts.5.down_proj", "model.layers.84.mlp.experts.6.down_proj", "model.layers.84.mlp.experts.7.down_proj", "model.layers.84.mlp.experts.8.down_proj", "model.layers.84.mlp.experts.9.down_proj", "model.layers.84.mlp.experts.10.down_proj", "model.layers.84.mlp.experts.11.down_proj", "model.layers.84.mlp.experts.12.down_proj", "model.layers.84.mlp.experts.13.down_proj", "model.layers.84.mlp.experts.14.down_proj", "model.layers.84.mlp.experts.15.down_proj", "model.layers.84.mlp.experts.16.down_proj", "model.layers.84.mlp.experts.17.down_proj", "model.layers.84.mlp.experts.18.down_proj", "model.layers.84.mlp.experts.19.down_proj", "model.layers.84.mlp.experts.20.down_proj", "model.layers.84.mlp.experts.21.down_proj", "model.layers.84.mlp.experts.22.down_proj", "model.layers.84.mlp.experts.23.down_proj", "model.layers.84.mlp.experts.24.down_proj", "model.layers.84.mlp.experts.25.down_proj", "model.layers.84.mlp.experts.26.down_proj", "model.layers.84.mlp.experts.27.down_proj", "model.layers.84.mlp.experts.28.down_proj", "model.layers.84.mlp.experts.29.down_proj", "model.layers.84.mlp.experts.30.down_proj", "model.layers.84.mlp.experts.31.down_proj", "model.layers.84.mlp.experts.32.down_proj", "model.layers.84.mlp.experts.33.down_proj", "model.layers.84.mlp.experts.34.down_proj", "model.layers.84.mlp.experts.35.down_proj", "model.layers.84.mlp.experts.36.down_proj", "model.layers.84.mlp.experts.37.down_proj", "model.layers.84.mlp.experts.38.down_proj", "model.layers.84.mlp.experts.39.down_proj", "model.layers.84.mlp.experts.40.down_proj", "model.layers.84.mlp.experts.41.down_proj", "model.layers.84.mlp.experts.42.down_proj", "model.layers.84.mlp.experts.43.down_proj", "model.layers.84.mlp.experts.44.down_proj", "model.layers.84.mlp.experts.45.down_proj", "model.layers.84.mlp.experts.46.down_proj", "model.layers.84.mlp.experts.47.down_proj", "model.layers.84.mlp.experts.48.down_proj", "model.layers.84.mlp.experts.49.down_proj", "model.layers.84.mlp.experts.50.down_proj", "model.layers.84.mlp.experts.51.down_proj", "model.layers.84.mlp.experts.52.down_proj", "model.layers.84.mlp.experts.53.down_proj", "model.layers.84.mlp.experts.54.down_proj", "model.layers.84.mlp.experts.55.down_proj", "model.layers.84.mlp.experts.56.down_proj", "model.layers.84.mlp.experts.57.down_proj", "model.layers.84.mlp.experts.58.down_proj", "model.layers.84.mlp.experts.59.down_proj", "model.layers.84.mlp.experts.60.down_proj", "model.layers.84.mlp.experts.61.down_proj", "model.layers.84.mlp.experts.62.down_proj", "model.layers.84.mlp.experts.63.down_proj", "model.layers.84.mlp.experts.64.down_proj", "model.layers.84.mlp.experts.65.down_proj", "model.layers.84.mlp.experts.66.down_proj", "model.layers.84.mlp.experts.67.down_proj", "model.layers.84.mlp.experts.68.down_proj", "model.layers.84.mlp.experts.69.down_proj", "model.layers.84.mlp.experts.70.down_proj", "model.layers.84.mlp.experts.71.down_proj", "model.layers.84.mlp.experts.72.down_proj", "model.layers.84.mlp.experts.73.down_proj", "model.layers.84.mlp.experts.74.down_proj", "model.layers.84.mlp.experts.75.down_proj", "model.layers.84.mlp.experts.76.down_proj", "model.layers.84.mlp.experts.77.down_proj", "model.layers.84.mlp.experts.78.down_proj", "model.layers.84.mlp.experts.79.down_proj", "model.layers.84.mlp.experts.80.down_proj", "model.layers.84.mlp.experts.81.down_proj", "model.layers.84.mlp.experts.82.down_proj", "model.layers.84.mlp.experts.83.down_proj", "model.layers.84.mlp.experts.84.down_proj", "model.layers.84.mlp.experts.85.down_proj", "model.layers.84.mlp.experts.86.down_proj", "model.layers.84.mlp.experts.87.down_proj", "model.layers.84.mlp.experts.88.down_proj", "model.layers.84.mlp.experts.89.down_proj", "model.layers.84.mlp.experts.90.down_proj", "model.layers.84.mlp.experts.91.down_proj", "model.layers.84.mlp.experts.92.down_proj", "model.layers.84.mlp.experts.93.down_proj", "model.layers.84.mlp.experts.94.down_proj", "model.layers.84.mlp.experts.95.down_proj", "model.layers.84.mlp.experts.96.down_proj", "model.layers.84.mlp.experts.97.down_proj", "model.layers.84.mlp.experts.98.down_proj", "model.layers.84.mlp.experts.99.down_proj", "model.layers.84.mlp.experts.100.down_proj", "model.layers.84.mlp.experts.101.down_proj", "model.layers.84.mlp.experts.102.down_proj", "model.layers.84.mlp.experts.103.down_proj", "model.layers.84.mlp.experts.104.down_proj", "model.layers.84.mlp.experts.105.down_proj", "model.layers.84.mlp.experts.106.down_proj", "model.layers.84.mlp.experts.107.down_proj", "model.layers.84.mlp.experts.108.down_proj", "model.layers.84.mlp.experts.109.down_proj", "model.layers.84.mlp.experts.110.down_proj", "model.layers.84.mlp.experts.111.down_proj", "model.layers.84.mlp.experts.112.down_proj", "model.layers.84.mlp.experts.113.down_proj", "model.layers.84.mlp.experts.114.down_proj", "model.layers.84.mlp.experts.115.down_proj", "model.layers.84.mlp.experts.116.down_proj", "model.layers.84.mlp.experts.117.down_proj", "model.layers.84.mlp.experts.118.down_proj", "model.layers.84.mlp.experts.119.down_proj", "model.layers.84.mlp.experts.120.down_proj", "model.layers.84.mlp.experts.121.down_proj", "model.layers.84.mlp.experts.122.down_proj", "model.layers.84.mlp.experts.123.down_proj", "model.layers.84.mlp.experts.124.down_proj", "model.layers.84.mlp.experts.125.down_proj", "model.layers.84.mlp.experts.126.down_proj", "model.layers.84.mlp.experts.127.down_proj", "model.layers.84.mlp.experts.128.down_proj", "model.layers.84.mlp.experts.129.down_proj", "model.layers.84.mlp.experts.130.down_proj", "model.layers.84.mlp.experts.131.down_proj", "model.layers.84.mlp.experts.132.down_proj", "model.layers.84.mlp.experts.133.down_proj", "model.layers.84.mlp.experts.134.down_proj", "model.layers.84.mlp.experts.135.down_proj", "model.layers.84.mlp.experts.136.down_proj", "model.layers.84.mlp.experts.137.down_proj", "model.layers.84.mlp.experts.138.down_proj", "model.layers.84.mlp.experts.139.down_proj", "model.layers.84.mlp.experts.140.down_proj", "model.layers.84.mlp.experts.141.down_proj", "model.layers.84.mlp.experts.142.down_proj", "model.layers.84.mlp.experts.143.down_proj", "model.layers.84.mlp.experts.144.down_proj", "model.layers.84.mlp.experts.145.down_proj", "model.layers.84.mlp.experts.146.down_proj", "model.layers.84.mlp.experts.147.down_proj", "model.layers.84.mlp.experts.148.down_proj", "model.layers.84.mlp.experts.149.down_proj", "model.layers.84.mlp.experts.150.down_proj", "model.layers.84.mlp.experts.151.down_proj", "model.layers.84.mlp.experts.152.down_proj", "model.layers.84.mlp.experts.153.down_proj", "model.layers.84.mlp.experts.154.down_proj", "model.layers.84.mlp.experts.155.down_proj", "model.layers.84.mlp.experts.156.down_proj", "model.layers.84.mlp.experts.157.down_proj", "model.layers.84.mlp.experts.158.down_proj", "model.layers.84.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -1.8031336367213546e-06, "dbits": 1258291200 }, { "dkld": -2.5972165167337247e-05, "dbits": 2516582400 }, { "dkld": -3.061043098569194e-05, "dbits": 3774873600 }, { "dkld": -2.2106058895590697e-05, "dbits": 6291456000 } ] }, { "idx": 507, "layers": [ "model.layers.85.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0004691788926720619, "dbits": 62914560 }, { "dkld": -0.0005176883190870368, "dbits": 125829120 }, { "dkld": -0.0007233385927975233, "dbits": 188743680 }, { "dkld": -0.000840206723660239, "dbits": 314572800 } ] }, { "idx": 508, "layers": [ "model.layers.85.self_attn.k_proj", "model.layers.85.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0006534513086080551, "dbits": 10485760 }, { "dkld": -0.0008580131456255996, "dbits": 20971520 }, { "dkld": -0.0010214449837803868, "dbits": 31457280 }, { "dkld": -0.0010003375820815674, "dbits": 52428800 } ] }, { "idx": 509, "layers": [ "model.layers.85.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0004882584325969219, "dbits": 62914560 }, { "dkld": -0.0006626525893807522, "dbits": 125829120 }, { "dkld": -0.0006920461542904432, "dbits": 188743680 }, { "dkld": -0.0006923511624336298, "dbits": 314572800 } ] }, { "idx": 510, "layers": [ "model.layers.85.mlp.shared_experts.gate_proj", "model.layers.85.mlp.shared_experts.up_proj", "model.layers.85.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.00024020392447710037, "dbits": 23592960 }, { "dkld": -0.00010501593351364136, "dbits": 47185920 }, { "dkld": -0.00012296885252000012, "dbits": 70778880 }, { "dkld": -0.000126589927822357, "dbits": 117964800 } ] }, { "idx": 511, "layers": [ "model.layers.85.mlp.experts.0.gate_proj", "model.layers.85.mlp.experts.1.gate_proj", "model.layers.85.mlp.experts.2.gate_proj", "model.layers.85.mlp.experts.3.gate_proj", "model.layers.85.mlp.experts.4.gate_proj", "model.layers.85.mlp.experts.5.gate_proj", "model.layers.85.mlp.experts.6.gate_proj", "model.layers.85.mlp.experts.7.gate_proj", "model.layers.85.mlp.experts.8.gate_proj", "model.layers.85.mlp.experts.9.gate_proj", "model.layers.85.mlp.experts.10.gate_proj", "model.layers.85.mlp.experts.11.gate_proj", "model.layers.85.mlp.experts.12.gate_proj", "model.layers.85.mlp.experts.13.gate_proj", "model.layers.85.mlp.experts.14.gate_proj", "model.layers.85.mlp.experts.15.gate_proj", "model.layers.85.mlp.experts.16.gate_proj", "model.layers.85.mlp.experts.17.gate_proj", "model.layers.85.mlp.experts.18.gate_proj", "model.layers.85.mlp.experts.19.gate_proj", "model.layers.85.mlp.experts.20.gate_proj", "model.layers.85.mlp.experts.21.gate_proj", "model.layers.85.mlp.experts.22.gate_proj", "model.layers.85.mlp.experts.23.gate_proj", "model.layers.85.mlp.experts.24.gate_proj", "model.layers.85.mlp.experts.25.gate_proj", "model.layers.85.mlp.experts.26.gate_proj", "model.layers.85.mlp.experts.27.gate_proj", "model.layers.85.mlp.experts.28.gate_proj", "model.layers.85.mlp.experts.29.gate_proj", "model.layers.85.mlp.experts.30.gate_proj", "model.layers.85.mlp.experts.31.gate_proj", "model.layers.85.mlp.experts.32.gate_proj", "model.layers.85.mlp.experts.33.gate_proj", "model.layers.85.mlp.experts.34.gate_proj", "model.layers.85.mlp.experts.35.gate_proj", "model.layers.85.mlp.experts.36.gate_proj", "model.layers.85.mlp.experts.37.gate_proj", "model.layers.85.mlp.experts.38.gate_proj", "model.layers.85.mlp.experts.39.gate_proj", "model.layers.85.mlp.experts.40.gate_proj", "model.layers.85.mlp.experts.41.gate_proj", "model.layers.85.mlp.experts.42.gate_proj", "model.layers.85.mlp.experts.43.gate_proj", "model.layers.85.mlp.experts.44.gate_proj", "model.layers.85.mlp.experts.45.gate_proj", "model.layers.85.mlp.experts.46.gate_proj", "model.layers.85.mlp.experts.47.gate_proj", "model.layers.85.mlp.experts.48.gate_proj", "model.layers.85.mlp.experts.49.gate_proj", "model.layers.85.mlp.experts.50.gate_proj", "model.layers.85.mlp.experts.51.gate_proj", "model.layers.85.mlp.experts.52.gate_proj", "model.layers.85.mlp.experts.53.gate_proj", "model.layers.85.mlp.experts.54.gate_proj", "model.layers.85.mlp.experts.55.gate_proj", "model.layers.85.mlp.experts.56.gate_proj", "model.layers.85.mlp.experts.57.gate_proj", "model.layers.85.mlp.experts.58.gate_proj", "model.layers.85.mlp.experts.59.gate_proj", "model.layers.85.mlp.experts.60.gate_proj", "model.layers.85.mlp.experts.61.gate_proj", "model.layers.85.mlp.experts.62.gate_proj", "model.layers.85.mlp.experts.63.gate_proj", "model.layers.85.mlp.experts.64.gate_proj", "model.layers.85.mlp.experts.65.gate_proj", "model.layers.85.mlp.experts.66.gate_proj", "model.layers.85.mlp.experts.67.gate_proj", "model.layers.85.mlp.experts.68.gate_proj", "model.layers.85.mlp.experts.69.gate_proj", "model.layers.85.mlp.experts.70.gate_proj", "model.layers.85.mlp.experts.71.gate_proj", "model.layers.85.mlp.experts.72.gate_proj", "model.layers.85.mlp.experts.73.gate_proj", "model.layers.85.mlp.experts.74.gate_proj", "model.layers.85.mlp.experts.75.gate_proj", "model.layers.85.mlp.experts.76.gate_proj", "model.layers.85.mlp.experts.77.gate_proj", "model.layers.85.mlp.experts.78.gate_proj", "model.layers.85.mlp.experts.79.gate_proj", "model.layers.85.mlp.experts.80.gate_proj", "model.layers.85.mlp.experts.81.gate_proj", "model.layers.85.mlp.experts.82.gate_proj", "model.layers.85.mlp.experts.83.gate_proj", "model.layers.85.mlp.experts.84.gate_proj", "model.layers.85.mlp.experts.85.gate_proj", "model.layers.85.mlp.experts.86.gate_proj", "model.layers.85.mlp.experts.87.gate_proj", "model.layers.85.mlp.experts.88.gate_proj", "model.layers.85.mlp.experts.89.gate_proj", "model.layers.85.mlp.experts.90.gate_proj", "model.layers.85.mlp.experts.91.gate_proj", "model.layers.85.mlp.experts.92.gate_proj", "model.layers.85.mlp.experts.93.gate_proj", "model.layers.85.mlp.experts.94.gate_proj", "model.layers.85.mlp.experts.95.gate_proj", "model.layers.85.mlp.experts.96.gate_proj", "model.layers.85.mlp.experts.97.gate_proj", "model.layers.85.mlp.experts.98.gate_proj", "model.layers.85.mlp.experts.99.gate_proj", "model.layers.85.mlp.experts.100.gate_proj", "model.layers.85.mlp.experts.101.gate_proj", "model.layers.85.mlp.experts.102.gate_proj", "model.layers.85.mlp.experts.103.gate_proj", "model.layers.85.mlp.experts.104.gate_proj", "model.layers.85.mlp.experts.105.gate_proj", "model.layers.85.mlp.experts.106.gate_proj", "model.layers.85.mlp.experts.107.gate_proj", "model.layers.85.mlp.experts.108.gate_proj", "model.layers.85.mlp.experts.109.gate_proj", "model.layers.85.mlp.experts.110.gate_proj", "model.layers.85.mlp.experts.111.gate_proj", "model.layers.85.mlp.experts.112.gate_proj", "model.layers.85.mlp.experts.113.gate_proj", "model.layers.85.mlp.experts.114.gate_proj", "model.layers.85.mlp.experts.115.gate_proj", "model.layers.85.mlp.experts.116.gate_proj", "model.layers.85.mlp.experts.117.gate_proj", "model.layers.85.mlp.experts.118.gate_proj", "model.layers.85.mlp.experts.119.gate_proj", "model.layers.85.mlp.experts.120.gate_proj", "model.layers.85.mlp.experts.121.gate_proj", "model.layers.85.mlp.experts.122.gate_proj", "model.layers.85.mlp.experts.123.gate_proj", "model.layers.85.mlp.experts.124.gate_proj", "model.layers.85.mlp.experts.125.gate_proj", "model.layers.85.mlp.experts.126.gate_proj", "model.layers.85.mlp.experts.127.gate_proj", "model.layers.85.mlp.experts.128.gate_proj", "model.layers.85.mlp.experts.129.gate_proj", "model.layers.85.mlp.experts.130.gate_proj", "model.layers.85.mlp.experts.131.gate_proj", "model.layers.85.mlp.experts.132.gate_proj", "model.layers.85.mlp.experts.133.gate_proj", "model.layers.85.mlp.experts.134.gate_proj", "model.layers.85.mlp.experts.135.gate_proj", "model.layers.85.mlp.experts.136.gate_proj", "model.layers.85.mlp.experts.137.gate_proj", "model.layers.85.mlp.experts.138.gate_proj", "model.layers.85.mlp.experts.139.gate_proj", "model.layers.85.mlp.experts.140.gate_proj", "model.layers.85.mlp.experts.141.gate_proj", "model.layers.85.mlp.experts.142.gate_proj", "model.layers.85.mlp.experts.143.gate_proj", "model.layers.85.mlp.experts.144.gate_proj", "model.layers.85.mlp.experts.145.gate_proj", "model.layers.85.mlp.experts.146.gate_proj", "model.layers.85.mlp.experts.147.gate_proj", "model.layers.85.mlp.experts.148.gate_proj", "model.layers.85.mlp.experts.149.gate_proj", "model.layers.85.mlp.experts.150.gate_proj", "model.layers.85.mlp.experts.151.gate_proj", "model.layers.85.mlp.experts.152.gate_proj", "model.layers.85.mlp.experts.153.gate_proj", "model.layers.85.mlp.experts.154.gate_proj", "model.layers.85.mlp.experts.155.gate_proj", "model.layers.85.mlp.experts.156.gate_proj", "model.layers.85.mlp.experts.157.gate_proj", "model.layers.85.mlp.experts.158.gate_proj", "model.layers.85.mlp.experts.159.gate_proj", "model.layers.85.mlp.experts.0.up_proj", "model.layers.85.mlp.experts.1.up_proj", "model.layers.85.mlp.experts.2.up_proj", "model.layers.85.mlp.experts.3.up_proj", "model.layers.85.mlp.experts.4.up_proj", "model.layers.85.mlp.experts.5.up_proj", "model.layers.85.mlp.experts.6.up_proj", "model.layers.85.mlp.experts.7.up_proj", "model.layers.85.mlp.experts.8.up_proj", "model.layers.85.mlp.experts.9.up_proj", "model.layers.85.mlp.experts.10.up_proj", "model.layers.85.mlp.experts.11.up_proj", "model.layers.85.mlp.experts.12.up_proj", "model.layers.85.mlp.experts.13.up_proj", "model.layers.85.mlp.experts.14.up_proj", "model.layers.85.mlp.experts.15.up_proj", "model.layers.85.mlp.experts.16.up_proj", "model.layers.85.mlp.experts.17.up_proj", "model.layers.85.mlp.experts.18.up_proj", "model.layers.85.mlp.experts.19.up_proj", "model.layers.85.mlp.experts.20.up_proj", "model.layers.85.mlp.experts.21.up_proj", "model.layers.85.mlp.experts.22.up_proj", "model.layers.85.mlp.experts.23.up_proj", "model.layers.85.mlp.experts.24.up_proj", "model.layers.85.mlp.experts.25.up_proj", "model.layers.85.mlp.experts.26.up_proj", "model.layers.85.mlp.experts.27.up_proj", "model.layers.85.mlp.experts.28.up_proj", "model.layers.85.mlp.experts.29.up_proj", "model.layers.85.mlp.experts.30.up_proj", "model.layers.85.mlp.experts.31.up_proj", "model.layers.85.mlp.experts.32.up_proj", "model.layers.85.mlp.experts.33.up_proj", "model.layers.85.mlp.experts.34.up_proj", "model.layers.85.mlp.experts.35.up_proj", "model.layers.85.mlp.experts.36.up_proj", "model.layers.85.mlp.experts.37.up_proj", "model.layers.85.mlp.experts.38.up_proj", "model.layers.85.mlp.experts.39.up_proj", "model.layers.85.mlp.experts.40.up_proj", "model.layers.85.mlp.experts.41.up_proj", "model.layers.85.mlp.experts.42.up_proj", "model.layers.85.mlp.experts.43.up_proj", "model.layers.85.mlp.experts.44.up_proj", "model.layers.85.mlp.experts.45.up_proj", "model.layers.85.mlp.experts.46.up_proj", "model.layers.85.mlp.experts.47.up_proj", "model.layers.85.mlp.experts.48.up_proj", "model.layers.85.mlp.experts.49.up_proj", "model.layers.85.mlp.experts.50.up_proj", "model.layers.85.mlp.experts.51.up_proj", "model.layers.85.mlp.experts.52.up_proj", "model.layers.85.mlp.experts.53.up_proj", "model.layers.85.mlp.experts.54.up_proj", "model.layers.85.mlp.experts.55.up_proj", "model.layers.85.mlp.experts.56.up_proj", "model.layers.85.mlp.experts.57.up_proj", "model.layers.85.mlp.experts.58.up_proj", "model.layers.85.mlp.experts.59.up_proj", "model.layers.85.mlp.experts.60.up_proj", "model.layers.85.mlp.experts.61.up_proj", "model.layers.85.mlp.experts.62.up_proj", "model.layers.85.mlp.experts.63.up_proj", "model.layers.85.mlp.experts.64.up_proj", "model.layers.85.mlp.experts.65.up_proj", "model.layers.85.mlp.experts.66.up_proj", "model.layers.85.mlp.experts.67.up_proj", "model.layers.85.mlp.experts.68.up_proj", "model.layers.85.mlp.experts.69.up_proj", "model.layers.85.mlp.experts.70.up_proj", "model.layers.85.mlp.experts.71.up_proj", "model.layers.85.mlp.experts.72.up_proj", "model.layers.85.mlp.experts.73.up_proj", "model.layers.85.mlp.experts.74.up_proj", "model.layers.85.mlp.experts.75.up_proj", "model.layers.85.mlp.experts.76.up_proj", "model.layers.85.mlp.experts.77.up_proj", "model.layers.85.mlp.experts.78.up_proj", "model.layers.85.mlp.experts.79.up_proj", "model.layers.85.mlp.experts.80.up_proj", "model.layers.85.mlp.experts.81.up_proj", "model.layers.85.mlp.experts.82.up_proj", "model.layers.85.mlp.experts.83.up_proj", "model.layers.85.mlp.experts.84.up_proj", "model.layers.85.mlp.experts.85.up_proj", "model.layers.85.mlp.experts.86.up_proj", "model.layers.85.mlp.experts.87.up_proj", "model.layers.85.mlp.experts.88.up_proj", "model.layers.85.mlp.experts.89.up_proj", "model.layers.85.mlp.experts.90.up_proj", "model.layers.85.mlp.experts.91.up_proj", "model.layers.85.mlp.experts.92.up_proj", "model.layers.85.mlp.experts.93.up_proj", "model.layers.85.mlp.experts.94.up_proj", "model.layers.85.mlp.experts.95.up_proj", "model.layers.85.mlp.experts.96.up_proj", "model.layers.85.mlp.experts.97.up_proj", "model.layers.85.mlp.experts.98.up_proj", "model.layers.85.mlp.experts.99.up_proj", "model.layers.85.mlp.experts.100.up_proj", "model.layers.85.mlp.experts.101.up_proj", "model.layers.85.mlp.experts.102.up_proj", "model.layers.85.mlp.experts.103.up_proj", "model.layers.85.mlp.experts.104.up_proj", "model.layers.85.mlp.experts.105.up_proj", "model.layers.85.mlp.experts.106.up_proj", "model.layers.85.mlp.experts.107.up_proj", "model.layers.85.mlp.experts.108.up_proj", "model.layers.85.mlp.experts.109.up_proj", "model.layers.85.mlp.experts.110.up_proj", "model.layers.85.mlp.experts.111.up_proj", "model.layers.85.mlp.experts.112.up_proj", "model.layers.85.mlp.experts.113.up_proj", "model.layers.85.mlp.experts.114.up_proj", "model.layers.85.mlp.experts.115.up_proj", "model.layers.85.mlp.experts.116.up_proj", "model.layers.85.mlp.experts.117.up_proj", "model.layers.85.mlp.experts.118.up_proj", "model.layers.85.mlp.experts.119.up_proj", "model.layers.85.mlp.experts.120.up_proj", "model.layers.85.mlp.experts.121.up_proj", "model.layers.85.mlp.experts.122.up_proj", "model.layers.85.mlp.experts.123.up_proj", "model.layers.85.mlp.experts.124.up_proj", "model.layers.85.mlp.experts.125.up_proj", "model.layers.85.mlp.experts.126.up_proj", "model.layers.85.mlp.experts.127.up_proj", "model.layers.85.mlp.experts.128.up_proj", "model.layers.85.mlp.experts.129.up_proj", "model.layers.85.mlp.experts.130.up_proj", "model.layers.85.mlp.experts.131.up_proj", "model.layers.85.mlp.experts.132.up_proj", "model.layers.85.mlp.experts.133.up_proj", "model.layers.85.mlp.experts.134.up_proj", "model.layers.85.mlp.experts.135.up_proj", "model.layers.85.mlp.experts.136.up_proj", "model.layers.85.mlp.experts.137.up_proj", "model.layers.85.mlp.experts.138.up_proj", "model.layers.85.mlp.experts.139.up_proj", "model.layers.85.mlp.experts.140.up_proj", "model.layers.85.mlp.experts.141.up_proj", "model.layers.85.mlp.experts.142.up_proj", "model.layers.85.mlp.experts.143.up_proj", "model.layers.85.mlp.experts.144.up_proj", "model.layers.85.mlp.experts.145.up_proj", "model.layers.85.mlp.experts.146.up_proj", "model.layers.85.mlp.experts.147.up_proj", "model.layers.85.mlp.experts.148.up_proj", "model.layers.85.mlp.experts.149.up_proj", "model.layers.85.mlp.experts.150.up_proj", "model.layers.85.mlp.experts.151.up_proj", "model.layers.85.mlp.experts.152.up_proj", "model.layers.85.mlp.experts.153.up_proj", "model.layers.85.mlp.experts.154.up_proj", "model.layers.85.mlp.experts.155.up_proj", "model.layers.85.mlp.experts.156.up_proj", "model.layers.85.mlp.experts.157.up_proj", "model.layers.85.mlp.experts.158.up_proj", "model.layers.85.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": 6.805816665290954e-05, "dbits": 2516582400 }, { "dkld": 0.00023750113323330047, "dbits": 5033164800 }, { "dkld": 0.00012985169887541614, "dbits": 7549747200 }, { "dkld": 0.00010263845324515186, "dbits": 12582912000 } ] }, { "idx": 512, "layers": [ "model.layers.85.mlp.experts.0.down_proj", "model.layers.85.mlp.experts.1.down_proj", "model.layers.85.mlp.experts.2.down_proj", "model.layers.85.mlp.experts.3.down_proj", "model.layers.85.mlp.experts.4.down_proj", "model.layers.85.mlp.experts.5.down_proj", "model.layers.85.mlp.experts.6.down_proj", "model.layers.85.mlp.experts.7.down_proj", "model.layers.85.mlp.experts.8.down_proj", "model.layers.85.mlp.experts.9.down_proj", "model.layers.85.mlp.experts.10.down_proj", "model.layers.85.mlp.experts.11.down_proj", "model.layers.85.mlp.experts.12.down_proj", "model.layers.85.mlp.experts.13.down_proj", "model.layers.85.mlp.experts.14.down_proj", "model.layers.85.mlp.experts.15.down_proj", "model.layers.85.mlp.experts.16.down_proj", "model.layers.85.mlp.experts.17.down_proj", "model.layers.85.mlp.experts.18.down_proj", "model.layers.85.mlp.experts.19.down_proj", "model.layers.85.mlp.experts.20.down_proj", "model.layers.85.mlp.experts.21.down_proj", "model.layers.85.mlp.experts.22.down_proj", "model.layers.85.mlp.experts.23.down_proj", "model.layers.85.mlp.experts.24.down_proj", "model.layers.85.mlp.experts.25.down_proj", "model.layers.85.mlp.experts.26.down_proj", "model.layers.85.mlp.experts.27.down_proj", "model.layers.85.mlp.experts.28.down_proj", "model.layers.85.mlp.experts.29.down_proj", "model.layers.85.mlp.experts.30.down_proj", "model.layers.85.mlp.experts.31.down_proj", "model.layers.85.mlp.experts.32.down_proj", "model.layers.85.mlp.experts.33.down_proj", "model.layers.85.mlp.experts.34.down_proj", "model.layers.85.mlp.experts.35.down_proj", "model.layers.85.mlp.experts.36.down_proj", "model.layers.85.mlp.experts.37.down_proj", "model.layers.85.mlp.experts.38.down_proj", "model.layers.85.mlp.experts.39.down_proj", "model.layers.85.mlp.experts.40.down_proj", "model.layers.85.mlp.experts.41.down_proj", "model.layers.85.mlp.experts.42.down_proj", "model.layers.85.mlp.experts.43.down_proj", "model.layers.85.mlp.experts.44.down_proj", "model.layers.85.mlp.experts.45.down_proj", "model.layers.85.mlp.experts.46.down_proj", "model.layers.85.mlp.experts.47.down_proj", "model.layers.85.mlp.experts.48.down_proj", "model.layers.85.mlp.experts.49.down_proj", "model.layers.85.mlp.experts.50.down_proj", "model.layers.85.mlp.experts.51.down_proj", "model.layers.85.mlp.experts.52.down_proj", "model.layers.85.mlp.experts.53.down_proj", "model.layers.85.mlp.experts.54.down_proj", "model.layers.85.mlp.experts.55.down_proj", "model.layers.85.mlp.experts.56.down_proj", "model.layers.85.mlp.experts.57.down_proj", "model.layers.85.mlp.experts.58.down_proj", "model.layers.85.mlp.experts.59.down_proj", "model.layers.85.mlp.experts.60.down_proj", "model.layers.85.mlp.experts.61.down_proj", "model.layers.85.mlp.experts.62.down_proj", "model.layers.85.mlp.experts.63.down_proj", "model.layers.85.mlp.experts.64.down_proj", "model.layers.85.mlp.experts.65.down_proj", "model.layers.85.mlp.experts.66.down_proj", "model.layers.85.mlp.experts.67.down_proj", "model.layers.85.mlp.experts.68.down_proj", "model.layers.85.mlp.experts.69.down_proj", "model.layers.85.mlp.experts.70.down_proj", "model.layers.85.mlp.experts.71.down_proj", "model.layers.85.mlp.experts.72.down_proj", "model.layers.85.mlp.experts.73.down_proj", "model.layers.85.mlp.experts.74.down_proj", "model.layers.85.mlp.experts.75.down_proj", "model.layers.85.mlp.experts.76.down_proj", "model.layers.85.mlp.experts.77.down_proj", "model.layers.85.mlp.experts.78.down_proj", "model.layers.85.mlp.experts.79.down_proj", "model.layers.85.mlp.experts.80.down_proj", "model.layers.85.mlp.experts.81.down_proj", "model.layers.85.mlp.experts.82.down_proj", "model.layers.85.mlp.experts.83.down_proj", "model.layers.85.mlp.experts.84.down_proj", "model.layers.85.mlp.experts.85.down_proj", "model.layers.85.mlp.experts.86.down_proj", "model.layers.85.mlp.experts.87.down_proj", "model.layers.85.mlp.experts.88.down_proj", "model.layers.85.mlp.experts.89.down_proj", "model.layers.85.mlp.experts.90.down_proj", "model.layers.85.mlp.experts.91.down_proj", "model.layers.85.mlp.experts.92.down_proj", "model.layers.85.mlp.experts.93.down_proj", "model.layers.85.mlp.experts.94.down_proj", "model.layers.85.mlp.experts.95.down_proj", "model.layers.85.mlp.experts.96.down_proj", "model.layers.85.mlp.experts.97.down_proj", "model.layers.85.mlp.experts.98.down_proj", "model.layers.85.mlp.experts.99.down_proj", "model.layers.85.mlp.experts.100.down_proj", "model.layers.85.mlp.experts.101.down_proj", "model.layers.85.mlp.experts.102.down_proj", "model.layers.85.mlp.experts.103.down_proj", "model.layers.85.mlp.experts.104.down_proj", "model.layers.85.mlp.experts.105.down_proj", "model.layers.85.mlp.experts.106.down_proj", "model.layers.85.mlp.experts.107.down_proj", "model.layers.85.mlp.experts.108.down_proj", "model.layers.85.mlp.experts.109.down_proj", "model.layers.85.mlp.experts.110.down_proj", "model.layers.85.mlp.experts.111.down_proj", "model.layers.85.mlp.experts.112.down_proj", "model.layers.85.mlp.experts.113.down_proj", "model.layers.85.mlp.experts.114.down_proj", "model.layers.85.mlp.experts.115.down_proj", "model.layers.85.mlp.experts.116.down_proj", "model.layers.85.mlp.experts.117.down_proj", "model.layers.85.mlp.experts.118.down_proj", "model.layers.85.mlp.experts.119.down_proj", "model.layers.85.mlp.experts.120.down_proj", "model.layers.85.mlp.experts.121.down_proj", "model.layers.85.mlp.experts.122.down_proj", "model.layers.85.mlp.experts.123.down_proj", "model.layers.85.mlp.experts.124.down_proj", "model.layers.85.mlp.experts.125.down_proj", "model.layers.85.mlp.experts.126.down_proj", "model.layers.85.mlp.experts.127.down_proj", "model.layers.85.mlp.experts.128.down_proj", "model.layers.85.mlp.experts.129.down_proj", "model.layers.85.mlp.experts.130.down_proj", "model.layers.85.mlp.experts.131.down_proj", "model.layers.85.mlp.experts.132.down_proj", "model.layers.85.mlp.experts.133.down_proj", "model.layers.85.mlp.experts.134.down_proj", "model.layers.85.mlp.experts.135.down_proj", "model.layers.85.mlp.experts.136.down_proj", "model.layers.85.mlp.experts.137.down_proj", "model.layers.85.mlp.experts.138.down_proj", "model.layers.85.mlp.experts.139.down_proj", "model.layers.85.mlp.experts.140.down_proj", "model.layers.85.mlp.experts.141.down_proj", "model.layers.85.mlp.experts.142.down_proj", "model.layers.85.mlp.experts.143.down_proj", "model.layers.85.mlp.experts.144.down_proj", "model.layers.85.mlp.experts.145.down_proj", "model.layers.85.mlp.experts.146.down_proj", "model.layers.85.mlp.experts.147.down_proj", "model.layers.85.mlp.experts.148.down_proj", "model.layers.85.mlp.experts.149.down_proj", "model.layers.85.mlp.experts.150.down_proj", "model.layers.85.mlp.experts.151.down_proj", "model.layers.85.mlp.experts.152.down_proj", "model.layers.85.mlp.experts.153.down_proj", "model.layers.85.mlp.experts.154.down_proj", "model.layers.85.mlp.experts.155.down_proj", "model.layers.85.mlp.experts.156.down_proj", "model.layers.85.mlp.experts.157.down_proj", "model.layers.85.mlp.experts.158.down_proj", "model.layers.85.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -7.009990513325015e-05, "dbits": 1258291200 }, { "dkld": -4.7489255666735564e-05, "dbits": 2516582400 }, { "dkld": -6.446279585362036e-05, "dbits": 3774873600 }, { "dkld": -5.442174151540358e-05, "dbits": 6291456000 } ] }, { "idx": 513, "layers": [ "model.layers.86.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00048508308827877045, "dbits": 62914560 }, { "dkld": -0.0003969860263168812, "dbits": 125829120 }, { "dkld": -0.0004767698235809831, "dbits": 188743680 }, { "dkld": -0.0004672875627875356, "dbits": 314572800 } ] }, { "idx": 514, "layers": [ "model.layers.86.self_attn.k_proj", "model.layers.86.self_attn.v_proj" ], "candidates": [ { "dkld": 3.3750198781490326e-05, "dbits": 10485760 }, { "dkld": -0.0003647599369287491, "dbits": 20971520 }, { "dkld": -0.0003408805467188386, "dbits": 31457280 }, { "dkld": -0.00040891654789448895, "dbits": 52428800 } ] }, { "idx": 515, "layers": [ "model.layers.86.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00015688650310038965, "dbits": 62914560 }, { "dkld": -5.0526671111592036e-05, "dbits": 125829120 }, { "dkld": -2.4501606822019406e-05, "dbits": 188743680 }, { "dkld": -2.622054889798997e-05, "dbits": 314572800 } ] }, { "idx": 516, "layers": [ "model.layers.86.mlp.shared_experts.gate_proj", "model.layers.86.mlp.shared_experts.up_proj", "model.layers.86.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0005745453760027941, "dbits": 23592960 }, { "dkld": -0.0005313243716955213, "dbits": 47185920 }, { "dkld": -0.0005798816680908231, "dbits": 70778880 }, { "dkld": -0.0005642732605338208, "dbits": 117964800 } ] }, { "idx": 517, "layers": [ "model.layers.86.mlp.experts.0.gate_proj", "model.layers.86.mlp.experts.1.gate_proj", "model.layers.86.mlp.experts.2.gate_proj", "model.layers.86.mlp.experts.3.gate_proj", "model.layers.86.mlp.experts.4.gate_proj", "model.layers.86.mlp.experts.5.gate_proj", "model.layers.86.mlp.experts.6.gate_proj", "model.layers.86.mlp.experts.7.gate_proj", "model.layers.86.mlp.experts.8.gate_proj", "model.layers.86.mlp.experts.9.gate_proj", "model.layers.86.mlp.experts.10.gate_proj", "model.layers.86.mlp.experts.11.gate_proj", "model.layers.86.mlp.experts.12.gate_proj", "model.layers.86.mlp.experts.13.gate_proj", "model.layers.86.mlp.experts.14.gate_proj", "model.layers.86.mlp.experts.15.gate_proj", "model.layers.86.mlp.experts.16.gate_proj", "model.layers.86.mlp.experts.17.gate_proj", "model.layers.86.mlp.experts.18.gate_proj", "model.layers.86.mlp.experts.19.gate_proj", "model.layers.86.mlp.experts.20.gate_proj", "model.layers.86.mlp.experts.21.gate_proj", "model.layers.86.mlp.experts.22.gate_proj", "model.layers.86.mlp.experts.23.gate_proj", "model.layers.86.mlp.experts.24.gate_proj", "model.layers.86.mlp.experts.25.gate_proj", "model.layers.86.mlp.experts.26.gate_proj", "model.layers.86.mlp.experts.27.gate_proj", "model.layers.86.mlp.experts.28.gate_proj", "model.layers.86.mlp.experts.29.gate_proj", "model.layers.86.mlp.experts.30.gate_proj", "model.layers.86.mlp.experts.31.gate_proj", "model.layers.86.mlp.experts.32.gate_proj", "model.layers.86.mlp.experts.33.gate_proj", "model.layers.86.mlp.experts.34.gate_proj", "model.layers.86.mlp.experts.35.gate_proj", "model.layers.86.mlp.experts.36.gate_proj", "model.layers.86.mlp.experts.37.gate_proj", "model.layers.86.mlp.experts.38.gate_proj", "model.layers.86.mlp.experts.39.gate_proj", "model.layers.86.mlp.experts.40.gate_proj", "model.layers.86.mlp.experts.41.gate_proj", "model.layers.86.mlp.experts.42.gate_proj", "model.layers.86.mlp.experts.43.gate_proj", "model.layers.86.mlp.experts.44.gate_proj", "model.layers.86.mlp.experts.45.gate_proj", "model.layers.86.mlp.experts.46.gate_proj", "model.layers.86.mlp.experts.47.gate_proj", "model.layers.86.mlp.experts.48.gate_proj", "model.layers.86.mlp.experts.49.gate_proj", "model.layers.86.mlp.experts.50.gate_proj", "model.layers.86.mlp.experts.51.gate_proj", "model.layers.86.mlp.experts.52.gate_proj", "model.layers.86.mlp.experts.53.gate_proj", "model.layers.86.mlp.experts.54.gate_proj", "model.layers.86.mlp.experts.55.gate_proj", "model.layers.86.mlp.experts.56.gate_proj", "model.layers.86.mlp.experts.57.gate_proj", "model.layers.86.mlp.experts.58.gate_proj", "model.layers.86.mlp.experts.59.gate_proj", "model.layers.86.mlp.experts.60.gate_proj", "model.layers.86.mlp.experts.61.gate_proj", "model.layers.86.mlp.experts.62.gate_proj", "model.layers.86.mlp.experts.63.gate_proj", "model.layers.86.mlp.experts.64.gate_proj", "model.layers.86.mlp.experts.65.gate_proj", "model.layers.86.mlp.experts.66.gate_proj", "model.layers.86.mlp.experts.67.gate_proj", "model.layers.86.mlp.experts.68.gate_proj", "model.layers.86.mlp.experts.69.gate_proj", "model.layers.86.mlp.experts.70.gate_proj", "model.layers.86.mlp.experts.71.gate_proj", "model.layers.86.mlp.experts.72.gate_proj", "model.layers.86.mlp.experts.73.gate_proj", "model.layers.86.mlp.experts.74.gate_proj", "model.layers.86.mlp.experts.75.gate_proj", "model.layers.86.mlp.experts.76.gate_proj", "model.layers.86.mlp.experts.77.gate_proj", "model.layers.86.mlp.experts.78.gate_proj", "model.layers.86.mlp.experts.79.gate_proj", "model.layers.86.mlp.experts.80.gate_proj", "model.layers.86.mlp.experts.81.gate_proj", "model.layers.86.mlp.experts.82.gate_proj", "model.layers.86.mlp.experts.83.gate_proj", "model.layers.86.mlp.experts.84.gate_proj", "model.layers.86.mlp.experts.85.gate_proj", "model.layers.86.mlp.experts.86.gate_proj", "model.layers.86.mlp.experts.87.gate_proj", "model.layers.86.mlp.experts.88.gate_proj", "model.layers.86.mlp.experts.89.gate_proj", "model.layers.86.mlp.experts.90.gate_proj", "model.layers.86.mlp.experts.91.gate_proj", "model.layers.86.mlp.experts.92.gate_proj", "model.layers.86.mlp.experts.93.gate_proj", "model.layers.86.mlp.experts.94.gate_proj", "model.layers.86.mlp.experts.95.gate_proj", "model.layers.86.mlp.experts.96.gate_proj", "model.layers.86.mlp.experts.97.gate_proj", "model.layers.86.mlp.experts.98.gate_proj", "model.layers.86.mlp.experts.99.gate_proj", "model.layers.86.mlp.experts.100.gate_proj", "model.layers.86.mlp.experts.101.gate_proj", "model.layers.86.mlp.experts.102.gate_proj", "model.layers.86.mlp.experts.103.gate_proj", "model.layers.86.mlp.experts.104.gate_proj", "model.layers.86.mlp.experts.105.gate_proj", "model.layers.86.mlp.experts.106.gate_proj", "model.layers.86.mlp.experts.107.gate_proj", "model.layers.86.mlp.experts.108.gate_proj", "model.layers.86.mlp.experts.109.gate_proj", "model.layers.86.mlp.experts.110.gate_proj", "model.layers.86.mlp.experts.111.gate_proj", "model.layers.86.mlp.experts.112.gate_proj", "model.layers.86.mlp.experts.113.gate_proj", "model.layers.86.mlp.experts.114.gate_proj", "model.layers.86.mlp.experts.115.gate_proj", "model.layers.86.mlp.experts.116.gate_proj", "model.layers.86.mlp.experts.117.gate_proj", "model.layers.86.mlp.experts.118.gate_proj", "model.layers.86.mlp.experts.119.gate_proj", "model.layers.86.mlp.experts.120.gate_proj", "model.layers.86.mlp.experts.121.gate_proj", "model.layers.86.mlp.experts.122.gate_proj", "model.layers.86.mlp.experts.123.gate_proj", "model.layers.86.mlp.experts.124.gate_proj", "model.layers.86.mlp.experts.125.gate_proj", "model.layers.86.mlp.experts.126.gate_proj", "model.layers.86.mlp.experts.127.gate_proj", "model.layers.86.mlp.experts.128.gate_proj", "model.layers.86.mlp.experts.129.gate_proj", "model.layers.86.mlp.experts.130.gate_proj", "model.layers.86.mlp.experts.131.gate_proj", "model.layers.86.mlp.experts.132.gate_proj", "model.layers.86.mlp.experts.133.gate_proj", "model.layers.86.mlp.experts.134.gate_proj", "model.layers.86.mlp.experts.135.gate_proj", "model.layers.86.mlp.experts.136.gate_proj", "model.layers.86.mlp.experts.137.gate_proj", "model.layers.86.mlp.experts.138.gate_proj", "model.layers.86.mlp.experts.139.gate_proj", "model.layers.86.mlp.experts.140.gate_proj", "model.layers.86.mlp.experts.141.gate_proj", "model.layers.86.mlp.experts.142.gate_proj", "model.layers.86.mlp.experts.143.gate_proj", "model.layers.86.mlp.experts.144.gate_proj", "model.layers.86.mlp.experts.145.gate_proj", "model.layers.86.mlp.experts.146.gate_proj", "model.layers.86.mlp.experts.147.gate_proj", "model.layers.86.mlp.experts.148.gate_proj", "model.layers.86.mlp.experts.149.gate_proj", "model.layers.86.mlp.experts.150.gate_proj", "model.layers.86.mlp.experts.151.gate_proj", "model.layers.86.mlp.experts.152.gate_proj", "model.layers.86.mlp.experts.153.gate_proj", "model.layers.86.mlp.experts.154.gate_proj", "model.layers.86.mlp.experts.155.gate_proj", "model.layers.86.mlp.experts.156.gate_proj", "model.layers.86.mlp.experts.157.gate_proj", "model.layers.86.mlp.experts.158.gate_proj", "model.layers.86.mlp.experts.159.gate_proj", "model.layers.86.mlp.experts.0.up_proj", "model.layers.86.mlp.experts.1.up_proj", "model.layers.86.mlp.experts.2.up_proj", "model.layers.86.mlp.experts.3.up_proj", "model.layers.86.mlp.experts.4.up_proj", "model.layers.86.mlp.experts.5.up_proj", "model.layers.86.mlp.experts.6.up_proj", "model.layers.86.mlp.experts.7.up_proj", "model.layers.86.mlp.experts.8.up_proj", "model.layers.86.mlp.experts.9.up_proj", "model.layers.86.mlp.experts.10.up_proj", "model.layers.86.mlp.experts.11.up_proj", "model.layers.86.mlp.experts.12.up_proj", "model.layers.86.mlp.experts.13.up_proj", "model.layers.86.mlp.experts.14.up_proj", "model.layers.86.mlp.experts.15.up_proj", "model.layers.86.mlp.experts.16.up_proj", "model.layers.86.mlp.experts.17.up_proj", "model.layers.86.mlp.experts.18.up_proj", "model.layers.86.mlp.experts.19.up_proj", "model.layers.86.mlp.experts.20.up_proj", "model.layers.86.mlp.experts.21.up_proj", "model.layers.86.mlp.experts.22.up_proj", "model.layers.86.mlp.experts.23.up_proj", "model.layers.86.mlp.experts.24.up_proj", "model.layers.86.mlp.experts.25.up_proj", "model.layers.86.mlp.experts.26.up_proj", "model.layers.86.mlp.experts.27.up_proj", "model.layers.86.mlp.experts.28.up_proj", "model.layers.86.mlp.experts.29.up_proj", "model.layers.86.mlp.experts.30.up_proj", "model.layers.86.mlp.experts.31.up_proj", "model.layers.86.mlp.experts.32.up_proj", "model.layers.86.mlp.experts.33.up_proj", "model.layers.86.mlp.experts.34.up_proj", "model.layers.86.mlp.experts.35.up_proj", "model.layers.86.mlp.experts.36.up_proj", "model.layers.86.mlp.experts.37.up_proj", "model.layers.86.mlp.experts.38.up_proj", "model.layers.86.mlp.experts.39.up_proj", "model.layers.86.mlp.experts.40.up_proj", "model.layers.86.mlp.experts.41.up_proj", "model.layers.86.mlp.experts.42.up_proj", "model.layers.86.mlp.experts.43.up_proj", "model.layers.86.mlp.experts.44.up_proj", "model.layers.86.mlp.experts.45.up_proj", "model.layers.86.mlp.experts.46.up_proj", "model.layers.86.mlp.experts.47.up_proj", "model.layers.86.mlp.experts.48.up_proj", "model.layers.86.mlp.experts.49.up_proj", "model.layers.86.mlp.experts.50.up_proj", "model.layers.86.mlp.experts.51.up_proj", "model.layers.86.mlp.experts.52.up_proj", "model.layers.86.mlp.experts.53.up_proj", "model.layers.86.mlp.experts.54.up_proj", "model.layers.86.mlp.experts.55.up_proj", "model.layers.86.mlp.experts.56.up_proj", "model.layers.86.mlp.experts.57.up_proj", "model.layers.86.mlp.experts.58.up_proj", "model.layers.86.mlp.experts.59.up_proj", "model.layers.86.mlp.experts.60.up_proj", "model.layers.86.mlp.experts.61.up_proj", "model.layers.86.mlp.experts.62.up_proj", "model.layers.86.mlp.experts.63.up_proj", "model.layers.86.mlp.experts.64.up_proj", "model.layers.86.mlp.experts.65.up_proj", "model.layers.86.mlp.experts.66.up_proj", "model.layers.86.mlp.experts.67.up_proj", "model.layers.86.mlp.experts.68.up_proj", "model.layers.86.mlp.experts.69.up_proj", "model.layers.86.mlp.experts.70.up_proj", "model.layers.86.mlp.experts.71.up_proj", "model.layers.86.mlp.experts.72.up_proj", "model.layers.86.mlp.experts.73.up_proj", "model.layers.86.mlp.experts.74.up_proj", "model.layers.86.mlp.experts.75.up_proj", "model.layers.86.mlp.experts.76.up_proj", "model.layers.86.mlp.experts.77.up_proj", "model.layers.86.mlp.experts.78.up_proj", "model.layers.86.mlp.experts.79.up_proj", "model.layers.86.mlp.experts.80.up_proj", "model.layers.86.mlp.experts.81.up_proj", "model.layers.86.mlp.experts.82.up_proj", "model.layers.86.mlp.experts.83.up_proj", "model.layers.86.mlp.experts.84.up_proj", "model.layers.86.mlp.experts.85.up_proj", "model.layers.86.mlp.experts.86.up_proj", "model.layers.86.mlp.experts.87.up_proj", "model.layers.86.mlp.experts.88.up_proj", "model.layers.86.mlp.experts.89.up_proj", "model.layers.86.mlp.experts.90.up_proj", "model.layers.86.mlp.experts.91.up_proj", "model.layers.86.mlp.experts.92.up_proj", "model.layers.86.mlp.experts.93.up_proj", "model.layers.86.mlp.experts.94.up_proj", "model.layers.86.mlp.experts.95.up_proj", "model.layers.86.mlp.experts.96.up_proj", "model.layers.86.mlp.experts.97.up_proj", "model.layers.86.mlp.experts.98.up_proj", "model.layers.86.mlp.experts.99.up_proj", "model.layers.86.mlp.experts.100.up_proj", "model.layers.86.mlp.experts.101.up_proj", "model.layers.86.mlp.experts.102.up_proj", "model.layers.86.mlp.experts.103.up_proj", "model.layers.86.mlp.experts.104.up_proj", "model.layers.86.mlp.experts.105.up_proj", "model.layers.86.mlp.experts.106.up_proj", "model.layers.86.mlp.experts.107.up_proj", "model.layers.86.mlp.experts.108.up_proj", "model.layers.86.mlp.experts.109.up_proj", "model.layers.86.mlp.experts.110.up_proj", "model.layers.86.mlp.experts.111.up_proj", "model.layers.86.mlp.experts.112.up_proj", "model.layers.86.mlp.experts.113.up_proj", "model.layers.86.mlp.experts.114.up_proj", "model.layers.86.mlp.experts.115.up_proj", "model.layers.86.mlp.experts.116.up_proj", "model.layers.86.mlp.experts.117.up_proj", "model.layers.86.mlp.experts.118.up_proj", "model.layers.86.mlp.experts.119.up_proj", "model.layers.86.mlp.experts.120.up_proj", "model.layers.86.mlp.experts.121.up_proj", "model.layers.86.mlp.experts.122.up_proj", "model.layers.86.mlp.experts.123.up_proj", "model.layers.86.mlp.experts.124.up_proj", "model.layers.86.mlp.experts.125.up_proj", "model.layers.86.mlp.experts.126.up_proj", "model.layers.86.mlp.experts.127.up_proj", "model.layers.86.mlp.experts.128.up_proj", "model.layers.86.mlp.experts.129.up_proj", "model.layers.86.mlp.experts.130.up_proj", "model.layers.86.mlp.experts.131.up_proj", "model.layers.86.mlp.experts.132.up_proj", "model.layers.86.mlp.experts.133.up_proj", "model.layers.86.mlp.experts.134.up_proj", "model.layers.86.mlp.experts.135.up_proj", "model.layers.86.mlp.experts.136.up_proj", "model.layers.86.mlp.experts.137.up_proj", "model.layers.86.mlp.experts.138.up_proj", "model.layers.86.mlp.experts.139.up_proj", "model.layers.86.mlp.experts.140.up_proj", "model.layers.86.mlp.experts.141.up_proj", "model.layers.86.mlp.experts.142.up_proj", "model.layers.86.mlp.experts.143.up_proj", "model.layers.86.mlp.experts.144.up_proj", "model.layers.86.mlp.experts.145.up_proj", "model.layers.86.mlp.experts.146.up_proj", "model.layers.86.mlp.experts.147.up_proj", "model.layers.86.mlp.experts.148.up_proj", "model.layers.86.mlp.experts.149.up_proj", "model.layers.86.mlp.experts.150.up_proj", "model.layers.86.mlp.experts.151.up_proj", "model.layers.86.mlp.experts.152.up_proj", "model.layers.86.mlp.experts.153.up_proj", "model.layers.86.mlp.experts.154.up_proj", "model.layers.86.mlp.experts.155.up_proj", "model.layers.86.mlp.experts.156.up_proj", "model.layers.86.mlp.experts.157.up_proj", "model.layers.86.mlp.experts.158.up_proj", "model.layers.86.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00012371717020870365, "dbits": 2516582400 }, { "dkld": -0.0002173963002860546, "dbits": 5033164800 }, { "dkld": -0.00020268140360713005, "dbits": 7549747200 }, { "dkld": -0.00018869750201702673, "dbits": 12582912000 } ] }, { "idx": 518, "layers": [ "model.layers.86.mlp.experts.0.down_proj", "model.layers.86.mlp.experts.1.down_proj", "model.layers.86.mlp.experts.2.down_proj", "model.layers.86.mlp.experts.3.down_proj", "model.layers.86.mlp.experts.4.down_proj", "model.layers.86.mlp.experts.5.down_proj", "model.layers.86.mlp.experts.6.down_proj", "model.layers.86.mlp.experts.7.down_proj", "model.layers.86.mlp.experts.8.down_proj", "model.layers.86.mlp.experts.9.down_proj", "model.layers.86.mlp.experts.10.down_proj", "model.layers.86.mlp.experts.11.down_proj", "model.layers.86.mlp.experts.12.down_proj", "model.layers.86.mlp.experts.13.down_proj", "model.layers.86.mlp.experts.14.down_proj", "model.layers.86.mlp.experts.15.down_proj", "model.layers.86.mlp.experts.16.down_proj", "model.layers.86.mlp.experts.17.down_proj", "model.layers.86.mlp.experts.18.down_proj", "model.layers.86.mlp.experts.19.down_proj", "model.layers.86.mlp.experts.20.down_proj", "model.layers.86.mlp.experts.21.down_proj", "model.layers.86.mlp.experts.22.down_proj", "model.layers.86.mlp.experts.23.down_proj", "model.layers.86.mlp.experts.24.down_proj", "model.layers.86.mlp.experts.25.down_proj", "model.layers.86.mlp.experts.26.down_proj", "model.layers.86.mlp.experts.27.down_proj", "model.layers.86.mlp.experts.28.down_proj", "model.layers.86.mlp.experts.29.down_proj", "model.layers.86.mlp.experts.30.down_proj", "model.layers.86.mlp.experts.31.down_proj", "model.layers.86.mlp.experts.32.down_proj", "model.layers.86.mlp.experts.33.down_proj", "model.layers.86.mlp.experts.34.down_proj", "model.layers.86.mlp.experts.35.down_proj", "model.layers.86.mlp.experts.36.down_proj", "model.layers.86.mlp.experts.37.down_proj", "model.layers.86.mlp.experts.38.down_proj", "model.layers.86.mlp.experts.39.down_proj", "model.layers.86.mlp.experts.40.down_proj", "model.layers.86.mlp.experts.41.down_proj", "model.layers.86.mlp.experts.42.down_proj", "model.layers.86.mlp.experts.43.down_proj", "model.layers.86.mlp.experts.44.down_proj", "model.layers.86.mlp.experts.45.down_proj", "model.layers.86.mlp.experts.46.down_proj", "model.layers.86.mlp.experts.47.down_proj", "model.layers.86.mlp.experts.48.down_proj", "model.layers.86.mlp.experts.49.down_proj", "model.layers.86.mlp.experts.50.down_proj", "model.layers.86.mlp.experts.51.down_proj", "model.layers.86.mlp.experts.52.down_proj", "model.layers.86.mlp.experts.53.down_proj", "model.layers.86.mlp.experts.54.down_proj", "model.layers.86.mlp.experts.55.down_proj", "model.layers.86.mlp.experts.56.down_proj", "model.layers.86.mlp.experts.57.down_proj", "model.layers.86.mlp.experts.58.down_proj", "model.layers.86.mlp.experts.59.down_proj", "model.layers.86.mlp.experts.60.down_proj", "model.layers.86.mlp.experts.61.down_proj", "model.layers.86.mlp.experts.62.down_proj", "model.layers.86.mlp.experts.63.down_proj", "model.layers.86.mlp.experts.64.down_proj", "model.layers.86.mlp.experts.65.down_proj", "model.layers.86.mlp.experts.66.down_proj", "model.layers.86.mlp.experts.67.down_proj", "model.layers.86.mlp.experts.68.down_proj", "model.layers.86.mlp.experts.69.down_proj", "model.layers.86.mlp.experts.70.down_proj", "model.layers.86.mlp.experts.71.down_proj", "model.layers.86.mlp.experts.72.down_proj", "model.layers.86.mlp.experts.73.down_proj", "model.layers.86.mlp.experts.74.down_proj", "model.layers.86.mlp.experts.75.down_proj", "model.layers.86.mlp.experts.76.down_proj", "model.layers.86.mlp.experts.77.down_proj", "model.layers.86.mlp.experts.78.down_proj", "model.layers.86.mlp.experts.79.down_proj", "model.layers.86.mlp.experts.80.down_proj", "model.layers.86.mlp.experts.81.down_proj", "model.layers.86.mlp.experts.82.down_proj", "model.layers.86.mlp.experts.83.down_proj", "model.layers.86.mlp.experts.84.down_proj", "model.layers.86.mlp.experts.85.down_proj", "model.layers.86.mlp.experts.86.down_proj", "model.layers.86.mlp.experts.87.down_proj", "model.layers.86.mlp.experts.88.down_proj", "model.layers.86.mlp.experts.89.down_proj", "model.layers.86.mlp.experts.90.down_proj", "model.layers.86.mlp.experts.91.down_proj", "model.layers.86.mlp.experts.92.down_proj", "model.layers.86.mlp.experts.93.down_proj", "model.layers.86.mlp.experts.94.down_proj", "model.layers.86.mlp.experts.95.down_proj", "model.layers.86.mlp.experts.96.down_proj", "model.layers.86.mlp.experts.97.down_proj", "model.layers.86.mlp.experts.98.down_proj", "model.layers.86.mlp.experts.99.down_proj", "model.layers.86.mlp.experts.100.down_proj", "model.layers.86.mlp.experts.101.down_proj", "model.layers.86.mlp.experts.102.down_proj", "model.layers.86.mlp.experts.103.down_proj", "model.layers.86.mlp.experts.104.down_proj", "model.layers.86.mlp.experts.105.down_proj", "model.layers.86.mlp.experts.106.down_proj", "model.layers.86.mlp.experts.107.down_proj", "model.layers.86.mlp.experts.108.down_proj", "model.layers.86.mlp.experts.109.down_proj", "model.layers.86.mlp.experts.110.down_proj", "model.layers.86.mlp.experts.111.down_proj", "model.layers.86.mlp.experts.112.down_proj", "model.layers.86.mlp.experts.113.down_proj", "model.layers.86.mlp.experts.114.down_proj", "model.layers.86.mlp.experts.115.down_proj", "model.layers.86.mlp.experts.116.down_proj", "model.layers.86.mlp.experts.117.down_proj", "model.layers.86.mlp.experts.118.down_proj", "model.layers.86.mlp.experts.119.down_proj", "model.layers.86.mlp.experts.120.down_proj", "model.layers.86.mlp.experts.121.down_proj", "model.layers.86.mlp.experts.122.down_proj", "model.layers.86.mlp.experts.123.down_proj", "model.layers.86.mlp.experts.124.down_proj", "model.layers.86.mlp.experts.125.down_proj", "model.layers.86.mlp.experts.126.down_proj", "model.layers.86.mlp.experts.127.down_proj", "model.layers.86.mlp.experts.128.down_proj", "model.layers.86.mlp.experts.129.down_proj", "model.layers.86.mlp.experts.130.down_proj", "model.layers.86.mlp.experts.131.down_proj", "model.layers.86.mlp.experts.132.down_proj", "model.layers.86.mlp.experts.133.down_proj", "model.layers.86.mlp.experts.134.down_proj", "model.layers.86.mlp.experts.135.down_proj", "model.layers.86.mlp.experts.136.down_proj", "model.layers.86.mlp.experts.137.down_proj", "model.layers.86.mlp.experts.138.down_proj", "model.layers.86.mlp.experts.139.down_proj", "model.layers.86.mlp.experts.140.down_proj", "model.layers.86.mlp.experts.141.down_proj", "model.layers.86.mlp.experts.142.down_proj", "model.layers.86.mlp.experts.143.down_proj", "model.layers.86.mlp.experts.144.down_proj", "model.layers.86.mlp.experts.145.down_proj", "model.layers.86.mlp.experts.146.down_proj", "model.layers.86.mlp.experts.147.down_proj", "model.layers.86.mlp.experts.148.down_proj", "model.layers.86.mlp.experts.149.down_proj", "model.layers.86.mlp.experts.150.down_proj", "model.layers.86.mlp.experts.151.down_proj", "model.layers.86.mlp.experts.152.down_proj", "model.layers.86.mlp.experts.153.down_proj", "model.layers.86.mlp.experts.154.down_proj", "model.layers.86.mlp.experts.155.down_proj", "model.layers.86.mlp.experts.156.down_proj", "model.layers.86.mlp.experts.157.down_proj", "model.layers.86.mlp.experts.158.down_proj", "model.layers.86.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -4.8920232802632246e-05, "dbits": 1258291200 }, { "dkld": -5.9546809643512555e-05, "dbits": 2516582400 }, { "dkld": -6.552459672093391e-05, "dbits": 3774873600 }, { "dkld": -8.9176185429099e-05, "dbits": 6291456000 } ] }, { "idx": 519, "layers": [ "model.layers.87.self_attn.q_proj" ], "candidates": [ { "dkld": 0.00021659331396221837, "dbits": 62914560 }, { "dkld": 0.00020209122449159345, "dbits": 125829120 }, { "dkld": 0.0002052775584161226, "dbits": 188743680 }, { "dkld": 0.00015994170680641573, "dbits": 314572800 } ] }, { "idx": 520, "layers": [ "model.layers.87.self_attn.k_proj", "model.layers.87.self_attn.v_proj" ], "candidates": [ { "dkld": 0.0004103614017367252, "dbits": 10485760 }, { "dkld": 0.00043580420315264545, "dbits": 20971520 }, { "dkld": -0.0002621199935674695, "dbits": 31457280 }, { "dkld": -7.895315065981068e-05, "dbits": 52428800 } ] }, { "idx": 521, "layers": [ "model.layers.87.self_attn.o_proj" ], "candidates": [ { "dkld": -0.000498458649963146, "dbits": 62914560 }, { "dkld": -0.0004175413399934852, "dbits": 125829120 }, { "dkld": -0.00031215958297253765, "dbits": 188743680 }, { "dkld": -0.00036138882860542454, "dbits": 314572800 } ] }, { "idx": 522, "layers": [ "model.layers.87.mlp.shared_experts.gate_proj", "model.layers.87.mlp.shared_experts.up_proj", "model.layers.87.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0002560422755777947, "dbits": 23592960 }, { "dkld": -0.00033239983022213815, "dbits": 47185920 }, { "dkld": -0.0003559943288564682, "dbits": 70778880 }, { "dkld": -0.00035025784745812416, "dbits": 117964800 } ] }, { "idx": 523, "layers": [ "model.layers.87.mlp.experts.0.gate_proj", "model.layers.87.mlp.experts.1.gate_proj", "model.layers.87.mlp.experts.2.gate_proj", "model.layers.87.mlp.experts.3.gate_proj", "model.layers.87.mlp.experts.4.gate_proj", "model.layers.87.mlp.experts.5.gate_proj", "model.layers.87.mlp.experts.6.gate_proj", "model.layers.87.mlp.experts.7.gate_proj", "model.layers.87.mlp.experts.8.gate_proj", "model.layers.87.mlp.experts.9.gate_proj", "model.layers.87.mlp.experts.10.gate_proj", "model.layers.87.mlp.experts.11.gate_proj", "model.layers.87.mlp.experts.12.gate_proj", "model.layers.87.mlp.experts.13.gate_proj", "model.layers.87.mlp.experts.14.gate_proj", "model.layers.87.mlp.experts.15.gate_proj", "model.layers.87.mlp.experts.16.gate_proj", "model.layers.87.mlp.experts.17.gate_proj", "model.layers.87.mlp.experts.18.gate_proj", "model.layers.87.mlp.experts.19.gate_proj", "model.layers.87.mlp.experts.20.gate_proj", "model.layers.87.mlp.experts.21.gate_proj", "model.layers.87.mlp.experts.22.gate_proj", "model.layers.87.mlp.experts.23.gate_proj", "model.layers.87.mlp.experts.24.gate_proj", "model.layers.87.mlp.experts.25.gate_proj", "model.layers.87.mlp.experts.26.gate_proj", "model.layers.87.mlp.experts.27.gate_proj", "model.layers.87.mlp.experts.28.gate_proj", "model.layers.87.mlp.experts.29.gate_proj", "model.layers.87.mlp.experts.30.gate_proj", "model.layers.87.mlp.experts.31.gate_proj", "model.layers.87.mlp.experts.32.gate_proj", "model.layers.87.mlp.experts.33.gate_proj", "model.layers.87.mlp.experts.34.gate_proj", "model.layers.87.mlp.experts.35.gate_proj", "model.layers.87.mlp.experts.36.gate_proj", "model.layers.87.mlp.experts.37.gate_proj", "model.layers.87.mlp.experts.38.gate_proj", "model.layers.87.mlp.experts.39.gate_proj", "model.layers.87.mlp.experts.40.gate_proj", "model.layers.87.mlp.experts.41.gate_proj", "model.layers.87.mlp.experts.42.gate_proj", "model.layers.87.mlp.experts.43.gate_proj", "model.layers.87.mlp.experts.44.gate_proj", "model.layers.87.mlp.experts.45.gate_proj", "model.layers.87.mlp.experts.46.gate_proj", "model.layers.87.mlp.experts.47.gate_proj", "model.layers.87.mlp.experts.48.gate_proj", "model.layers.87.mlp.experts.49.gate_proj", "model.layers.87.mlp.experts.50.gate_proj", "model.layers.87.mlp.experts.51.gate_proj", "model.layers.87.mlp.experts.52.gate_proj", "model.layers.87.mlp.experts.53.gate_proj", "model.layers.87.mlp.experts.54.gate_proj", "model.layers.87.mlp.experts.55.gate_proj", "model.layers.87.mlp.experts.56.gate_proj", "model.layers.87.mlp.experts.57.gate_proj", "model.layers.87.mlp.experts.58.gate_proj", "model.layers.87.mlp.experts.59.gate_proj", "model.layers.87.mlp.experts.60.gate_proj", "model.layers.87.mlp.experts.61.gate_proj", "model.layers.87.mlp.experts.62.gate_proj", "model.layers.87.mlp.experts.63.gate_proj", "model.layers.87.mlp.experts.64.gate_proj", "model.layers.87.mlp.experts.65.gate_proj", "model.layers.87.mlp.experts.66.gate_proj", "model.layers.87.mlp.experts.67.gate_proj", "model.layers.87.mlp.experts.68.gate_proj", "model.layers.87.mlp.experts.69.gate_proj", "model.layers.87.mlp.experts.70.gate_proj", "model.layers.87.mlp.experts.71.gate_proj", "model.layers.87.mlp.experts.72.gate_proj", "model.layers.87.mlp.experts.73.gate_proj", "model.layers.87.mlp.experts.74.gate_proj", "model.layers.87.mlp.experts.75.gate_proj", "model.layers.87.mlp.experts.76.gate_proj", "model.layers.87.mlp.experts.77.gate_proj", "model.layers.87.mlp.experts.78.gate_proj", "model.layers.87.mlp.experts.79.gate_proj", "model.layers.87.mlp.experts.80.gate_proj", "model.layers.87.mlp.experts.81.gate_proj", "model.layers.87.mlp.experts.82.gate_proj", "model.layers.87.mlp.experts.83.gate_proj", "model.layers.87.mlp.experts.84.gate_proj", "model.layers.87.mlp.experts.85.gate_proj", "model.layers.87.mlp.experts.86.gate_proj", "model.layers.87.mlp.experts.87.gate_proj", "model.layers.87.mlp.experts.88.gate_proj", "model.layers.87.mlp.experts.89.gate_proj", "model.layers.87.mlp.experts.90.gate_proj", "model.layers.87.mlp.experts.91.gate_proj", "model.layers.87.mlp.experts.92.gate_proj", "model.layers.87.mlp.experts.93.gate_proj", "model.layers.87.mlp.experts.94.gate_proj", "model.layers.87.mlp.experts.95.gate_proj", "model.layers.87.mlp.experts.96.gate_proj", "model.layers.87.mlp.experts.97.gate_proj", "model.layers.87.mlp.experts.98.gate_proj", "model.layers.87.mlp.experts.99.gate_proj", "model.layers.87.mlp.experts.100.gate_proj", "model.layers.87.mlp.experts.101.gate_proj", "model.layers.87.mlp.experts.102.gate_proj", "model.layers.87.mlp.experts.103.gate_proj", "model.layers.87.mlp.experts.104.gate_proj", "model.layers.87.mlp.experts.105.gate_proj", "model.layers.87.mlp.experts.106.gate_proj", "model.layers.87.mlp.experts.107.gate_proj", "model.layers.87.mlp.experts.108.gate_proj", "model.layers.87.mlp.experts.109.gate_proj", "model.layers.87.mlp.experts.110.gate_proj", "model.layers.87.mlp.experts.111.gate_proj", "model.layers.87.mlp.experts.112.gate_proj", "model.layers.87.mlp.experts.113.gate_proj", "model.layers.87.mlp.experts.114.gate_proj", "model.layers.87.mlp.experts.115.gate_proj", "model.layers.87.mlp.experts.116.gate_proj", "model.layers.87.mlp.experts.117.gate_proj", "model.layers.87.mlp.experts.118.gate_proj", "model.layers.87.mlp.experts.119.gate_proj", "model.layers.87.mlp.experts.120.gate_proj", "model.layers.87.mlp.experts.121.gate_proj", "model.layers.87.mlp.experts.122.gate_proj", "model.layers.87.mlp.experts.123.gate_proj", "model.layers.87.mlp.experts.124.gate_proj", "model.layers.87.mlp.experts.125.gate_proj", "model.layers.87.mlp.experts.126.gate_proj", "model.layers.87.mlp.experts.127.gate_proj", "model.layers.87.mlp.experts.128.gate_proj", "model.layers.87.mlp.experts.129.gate_proj", "model.layers.87.mlp.experts.130.gate_proj", "model.layers.87.mlp.experts.131.gate_proj", "model.layers.87.mlp.experts.132.gate_proj", "model.layers.87.mlp.experts.133.gate_proj", "model.layers.87.mlp.experts.134.gate_proj", "model.layers.87.mlp.experts.135.gate_proj", "model.layers.87.mlp.experts.136.gate_proj", "model.layers.87.mlp.experts.137.gate_proj", "model.layers.87.mlp.experts.138.gate_proj", "model.layers.87.mlp.experts.139.gate_proj", "model.layers.87.mlp.experts.140.gate_proj", "model.layers.87.mlp.experts.141.gate_proj", "model.layers.87.mlp.experts.142.gate_proj", "model.layers.87.mlp.experts.143.gate_proj", "model.layers.87.mlp.experts.144.gate_proj", "model.layers.87.mlp.experts.145.gate_proj", "model.layers.87.mlp.experts.146.gate_proj", "model.layers.87.mlp.experts.147.gate_proj", "model.layers.87.mlp.experts.148.gate_proj", "model.layers.87.mlp.experts.149.gate_proj", "model.layers.87.mlp.experts.150.gate_proj", "model.layers.87.mlp.experts.151.gate_proj", "model.layers.87.mlp.experts.152.gate_proj", "model.layers.87.mlp.experts.153.gate_proj", "model.layers.87.mlp.experts.154.gate_proj", "model.layers.87.mlp.experts.155.gate_proj", "model.layers.87.mlp.experts.156.gate_proj", "model.layers.87.mlp.experts.157.gate_proj", "model.layers.87.mlp.experts.158.gate_proj", "model.layers.87.mlp.experts.159.gate_proj", "model.layers.87.mlp.experts.0.up_proj", "model.layers.87.mlp.experts.1.up_proj", "model.layers.87.mlp.experts.2.up_proj", "model.layers.87.mlp.experts.3.up_proj", "model.layers.87.mlp.experts.4.up_proj", "model.layers.87.mlp.experts.5.up_proj", "model.layers.87.mlp.experts.6.up_proj", "model.layers.87.mlp.experts.7.up_proj", "model.layers.87.mlp.experts.8.up_proj", "model.layers.87.mlp.experts.9.up_proj", "model.layers.87.mlp.experts.10.up_proj", "model.layers.87.mlp.experts.11.up_proj", "model.layers.87.mlp.experts.12.up_proj", "model.layers.87.mlp.experts.13.up_proj", "model.layers.87.mlp.experts.14.up_proj", "model.layers.87.mlp.experts.15.up_proj", "model.layers.87.mlp.experts.16.up_proj", "model.layers.87.mlp.experts.17.up_proj", "model.layers.87.mlp.experts.18.up_proj", "model.layers.87.mlp.experts.19.up_proj", "model.layers.87.mlp.experts.20.up_proj", "model.layers.87.mlp.experts.21.up_proj", "model.layers.87.mlp.experts.22.up_proj", "model.layers.87.mlp.experts.23.up_proj", "model.layers.87.mlp.experts.24.up_proj", "model.layers.87.mlp.experts.25.up_proj", "model.layers.87.mlp.experts.26.up_proj", "model.layers.87.mlp.experts.27.up_proj", "model.layers.87.mlp.experts.28.up_proj", "model.layers.87.mlp.experts.29.up_proj", "model.layers.87.mlp.experts.30.up_proj", "model.layers.87.mlp.experts.31.up_proj", "model.layers.87.mlp.experts.32.up_proj", "model.layers.87.mlp.experts.33.up_proj", "model.layers.87.mlp.experts.34.up_proj", "model.layers.87.mlp.experts.35.up_proj", "model.layers.87.mlp.experts.36.up_proj", "model.layers.87.mlp.experts.37.up_proj", "model.layers.87.mlp.experts.38.up_proj", "model.layers.87.mlp.experts.39.up_proj", "model.layers.87.mlp.experts.40.up_proj", "model.layers.87.mlp.experts.41.up_proj", "model.layers.87.mlp.experts.42.up_proj", "model.layers.87.mlp.experts.43.up_proj", "model.layers.87.mlp.experts.44.up_proj", "model.layers.87.mlp.experts.45.up_proj", "model.layers.87.mlp.experts.46.up_proj", "model.layers.87.mlp.experts.47.up_proj", "model.layers.87.mlp.experts.48.up_proj", "model.layers.87.mlp.experts.49.up_proj", "model.layers.87.mlp.experts.50.up_proj", "model.layers.87.mlp.experts.51.up_proj", "model.layers.87.mlp.experts.52.up_proj", "model.layers.87.mlp.experts.53.up_proj", "model.layers.87.mlp.experts.54.up_proj", "model.layers.87.mlp.experts.55.up_proj", "model.layers.87.mlp.experts.56.up_proj", "model.layers.87.mlp.experts.57.up_proj", "model.layers.87.mlp.experts.58.up_proj", "model.layers.87.mlp.experts.59.up_proj", "model.layers.87.mlp.experts.60.up_proj", "model.layers.87.mlp.experts.61.up_proj", "model.layers.87.mlp.experts.62.up_proj", "model.layers.87.mlp.experts.63.up_proj", "model.layers.87.mlp.experts.64.up_proj", "model.layers.87.mlp.experts.65.up_proj", "model.layers.87.mlp.experts.66.up_proj", "model.layers.87.mlp.experts.67.up_proj", "model.layers.87.mlp.experts.68.up_proj", "model.layers.87.mlp.experts.69.up_proj", "model.layers.87.mlp.experts.70.up_proj", "model.layers.87.mlp.experts.71.up_proj", "model.layers.87.mlp.experts.72.up_proj", "model.layers.87.mlp.experts.73.up_proj", "model.layers.87.mlp.experts.74.up_proj", "model.layers.87.mlp.experts.75.up_proj", "model.layers.87.mlp.experts.76.up_proj", "model.layers.87.mlp.experts.77.up_proj", "model.layers.87.mlp.experts.78.up_proj", "model.layers.87.mlp.experts.79.up_proj", "model.layers.87.mlp.experts.80.up_proj", "model.layers.87.mlp.experts.81.up_proj", "model.layers.87.mlp.experts.82.up_proj", "model.layers.87.mlp.experts.83.up_proj", "model.layers.87.mlp.experts.84.up_proj", "model.layers.87.mlp.experts.85.up_proj", "model.layers.87.mlp.experts.86.up_proj", "model.layers.87.mlp.experts.87.up_proj", "model.layers.87.mlp.experts.88.up_proj", "model.layers.87.mlp.experts.89.up_proj", "model.layers.87.mlp.experts.90.up_proj", "model.layers.87.mlp.experts.91.up_proj", "model.layers.87.mlp.experts.92.up_proj", "model.layers.87.mlp.experts.93.up_proj", "model.layers.87.mlp.experts.94.up_proj", "model.layers.87.mlp.experts.95.up_proj", "model.layers.87.mlp.experts.96.up_proj", "model.layers.87.mlp.experts.97.up_proj", "model.layers.87.mlp.experts.98.up_proj", "model.layers.87.mlp.experts.99.up_proj", "model.layers.87.mlp.experts.100.up_proj", "model.layers.87.mlp.experts.101.up_proj", "model.layers.87.mlp.experts.102.up_proj", "model.layers.87.mlp.experts.103.up_proj", "model.layers.87.mlp.experts.104.up_proj", "model.layers.87.mlp.experts.105.up_proj", "model.layers.87.mlp.experts.106.up_proj", "model.layers.87.mlp.experts.107.up_proj", "model.layers.87.mlp.experts.108.up_proj", "model.layers.87.mlp.experts.109.up_proj", "model.layers.87.mlp.experts.110.up_proj", "model.layers.87.mlp.experts.111.up_proj", "model.layers.87.mlp.experts.112.up_proj", "model.layers.87.mlp.experts.113.up_proj", "model.layers.87.mlp.experts.114.up_proj", "model.layers.87.mlp.experts.115.up_proj", "model.layers.87.mlp.experts.116.up_proj", "model.layers.87.mlp.experts.117.up_proj", "model.layers.87.mlp.experts.118.up_proj", "model.layers.87.mlp.experts.119.up_proj", "model.layers.87.mlp.experts.120.up_proj", "model.layers.87.mlp.experts.121.up_proj", "model.layers.87.mlp.experts.122.up_proj", "model.layers.87.mlp.experts.123.up_proj", "model.layers.87.mlp.experts.124.up_proj", "model.layers.87.mlp.experts.125.up_proj", "model.layers.87.mlp.experts.126.up_proj", "model.layers.87.mlp.experts.127.up_proj", "model.layers.87.mlp.experts.128.up_proj", "model.layers.87.mlp.experts.129.up_proj", "model.layers.87.mlp.experts.130.up_proj", "model.layers.87.mlp.experts.131.up_proj", "model.layers.87.mlp.experts.132.up_proj", "model.layers.87.mlp.experts.133.up_proj", "model.layers.87.mlp.experts.134.up_proj", "model.layers.87.mlp.experts.135.up_proj", "model.layers.87.mlp.experts.136.up_proj", "model.layers.87.mlp.experts.137.up_proj", "model.layers.87.mlp.experts.138.up_proj", "model.layers.87.mlp.experts.139.up_proj", "model.layers.87.mlp.experts.140.up_proj", "model.layers.87.mlp.experts.141.up_proj", "model.layers.87.mlp.experts.142.up_proj", "model.layers.87.mlp.experts.143.up_proj", "model.layers.87.mlp.experts.144.up_proj", "model.layers.87.mlp.experts.145.up_proj", "model.layers.87.mlp.experts.146.up_proj", "model.layers.87.mlp.experts.147.up_proj", "model.layers.87.mlp.experts.148.up_proj", "model.layers.87.mlp.experts.149.up_proj", "model.layers.87.mlp.experts.150.up_proj", "model.layers.87.mlp.experts.151.up_proj", "model.layers.87.mlp.experts.152.up_proj", "model.layers.87.mlp.experts.153.up_proj", "model.layers.87.mlp.experts.154.up_proj", "model.layers.87.mlp.experts.155.up_proj", "model.layers.87.mlp.experts.156.up_proj", "model.layers.87.mlp.experts.157.up_proj", "model.layers.87.mlp.experts.158.up_proj", "model.layers.87.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -7.999204099179424e-05, "dbits": 2516582400 }, { "dkld": -0.00010029114782811321, "dbits": 5033164800 }, { "dkld": -9.526414796710847e-05, "dbits": 7549747200 }, { "dkld": -7.174713537097532e-05, "dbits": 12582912000 } ] }, { "idx": 524, "layers": [ "model.layers.87.mlp.experts.0.down_proj", "model.layers.87.mlp.experts.1.down_proj", "model.layers.87.mlp.experts.2.down_proj", "model.layers.87.mlp.experts.3.down_proj", "model.layers.87.mlp.experts.4.down_proj", "model.layers.87.mlp.experts.5.down_proj", "model.layers.87.mlp.experts.6.down_proj", "model.layers.87.mlp.experts.7.down_proj", "model.layers.87.mlp.experts.8.down_proj", "model.layers.87.mlp.experts.9.down_proj", "model.layers.87.mlp.experts.10.down_proj", "model.layers.87.mlp.experts.11.down_proj", "model.layers.87.mlp.experts.12.down_proj", "model.layers.87.mlp.experts.13.down_proj", "model.layers.87.mlp.experts.14.down_proj", "model.layers.87.mlp.experts.15.down_proj", "model.layers.87.mlp.experts.16.down_proj", "model.layers.87.mlp.experts.17.down_proj", "model.layers.87.mlp.experts.18.down_proj", "model.layers.87.mlp.experts.19.down_proj", "model.layers.87.mlp.experts.20.down_proj", "model.layers.87.mlp.experts.21.down_proj", "model.layers.87.mlp.experts.22.down_proj", "model.layers.87.mlp.experts.23.down_proj", "model.layers.87.mlp.experts.24.down_proj", "model.layers.87.mlp.experts.25.down_proj", "model.layers.87.mlp.experts.26.down_proj", "model.layers.87.mlp.experts.27.down_proj", "model.layers.87.mlp.experts.28.down_proj", "model.layers.87.mlp.experts.29.down_proj", "model.layers.87.mlp.experts.30.down_proj", "model.layers.87.mlp.experts.31.down_proj", "model.layers.87.mlp.experts.32.down_proj", "model.layers.87.mlp.experts.33.down_proj", "model.layers.87.mlp.experts.34.down_proj", "model.layers.87.mlp.experts.35.down_proj", "model.layers.87.mlp.experts.36.down_proj", "model.layers.87.mlp.experts.37.down_proj", "model.layers.87.mlp.experts.38.down_proj", "model.layers.87.mlp.experts.39.down_proj", "model.layers.87.mlp.experts.40.down_proj", "model.layers.87.mlp.experts.41.down_proj", "model.layers.87.mlp.experts.42.down_proj", "model.layers.87.mlp.experts.43.down_proj", "model.layers.87.mlp.experts.44.down_proj", "model.layers.87.mlp.experts.45.down_proj", "model.layers.87.mlp.experts.46.down_proj", "model.layers.87.mlp.experts.47.down_proj", "model.layers.87.mlp.experts.48.down_proj", "model.layers.87.mlp.experts.49.down_proj", "model.layers.87.mlp.experts.50.down_proj", "model.layers.87.mlp.experts.51.down_proj", "model.layers.87.mlp.experts.52.down_proj", "model.layers.87.mlp.experts.53.down_proj", "model.layers.87.mlp.experts.54.down_proj", "model.layers.87.mlp.experts.55.down_proj", "model.layers.87.mlp.experts.56.down_proj", "model.layers.87.mlp.experts.57.down_proj", "model.layers.87.mlp.experts.58.down_proj", "model.layers.87.mlp.experts.59.down_proj", "model.layers.87.mlp.experts.60.down_proj", "model.layers.87.mlp.experts.61.down_proj", "model.layers.87.mlp.experts.62.down_proj", "model.layers.87.mlp.experts.63.down_proj", "model.layers.87.mlp.experts.64.down_proj", "model.layers.87.mlp.experts.65.down_proj", "model.layers.87.mlp.experts.66.down_proj", "model.layers.87.mlp.experts.67.down_proj", "model.layers.87.mlp.experts.68.down_proj", "model.layers.87.mlp.experts.69.down_proj", "model.layers.87.mlp.experts.70.down_proj", "model.layers.87.mlp.experts.71.down_proj", "model.layers.87.mlp.experts.72.down_proj", "model.layers.87.mlp.experts.73.down_proj", "model.layers.87.mlp.experts.74.down_proj", "model.layers.87.mlp.experts.75.down_proj", "model.layers.87.mlp.experts.76.down_proj", "model.layers.87.mlp.experts.77.down_proj", "model.layers.87.mlp.experts.78.down_proj", "model.layers.87.mlp.experts.79.down_proj", "model.layers.87.mlp.experts.80.down_proj", "model.layers.87.mlp.experts.81.down_proj", "model.layers.87.mlp.experts.82.down_proj", "model.layers.87.mlp.experts.83.down_proj", "model.layers.87.mlp.experts.84.down_proj", "model.layers.87.mlp.experts.85.down_proj", "model.layers.87.mlp.experts.86.down_proj", "model.layers.87.mlp.experts.87.down_proj", "model.layers.87.mlp.experts.88.down_proj", "model.layers.87.mlp.experts.89.down_proj", "model.layers.87.mlp.experts.90.down_proj", "model.layers.87.mlp.experts.91.down_proj", "model.layers.87.mlp.experts.92.down_proj", "model.layers.87.mlp.experts.93.down_proj", "model.layers.87.mlp.experts.94.down_proj", "model.layers.87.mlp.experts.95.down_proj", "model.layers.87.mlp.experts.96.down_proj", "model.layers.87.mlp.experts.97.down_proj", "model.layers.87.mlp.experts.98.down_proj", "model.layers.87.mlp.experts.99.down_proj", "model.layers.87.mlp.experts.100.down_proj", "model.layers.87.mlp.experts.101.down_proj", "model.layers.87.mlp.experts.102.down_proj", "model.layers.87.mlp.experts.103.down_proj", "model.layers.87.mlp.experts.104.down_proj", "model.layers.87.mlp.experts.105.down_proj", "model.layers.87.mlp.experts.106.down_proj", "model.layers.87.mlp.experts.107.down_proj", "model.layers.87.mlp.experts.108.down_proj", "model.layers.87.mlp.experts.109.down_proj", "model.layers.87.mlp.experts.110.down_proj", "model.layers.87.mlp.experts.111.down_proj", "model.layers.87.mlp.experts.112.down_proj", "model.layers.87.mlp.experts.113.down_proj", "model.layers.87.mlp.experts.114.down_proj", "model.layers.87.mlp.experts.115.down_proj", "model.layers.87.mlp.experts.116.down_proj", "model.layers.87.mlp.experts.117.down_proj", "model.layers.87.mlp.experts.118.down_proj", "model.layers.87.mlp.experts.119.down_proj", "model.layers.87.mlp.experts.120.down_proj", "model.layers.87.mlp.experts.121.down_proj", "model.layers.87.mlp.experts.122.down_proj", "model.layers.87.mlp.experts.123.down_proj", "model.layers.87.mlp.experts.124.down_proj", "model.layers.87.mlp.experts.125.down_proj", "model.layers.87.mlp.experts.126.down_proj", "model.layers.87.mlp.experts.127.down_proj", "model.layers.87.mlp.experts.128.down_proj", "model.layers.87.mlp.experts.129.down_proj", "model.layers.87.mlp.experts.130.down_proj", "model.layers.87.mlp.experts.131.down_proj", "model.layers.87.mlp.experts.132.down_proj", "model.layers.87.mlp.experts.133.down_proj", "model.layers.87.mlp.experts.134.down_proj", "model.layers.87.mlp.experts.135.down_proj", "model.layers.87.mlp.experts.136.down_proj", "model.layers.87.mlp.experts.137.down_proj", "model.layers.87.mlp.experts.138.down_proj", "model.layers.87.mlp.experts.139.down_proj", "model.layers.87.mlp.experts.140.down_proj", "model.layers.87.mlp.experts.141.down_proj", "model.layers.87.mlp.experts.142.down_proj", "model.layers.87.mlp.experts.143.down_proj", "model.layers.87.mlp.experts.144.down_proj", "model.layers.87.mlp.experts.145.down_proj", "model.layers.87.mlp.experts.146.down_proj", "model.layers.87.mlp.experts.147.down_proj", "model.layers.87.mlp.experts.148.down_proj", "model.layers.87.mlp.experts.149.down_proj", "model.layers.87.mlp.experts.150.down_proj", "model.layers.87.mlp.experts.151.down_proj", "model.layers.87.mlp.experts.152.down_proj", "model.layers.87.mlp.experts.153.down_proj", "model.layers.87.mlp.experts.154.down_proj", "model.layers.87.mlp.experts.155.down_proj", "model.layers.87.mlp.experts.156.down_proj", "model.layers.87.mlp.experts.157.down_proj", "model.layers.87.mlp.experts.158.down_proj", "model.layers.87.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -9.855292737484533e-05, "dbits": 1258291200 }, { "dkld": -0.00011342661455274183, "dbits": 2516582400 }, { "dkld": -0.00011122673749924816, "dbits": 3774873600 }, { "dkld": -0.00011763712391257841, "dbits": 6291456000 } ] }, { "idx": 525, "layers": [ "model.layers.88.self_attn.q_proj" ], "candidates": [ { "dkld": -5.0496309995651245e-05, "dbits": 62914560 }, { "dkld": -6.734980270267643e-05, "dbits": 125829120 }, { "dkld": -5.3775310516368524e-05, "dbits": 188743680 }, { "dkld": -7.757460698486207e-05, "dbits": 314572800 } ] }, { "idx": 526, "layers": [ "model.layers.88.self_attn.k_proj", "model.layers.88.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007289796136319693, "dbits": 10485760 }, { "dkld": -0.0005119074136018809, "dbits": 20971520 }, { "dkld": -0.0005088381469249836, "dbits": 31457280 }, { "dkld": -0.0005718267522752396, "dbits": 52428800 } ] }, { "idx": 527, "layers": [ "model.layers.88.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00011662887409330525, "dbits": 62914560 }, { "dkld": -0.0001940020360052669, "dbits": 125829120 }, { "dkld": -0.00019493941217661737, "dbits": 188743680 }, { "dkld": -0.00018016155809164047, "dbits": 314572800 } ] }, { "idx": 528, "layers": [ "model.layers.88.mlp.shared_experts.gate_proj", "model.layers.88.mlp.shared_experts.up_proj", "model.layers.88.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -7.95499421656215e-05, "dbits": 23592960 }, { "dkld": -0.00017855484038592495, "dbits": 47185920 }, { "dkld": -0.00022679669782520173, "dbits": 70778880 }, { "dkld": -0.00023324713110924045, "dbits": 117964800 } ] }, { "idx": 529, "layers": [ "model.layers.88.mlp.experts.0.gate_proj", "model.layers.88.mlp.experts.1.gate_proj", "model.layers.88.mlp.experts.2.gate_proj", "model.layers.88.mlp.experts.3.gate_proj", "model.layers.88.mlp.experts.4.gate_proj", "model.layers.88.mlp.experts.5.gate_proj", "model.layers.88.mlp.experts.6.gate_proj", "model.layers.88.mlp.experts.7.gate_proj", "model.layers.88.mlp.experts.8.gate_proj", "model.layers.88.mlp.experts.9.gate_proj", "model.layers.88.mlp.experts.10.gate_proj", "model.layers.88.mlp.experts.11.gate_proj", "model.layers.88.mlp.experts.12.gate_proj", "model.layers.88.mlp.experts.13.gate_proj", "model.layers.88.mlp.experts.14.gate_proj", "model.layers.88.mlp.experts.15.gate_proj", "model.layers.88.mlp.experts.16.gate_proj", "model.layers.88.mlp.experts.17.gate_proj", "model.layers.88.mlp.experts.18.gate_proj", "model.layers.88.mlp.experts.19.gate_proj", "model.layers.88.mlp.experts.20.gate_proj", "model.layers.88.mlp.experts.21.gate_proj", "model.layers.88.mlp.experts.22.gate_proj", "model.layers.88.mlp.experts.23.gate_proj", "model.layers.88.mlp.experts.24.gate_proj", "model.layers.88.mlp.experts.25.gate_proj", "model.layers.88.mlp.experts.26.gate_proj", "model.layers.88.mlp.experts.27.gate_proj", "model.layers.88.mlp.experts.28.gate_proj", "model.layers.88.mlp.experts.29.gate_proj", "model.layers.88.mlp.experts.30.gate_proj", "model.layers.88.mlp.experts.31.gate_proj", "model.layers.88.mlp.experts.32.gate_proj", "model.layers.88.mlp.experts.33.gate_proj", "model.layers.88.mlp.experts.34.gate_proj", "model.layers.88.mlp.experts.35.gate_proj", "model.layers.88.mlp.experts.36.gate_proj", "model.layers.88.mlp.experts.37.gate_proj", "model.layers.88.mlp.experts.38.gate_proj", "model.layers.88.mlp.experts.39.gate_proj", "model.layers.88.mlp.experts.40.gate_proj", "model.layers.88.mlp.experts.41.gate_proj", "model.layers.88.mlp.experts.42.gate_proj", "model.layers.88.mlp.experts.43.gate_proj", "model.layers.88.mlp.experts.44.gate_proj", "model.layers.88.mlp.experts.45.gate_proj", "model.layers.88.mlp.experts.46.gate_proj", "model.layers.88.mlp.experts.47.gate_proj", "model.layers.88.mlp.experts.48.gate_proj", "model.layers.88.mlp.experts.49.gate_proj", "model.layers.88.mlp.experts.50.gate_proj", "model.layers.88.mlp.experts.51.gate_proj", "model.layers.88.mlp.experts.52.gate_proj", "model.layers.88.mlp.experts.53.gate_proj", "model.layers.88.mlp.experts.54.gate_proj", "model.layers.88.mlp.experts.55.gate_proj", "model.layers.88.mlp.experts.56.gate_proj", "model.layers.88.mlp.experts.57.gate_proj", "model.layers.88.mlp.experts.58.gate_proj", "model.layers.88.mlp.experts.59.gate_proj", "model.layers.88.mlp.experts.60.gate_proj", "model.layers.88.mlp.experts.61.gate_proj", "model.layers.88.mlp.experts.62.gate_proj", "model.layers.88.mlp.experts.63.gate_proj", "model.layers.88.mlp.experts.64.gate_proj", "model.layers.88.mlp.experts.65.gate_proj", "model.layers.88.mlp.experts.66.gate_proj", "model.layers.88.mlp.experts.67.gate_proj", "model.layers.88.mlp.experts.68.gate_proj", "model.layers.88.mlp.experts.69.gate_proj", "model.layers.88.mlp.experts.70.gate_proj", "model.layers.88.mlp.experts.71.gate_proj", "model.layers.88.mlp.experts.72.gate_proj", "model.layers.88.mlp.experts.73.gate_proj", "model.layers.88.mlp.experts.74.gate_proj", "model.layers.88.mlp.experts.75.gate_proj", "model.layers.88.mlp.experts.76.gate_proj", "model.layers.88.mlp.experts.77.gate_proj", "model.layers.88.mlp.experts.78.gate_proj", "model.layers.88.mlp.experts.79.gate_proj", "model.layers.88.mlp.experts.80.gate_proj", "model.layers.88.mlp.experts.81.gate_proj", "model.layers.88.mlp.experts.82.gate_proj", "model.layers.88.mlp.experts.83.gate_proj", "model.layers.88.mlp.experts.84.gate_proj", "model.layers.88.mlp.experts.85.gate_proj", "model.layers.88.mlp.experts.86.gate_proj", "model.layers.88.mlp.experts.87.gate_proj", "model.layers.88.mlp.experts.88.gate_proj", "model.layers.88.mlp.experts.89.gate_proj", "model.layers.88.mlp.experts.90.gate_proj", "model.layers.88.mlp.experts.91.gate_proj", "model.layers.88.mlp.experts.92.gate_proj", "model.layers.88.mlp.experts.93.gate_proj", "model.layers.88.mlp.experts.94.gate_proj", "model.layers.88.mlp.experts.95.gate_proj", "model.layers.88.mlp.experts.96.gate_proj", "model.layers.88.mlp.experts.97.gate_proj", "model.layers.88.mlp.experts.98.gate_proj", "model.layers.88.mlp.experts.99.gate_proj", "model.layers.88.mlp.experts.100.gate_proj", "model.layers.88.mlp.experts.101.gate_proj", "model.layers.88.mlp.experts.102.gate_proj", "model.layers.88.mlp.experts.103.gate_proj", "model.layers.88.mlp.experts.104.gate_proj", "model.layers.88.mlp.experts.105.gate_proj", "model.layers.88.mlp.experts.106.gate_proj", "model.layers.88.mlp.experts.107.gate_proj", "model.layers.88.mlp.experts.108.gate_proj", "model.layers.88.mlp.experts.109.gate_proj", "model.layers.88.mlp.experts.110.gate_proj", "model.layers.88.mlp.experts.111.gate_proj", "model.layers.88.mlp.experts.112.gate_proj", "model.layers.88.mlp.experts.113.gate_proj", "model.layers.88.mlp.experts.114.gate_proj", "model.layers.88.mlp.experts.115.gate_proj", "model.layers.88.mlp.experts.116.gate_proj", "model.layers.88.mlp.experts.117.gate_proj", "model.layers.88.mlp.experts.118.gate_proj", "model.layers.88.mlp.experts.119.gate_proj", "model.layers.88.mlp.experts.120.gate_proj", "model.layers.88.mlp.experts.121.gate_proj", "model.layers.88.mlp.experts.122.gate_proj", "model.layers.88.mlp.experts.123.gate_proj", "model.layers.88.mlp.experts.124.gate_proj", "model.layers.88.mlp.experts.125.gate_proj", "model.layers.88.mlp.experts.126.gate_proj", "model.layers.88.mlp.experts.127.gate_proj", "model.layers.88.mlp.experts.128.gate_proj", "model.layers.88.mlp.experts.129.gate_proj", "model.layers.88.mlp.experts.130.gate_proj", "model.layers.88.mlp.experts.131.gate_proj", "model.layers.88.mlp.experts.132.gate_proj", "model.layers.88.mlp.experts.133.gate_proj", "model.layers.88.mlp.experts.134.gate_proj", "model.layers.88.mlp.experts.135.gate_proj", "model.layers.88.mlp.experts.136.gate_proj", "model.layers.88.mlp.experts.137.gate_proj", "model.layers.88.mlp.experts.138.gate_proj", "model.layers.88.mlp.experts.139.gate_proj", "model.layers.88.mlp.experts.140.gate_proj", "model.layers.88.mlp.experts.141.gate_proj", "model.layers.88.mlp.experts.142.gate_proj", "model.layers.88.mlp.experts.143.gate_proj", "model.layers.88.mlp.experts.144.gate_proj", "model.layers.88.mlp.experts.145.gate_proj", "model.layers.88.mlp.experts.146.gate_proj", "model.layers.88.mlp.experts.147.gate_proj", "model.layers.88.mlp.experts.148.gate_proj", "model.layers.88.mlp.experts.149.gate_proj", "model.layers.88.mlp.experts.150.gate_proj", "model.layers.88.mlp.experts.151.gate_proj", "model.layers.88.mlp.experts.152.gate_proj", "model.layers.88.mlp.experts.153.gate_proj", "model.layers.88.mlp.experts.154.gate_proj", "model.layers.88.mlp.experts.155.gate_proj", "model.layers.88.mlp.experts.156.gate_proj", "model.layers.88.mlp.experts.157.gate_proj", "model.layers.88.mlp.experts.158.gate_proj", "model.layers.88.mlp.experts.159.gate_proj", "model.layers.88.mlp.experts.0.up_proj", "model.layers.88.mlp.experts.1.up_proj", "model.layers.88.mlp.experts.2.up_proj", "model.layers.88.mlp.experts.3.up_proj", "model.layers.88.mlp.experts.4.up_proj", "model.layers.88.mlp.experts.5.up_proj", "model.layers.88.mlp.experts.6.up_proj", "model.layers.88.mlp.experts.7.up_proj", "model.layers.88.mlp.experts.8.up_proj", "model.layers.88.mlp.experts.9.up_proj", "model.layers.88.mlp.experts.10.up_proj", "model.layers.88.mlp.experts.11.up_proj", "model.layers.88.mlp.experts.12.up_proj", "model.layers.88.mlp.experts.13.up_proj", "model.layers.88.mlp.experts.14.up_proj", "model.layers.88.mlp.experts.15.up_proj", "model.layers.88.mlp.experts.16.up_proj", "model.layers.88.mlp.experts.17.up_proj", "model.layers.88.mlp.experts.18.up_proj", "model.layers.88.mlp.experts.19.up_proj", "model.layers.88.mlp.experts.20.up_proj", "model.layers.88.mlp.experts.21.up_proj", "model.layers.88.mlp.experts.22.up_proj", "model.layers.88.mlp.experts.23.up_proj", "model.layers.88.mlp.experts.24.up_proj", "model.layers.88.mlp.experts.25.up_proj", "model.layers.88.mlp.experts.26.up_proj", "model.layers.88.mlp.experts.27.up_proj", "model.layers.88.mlp.experts.28.up_proj", "model.layers.88.mlp.experts.29.up_proj", "model.layers.88.mlp.experts.30.up_proj", "model.layers.88.mlp.experts.31.up_proj", "model.layers.88.mlp.experts.32.up_proj", "model.layers.88.mlp.experts.33.up_proj", "model.layers.88.mlp.experts.34.up_proj", "model.layers.88.mlp.experts.35.up_proj", "model.layers.88.mlp.experts.36.up_proj", "model.layers.88.mlp.experts.37.up_proj", "model.layers.88.mlp.experts.38.up_proj", "model.layers.88.mlp.experts.39.up_proj", "model.layers.88.mlp.experts.40.up_proj", "model.layers.88.mlp.experts.41.up_proj", "model.layers.88.mlp.experts.42.up_proj", "model.layers.88.mlp.experts.43.up_proj", "model.layers.88.mlp.experts.44.up_proj", "model.layers.88.mlp.experts.45.up_proj", "model.layers.88.mlp.experts.46.up_proj", "model.layers.88.mlp.experts.47.up_proj", "model.layers.88.mlp.experts.48.up_proj", "model.layers.88.mlp.experts.49.up_proj", "model.layers.88.mlp.experts.50.up_proj", "model.layers.88.mlp.experts.51.up_proj", "model.layers.88.mlp.experts.52.up_proj", "model.layers.88.mlp.experts.53.up_proj", "model.layers.88.mlp.experts.54.up_proj", "model.layers.88.mlp.experts.55.up_proj", "model.layers.88.mlp.experts.56.up_proj", "model.layers.88.mlp.experts.57.up_proj", "model.layers.88.mlp.experts.58.up_proj", "model.layers.88.mlp.experts.59.up_proj", "model.layers.88.mlp.experts.60.up_proj", "model.layers.88.mlp.experts.61.up_proj", "model.layers.88.mlp.experts.62.up_proj", "model.layers.88.mlp.experts.63.up_proj", "model.layers.88.mlp.experts.64.up_proj", "model.layers.88.mlp.experts.65.up_proj", "model.layers.88.mlp.experts.66.up_proj", "model.layers.88.mlp.experts.67.up_proj", "model.layers.88.mlp.experts.68.up_proj", "model.layers.88.mlp.experts.69.up_proj", "model.layers.88.mlp.experts.70.up_proj", "model.layers.88.mlp.experts.71.up_proj", "model.layers.88.mlp.experts.72.up_proj", "model.layers.88.mlp.experts.73.up_proj", "model.layers.88.mlp.experts.74.up_proj", "model.layers.88.mlp.experts.75.up_proj", "model.layers.88.mlp.experts.76.up_proj", "model.layers.88.mlp.experts.77.up_proj", "model.layers.88.mlp.experts.78.up_proj", "model.layers.88.mlp.experts.79.up_proj", "model.layers.88.mlp.experts.80.up_proj", "model.layers.88.mlp.experts.81.up_proj", "model.layers.88.mlp.experts.82.up_proj", "model.layers.88.mlp.experts.83.up_proj", "model.layers.88.mlp.experts.84.up_proj", "model.layers.88.mlp.experts.85.up_proj", "model.layers.88.mlp.experts.86.up_proj", "model.layers.88.mlp.experts.87.up_proj", "model.layers.88.mlp.experts.88.up_proj", "model.layers.88.mlp.experts.89.up_proj", "model.layers.88.mlp.experts.90.up_proj", "model.layers.88.mlp.experts.91.up_proj", "model.layers.88.mlp.experts.92.up_proj", "model.layers.88.mlp.experts.93.up_proj", "model.layers.88.mlp.experts.94.up_proj", "model.layers.88.mlp.experts.95.up_proj", "model.layers.88.mlp.experts.96.up_proj", "model.layers.88.mlp.experts.97.up_proj", "model.layers.88.mlp.experts.98.up_proj", "model.layers.88.mlp.experts.99.up_proj", "model.layers.88.mlp.experts.100.up_proj", "model.layers.88.mlp.experts.101.up_proj", "model.layers.88.mlp.experts.102.up_proj", "model.layers.88.mlp.experts.103.up_proj", "model.layers.88.mlp.experts.104.up_proj", "model.layers.88.mlp.experts.105.up_proj", "model.layers.88.mlp.experts.106.up_proj", "model.layers.88.mlp.experts.107.up_proj", "model.layers.88.mlp.experts.108.up_proj", "model.layers.88.mlp.experts.109.up_proj", "model.layers.88.mlp.experts.110.up_proj", "model.layers.88.mlp.experts.111.up_proj", "model.layers.88.mlp.experts.112.up_proj", "model.layers.88.mlp.experts.113.up_proj", "model.layers.88.mlp.experts.114.up_proj", "model.layers.88.mlp.experts.115.up_proj", "model.layers.88.mlp.experts.116.up_proj", "model.layers.88.mlp.experts.117.up_proj", "model.layers.88.mlp.experts.118.up_proj", "model.layers.88.mlp.experts.119.up_proj", "model.layers.88.mlp.experts.120.up_proj", "model.layers.88.mlp.experts.121.up_proj", "model.layers.88.mlp.experts.122.up_proj", "model.layers.88.mlp.experts.123.up_proj", "model.layers.88.mlp.experts.124.up_proj", "model.layers.88.mlp.experts.125.up_proj", "model.layers.88.mlp.experts.126.up_proj", "model.layers.88.mlp.experts.127.up_proj", "model.layers.88.mlp.experts.128.up_proj", "model.layers.88.mlp.experts.129.up_proj", "model.layers.88.mlp.experts.130.up_proj", "model.layers.88.mlp.experts.131.up_proj", "model.layers.88.mlp.experts.132.up_proj", "model.layers.88.mlp.experts.133.up_proj", "model.layers.88.mlp.experts.134.up_proj", "model.layers.88.mlp.experts.135.up_proj", "model.layers.88.mlp.experts.136.up_proj", "model.layers.88.mlp.experts.137.up_proj", "model.layers.88.mlp.experts.138.up_proj", "model.layers.88.mlp.experts.139.up_proj", "model.layers.88.mlp.experts.140.up_proj", "model.layers.88.mlp.experts.141.up_proj", "model.layers.88.mlp.experts.142.up_proj", "model.layers.88.mlp.experts.143.up_proj", "model.layers.88.mlp.experts.144.up_proj", "model.layers.88.mlp.experts.145.up_proj", "model.layers.88.mlp.experts.146.up_proj", "model.layers.88.mlp.experts.147.up_proj", "model.layers.88.mlp.experts.148.up_proj", "model.layers.88.mlp.experts.149.up_proj", "model.layers.88.mlp.experts.150.up_proj", "model.layers.88.mlp.experts.151.up_proj", "model.layers.88.mlp.experts.152.up_proj", "model.layers.88.mlp.experts.153.up_proj", "model.layers.88.mlp.experts.154.up_proj", "model.layers.88.mlp.experts.155.up_proj", "model.layers.88.mlp.experts.156.up_proj", "model.layers.88.mlp.experts.157.up_proj", "model.layers.88.mlp.experts.158.up_proj", "model.layers.88.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00011409930884838382, "dbits": 2516582400 }, { "dkld": -0.00013692835345864574, "dbits": 5033164800 }, { "dkld": -0.00011942600831389705, "dbits": 7549747200 }, { "dkld": -0.00013306103646755496, "dbits": 12582912000 } ] }, { "idx": 530, "layers": [ "model.layers.88.mlp.experts.0.down_proj", "model.layers.88.mlp.experts.1.down_proj", "model.layers.88.mlp.experts.2.down_proj", "model.layers.88.mlp.experts.3.down_proj", "model.layers.88.mlp.experts.4.down_proj", "model.layers.88.mlp.experts.5.down_proj", "model.layers.88.mlp.experts.6.down_proj", "model.layers.88.mlp.experts.7.down_proj", "model.layers.88.mlp.experts.8.down_proj", "model.layers.88.mlp.experts.9.down_proj", "model.layers.88.mlp.experts.10.down_proj", "model.layers.88.mlp.experts.11.down_proj", "model.layers.88.mlp.experts.12.down_proj", "model.layers.88.mlp.experts.13.down_proj", "model.layers.88.mlp.experts.14.down_proj", "model.layers.88.mlp.experts.15.down_proj", "model.layers.88.mlp.experts.16.down_proj", "model.layers.88.mlp.experts.17.down_proj", "model.layers.88.mlp.experts.18.down_proj", "model.layers.88.mlp.experts.19.down_proj", "model.layers.88.mlp.experts.20.down_proj", "model.layers.88.mlp.experts.21.down_proj", "model.layers.88.mlp.experts.22.down_proj", "model.layers.88.mlp.experts.23.down_proj", "model.layers.88.mlp.experts.24.down_proj", "model.layers.88.mlp.experts.25.down_proj", "model.layers.88.mlp.experts.26.down_proj", "model.layers.88.mlp.experts.27.down_proj", "model.layers.88.mlp.experts.28.down_proj", "model.layers.88.mlp.experts.29.down_proj", "model.layers.88.mlp.experts.30.down_proj", "model.layers.88.mlp.experts.31.down_proj", "model.layers.88.mlp.experts.32.down_proj", "model.layers.88.mlp.experts.33.down_proj", "model.layers.88.mlp.experts.34.down_proj", "model.layers.88.mlp.experts.35.down_proj", "model.layers.88.mlp.experts.36.down_proj", "model.layers.88.mlp.experts.37.down_proj", "model.layers.88.mlp.experts.38.down_proj", "model.layers.88.mlp.experts.39.down_proj", "model.layers.88.mlp.experts.40.down_proj", "model.layers.88.mlp.experts.41.down_proj", "model.layers.88.mlp.experts.42.down_proj", "model.layers.88.mlp.experts.43.down_proj", "model.layers.88.mlp.experts.44.down_proj", "model.layers.88.mlp.experts.45.down_proj", "model.layers.88.mlp.experts.46.down_proj", "model.layers.88.mlp.experts.47.down_proj", "model.layers.88.mlp.experts.48.down_proj", "model.layers.88.mlp.experts.49.down_proj", "model.layers.88.mlp.experts.50.down_proj", "model.layers.88.mlp.experts.51.down_proj", "model.layers.88.mlp.experts.52.down_proj", "model.layers.88.mlp.experts.53.down_proj", "model.layers.88.mlp.experts.54.down_proj", "model.layers.88.mlp.experts.55.down_proj", "model.layers.88.mlp.experts.56.down_proj", "model.layers.88.mlp.experts.57.down_proj", "model.layers.88.mlp.experts.58.down_proj", "model.layers.88.mlp.experts.59.down_proj", "model.layers.88.mlp.experts.60.down_proj", "model.layers.88.mlp.experts.61.down_proj", "model.layers.88.mlp.experts.62.down_proj", "model.layers.88.mlp.experts.63.down_proj", "model.layers.88.mlp.experts.64.down_proj", "model.layers.88.mlp.experts.65.down_proj", "model.layers.88.mlp.experts.66.down_proj", "model.layers.88.mlp.experts.67.down_proj", "model.layers.88.mlp.experts.68.down_proj", "model.layers.88.mlp.experts.69.down_proj", "model.layers.88.mlp.experts.70.down_proj", "model.layers.88.mlp.experts.71.down_proj", "model.layers.88.mlp.experts.72.down_proj", "model.layers.88.mlp.experts.73.down_proj", "model.layers.88.mlp.experts.74.down_proj", "model.layers.88.mlp.experts.75.down_proj", "model.layers.88.mlp.experts.76.down_proj", "model.layers.88.mlp.experts.77.down_proj", "model.layers.88.mlp.experts.78.down_proj", "model.layers.88.mlp.experts.79.down_proj", "model.layers.88.mlp.experts.80.down_proj", "model.layers.88.mlp.experts.81.down_proj", "model.layers.88.mlp.experts.82.down_proj", "model.layers.88.mlp.experts.83.down_proj", "model.layers.88.mlp.experts.84.down_proj", "model.layers.88.mlp.experts.85.down_proj", "model.layers.88.mlp.experts.86.down_proj", "model.layers.88.mlp.experts.87.down_proj", "model.layers.88.mlp.experts.88.down_proj", "model.layers.88.mlp.experts.89.down_proj", "model.layers.88.mlp.experts.90.down_proj", "model.layers.88.mlp.experts.91.down_proj", "model.layers.88.mlp.experts.92.down_proj", "model.layers.88.mlp.experts.93.down_proj", "model.layers.88.mlp.experts.94.down_proj", "model.layers.88.mlp.experts.95.down_proj", "model.layers.88.mlp.experts.96.down_proj", "model.layers.88.mlp.experts.97.down_proj", "model.layers.88.mlp.experts.98.down_proj", "model.layers.88.mlp.experts.99.down_proj", "model.layers.88.mlp.experts.100.down_proj", "model.layers.88.mlp.experts.101.down_proj", "model.layers.88.mlp.experts.102.down_proj", "model.layers.88.mlp.experts.103.down_proj", "model.layers.88.mlp.experts.104.down_proj", "model.layers.88.mlp.experts.105.down_proj", "model.layers.88.mlp.experts.106.down_proj", "model.layers.88.mlp.experts.107.down_proj", "model.layers.88.mlp.experts.108.down_proj", "model.layers.88.mlp.experts.109.down_proj", "model.layers.88.mlp.experts.110.down_proj", "model.layers.88.mlp.experts.111.down_proj", "model.layers.88.mlp.experts.112.down_proj", "model.layers.88.mlp.experts.113.down_proj", "model.layers.88.mlp.experts.114.down_proj", "model.layers.88.mlp.experts.115.down_proj", "model.layers.88.mlp.experts.116.down_proj", "model.layers.88.mlp.experts.117.down_proj", "model.layers.88.mlp.experts.118.down_proj", "model.layers.88.mlp.experts.119.down_proj", "model.layers.88.mlp.experts.120.down_proj", "model.layers.88.mlp.experts.121.down_proj", "model.layers.88.mlp.experts.122.down_proj", "model.layers.88.mlp.experts.123.down_proj", "model.layers.88.mlp.experts.124.down_proj", "model.layers.88.mlp.experts.125.down_proj", "model.layers.88.mlp.experts.126.down_proj", "model.layers.88.mlp.experts.127.down_proj", "model.layers.88.mlp.experts.128.down_proj", "model.layers.88.mlp.experts.129.down_proj", "model.layers.88.mlp.experts.130.down_proj", "model.layers.88.mlp.experts.131.down_proj", "model.layers.88.mlp.experts.132.down_proj", "model.layers.88.mlp.experts.133.down_proj", "model.layers.88.mlp.experts.134.down_proj", "model.layers.88.mlp.experts.135.down_proj", "model.layers.88.mlp.experts.136.down_proj", "model.layers.88.mlp.experts.137.down_proj", "model.layers.88.mlp.experts.138.down_proj", "model.layers.88.mlp.experts.139.down_proj", "model.layers.88.mlp.experts.140.down_proj", "model.layers.88.mlp.experts.141.down_proj", "model.layers.88.mlp.experts.142.down_proj", "model.layers.88.mlp.experts.143.down_proj", "model.layers.88.mlp.experts.144.down_proj", "model.layers.88.mlp.experts.145.down_proj", "model.layers.88.mlp.experts.146.down_proj", "model.layers.88.mlp.experts.147.down_proj", "model.layers.88.mlp.experts.148.down_proj", "model.layers.88.mlp.experts.149.down_proj", "model.layers.88.mlp.experts.150.down_proj", "model.layers.88.mlp.experts.151.down_proj", "model.layers.88.mlp.experts.152.down_proj", "model.layers.88.mlp.experts.153.down_proj", "model.layers.88.mlp.experts.154.down_proj", "model.layers.88.mlp.experts.155.down_proj", "model.layers.88.mlp.experts.156.down_proj", "model.layers.88.mlp.experts.157.down_proj", "model.layers.88.mlp.experts.158.down_proj", "model.layers.88.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -8.756630122662423e-05, "dbits": 1258291200 }, { "dkld": -0.00010633077472449459, "dbits": 2516582400 }, { "dkld": -9.496910497547306e-05, "dbits": 3774873600 }, { "dkld": -0.00010943170636892874, "dbits": 6291456000 } ] }, { "idx": 531, "layers": [ "model.layers.89.self_attn.q_proj" ], "candidates": [ { "dkld": -0.0001702778972685337, "dbits": 62914560 }, { "dkld": -0.00022527426481247503, "dbits": 125829120 }, { "dkld": -0.0002809572033584118, "dbits": 188743680 }, { "dkld": -0.00026005906984211125, "dbits": 314572800 } ] }, { "idx": 532, "layers": [ "model.layers.89.self_attn.k_proj", "model.layers.89.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0005382791161537281, "dbits": 10485760 }, { "dkld": -0.0007004011422395706, "dbits": 20971520 }, { "dkld": -0.0004942458122968729, "dbits": 31457280 }, { "dkld": -0.0005662547424435671, "dbits": 52428800 } ] }, { "idx": 533, "layers": [ "model.layers.89.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00028682285919785777, "dbits": 62914560 }, { "dkld": -0.0002292322926223389, "dbits": 125829120 }, { "dkld": -0.0002803510986268548, "dbits": 188743680 }, { "dkld": -0.000265447329729801, "dbits": 314572800 } ] }, { "idx": 534, "layers": [ "model.layers.89.mlp.shared_experts.gate_proj", "model.layers.89.mlp.shared_experts.up_proj", "model.layers.89.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0001594781875610407, "dbits": 23592960 }, { "dkld": -0.0002971441484987819, "dbits": 47185920 }, { "dkld": -0.00034017302095890045, "dbits": 70778880 }, { "dkld": -0.0002931372262537507, "dbits": 117964800 } ] }, { "idx": 535, "layers": [ "model.layers.89.mlp.experts.0.gate_proj", "model.layers.89.mlp.experts.1.gate_proj", "model.layers.89.mlp.experts.2.gate_proj", "model.layers.89.mlp.experts.3.gate_proj", "model.layers.89.mlp.experts.4.gate_proj", "model.layers.89.mlp.experts.5.gate_proj", "model.layers.89.mlp.experts.6.gate_proj", "model.layers.89.mlp.experts.7.gate_proj", "model.layers.89.mlp.experts.8.gate_proj", "model.layers.89.mlp.experts.9.gate_proj", "model.layers.89.mlp.experts.10.gate_proj", "model.layers.89.mlp.experts.11.gate_proj", "model.layers.89.mlp.experts.12.gate_proj", "model.layers.89.mlp.experts.13.gate_proj", "model.layers.89.mlp.experts.14.gate_proj", "model.layers.89.mlp.experts.15.gate_proj", "model.layers.89.mlp.experts.16.gate_proj", "model.layers.89.mlp.experts.17.gate_proj", "model.layers.89.mlp.experts.18.gate_proj", "model.layers.89.mlp.experts.19.gate_proj", "model.layers.89.mlp.experts.20.gate_proj", "model.layers.89.mlp.experts.21.gate_proj", "model.layers.89.mlp.experts.22.gate_proj", "model.layers.89.mlp.experts.23.gate_proj", "model.layers.89.mlp.experts.24.gate_proj", "model.layers.89.mlp.experts.25.gate_proj", "model.layers.89.mlp.experts.26.gate_proj", "model.layers.89.mlp.experts.27.gate_proj", "model.layers.89.mlp.experts.28.gate_proj", "model.layers.89.mlp.experts.29.gate_proj", "model.layers.89.mlp.experts.30.gate_proj", "model.layers.89.mlp.experts.31.gate_proj", "model.layers.89.mlp.experts.32.gate_proj", "model.layers.89.mlp.experts.33.gate_proj", "model.layers.89.mlp.experts.34.gate_proj", "model.layers.89.mlp.experts.35.gate_proj", "model.layers.89.mlp.experts.36.gate_proj", "model.layers.89.mlp.experts.37.gate_proj", "model.layers.89.mlp.experts.38.gate_proj", "model.layers.89.mlp.experts.39.gate_proj", "model.layers.89.mlp.experts.40.gate_proj", "model.layers.89.mlp.experts.41.gate_proj", "model.layers.89.mlp.experts.42.gate_proj", "model.layers.89.mlp.experts.43.gate_proj", "model.layers.89.mlp.experts.44.gate_proj", "model.layers.89.mlp.experts.45.gate_proj", "model.layers.89.mlp.experts.46.gate_proj", "model.layers.89.mlp.experts.47.gate_proj", "model.layers.89.mlp.experts.48.gate_proj", "model.layers.89.mlp.experts.49.gate_proj", "model.layers.89.mlp.experts.50.gate_proj", "model.layers.89.mlp.experts.51.gate_proj", "model.layers.89.mlp.experts.52.gate_proj", "model.layers.89.mlp.experts.53.gate_proj", "model.layers.89.mlp.experts.54.gate_proj", "model.layers.89.mlp.experts.55.gate_proj", "model.layers.89.mlp.experts.56.gate_proj", "model.layers.89.mlp.experts.57.gate_proj", "model.layers.89.mlp.experts.58.gate_proj", "model.layers.89.mlp.experts.59.gate_proj", "model.layers.89.mlp.experts.60.gate_proj", "model.layers.89.mlp.experts.61.gate_proj", "model.layers.89.mlp.experts.62.gate_proj", "model.layers.89.mlp.experts.63.gate_proj", "model.layers.89.mlp.experts.64.gate_proj", "model.layers.89.mlp.experts.65.gate_proj", "model.layers.89.mlp.experts.66.gate_proj", "model.layers.89.mlp.experts.67.gate_proj", "model.layers.89.mlp.experts.68.gate_proj", "model.layers.89.mlp.experts.69.gate_proj", "model.layers.89.mlp.experts.70.gate_proj", "model.layers.89.mlp.experts.71.gate_proj", "model.layers.89.mlp.experts.72.gate_proj", "model.layers.89.mlp.experts.73.gate_proj", "model.layers.89.mlp.experts.74.gate_proj", "model.layers.89.mlp.experts.75.gate_proj", "model.layers.89.mlp.experts.76.gate_proj", "model.layers.89.mlp.experts.77.gate_proj", "model.layers.89.mlp.experts.78.gate_proj", "model.layers.89.mlp.experts.79.gate_proj", "model.layers.89.mlp.experts.80.gate_proj", "model.layers.89.mlp.experts.81.gate_proj", "model.layers.89.mlp.experts.82.gate_proj", "model.layers.89.mlp.experts.83.gate_proj", "model.layers.89.mlp.experts.84.gate_proj", "model.layers.89.mlp.experts.85.gate_proj", "model.layers.89.mlp.experts.86.gate_proj", "model.layers.89.mlp.experts.87.gate_proj", "model.layers.89.mlp.experts.88.gate_proj", "model.layers.89.mlp.experts.89.gate_proj", "model.layers.89.mlp.experts.90.gate_proj", "model.layers.89.mlp.experts.91.gate_proj", "model.layers.89.mlp.experts.92.gate_proj", "model.layers.89.mlp.experts.93.gate_proj", "model.layers.89.mlp.experts.94.gate_proj", "model.layers.89.mlp.experts.95.gate_proj", "model.layers.89.mlp.experts.96.gate_proj", "model.layers.89.mlp.experts.97.gate_proj", "model.layers.89.mlp.experts.98.gate_proj", "model.layers.89.mlp.experts.99.gate_proj", "model.layers.89.mlp.experts.100.gate_proj", "model.layers.89.mlp.experts.101.gate_proj", "model.layers.89.mlp.experts.102.gate_proj", "model.layers.89.mlp.experts.103.gate_proj", "model.layers.89.mlp.experts.104.gate_proj", "model.layers.89.mlp.experts.105.gate_proj", "model.layers.89.mlp.experts.106.gate_proj", "model.layers.89.mlp.experts.107.gate_proj", "model.layers.89.mlp.experts.108.gate_proj", "model.layers.89.mlp.experts.109.gate_proj", "model.layers.89.mlp.experts.110.gate_proj", "model.layers.89.mlp.experts.111.gate_proj", "model.layers.89.mlp.experts.112.gate_proj", "model.layers.89.mlp.experts.113.gate_proj", "model.layers.89.mlp.experts.114.gate_proj", "model.layers.89.mlp.experts.115.gate_proj", "model.layers.89.mlp.experts.116.gate_proj", "model.layers.89.mlp.experts.117.gate_proj", "model.layers.89.mlp.experts.118.gate_proj", "model.layers.89.mlp.experts.119.gate_proj", "model.layers.89.mlp.experts.120.gate_proj", "model.layers.89.mlp.experts.121.gate_proj", "model.layers.89.mlp.experts.122.gate_proj", "model.layers.89.mlp.experts.123.gate_proj", "model.layers.89.mlp.experts.124.gate_proj", "model.layers.89.mlp.experts.125.gate_proj", "model.layers.89.mlp.experts.126.gate_proj", "model.layers.89.mlp.experts.127.gate_proj", "model.layers.89.mlp.experts.128.gate_proj", "model.layers.89.mlp.experts.129.gate_proj", "model.layers.89.mlp.experts.130.gate_proj", "model.layers.89.mlp.experts.131.gate_proj", "model.layers.89.mlp.experts.132.gate_proj", "model.layers.89.mlp.experts.133.gate_proj", "model.layers.89.mlp.experts.134.gate_proj", "model.layers.89.mlp.experts.135.gate_proj", "model.layers.89.mlp.experts.136.gate_proj", "model.layers.89.mlp.experts.137.gate_proj", "model.layers.89.mlp.experts.138.gate_proj", "model.layers.89.mlp.experts.139.gate_proj", "model.layers.89.mlp.experts.140.gate_proj", "model.layers.89.mlp.experts.141.gate_proj", "model.layers.89.mlp.experts.142.gate_proj", "model.layers.89.mlp.experts.143.gate_proj", "model.layers.89.mlp.experts.144.gate_proj", "model.layers.89.mlp.experts.145.gate_proj", "model.layers.89.mlp.experts.146.gate_proj", "model.layers.89.mlp.experts.147.gate_proj", "model.layers.89.mlp.experts.148.gate_proj", "model.layers.89.mlp.experts.149.gate_proj", "model.layers.89.mlp.experts.150.gate_proj", "model.layers.89.mlp.experts.151.gate_proj", "model.layers.89.mlp.experts.152.gate_proj", "model.layers.89.mlp.experts.153.gate_proj", "model.layers.89.mlp.experts.154.gate_proj", "model.layers.89.mlp.experts.155.gate_proj", "model.layers.89.mlp.experts.156.gate_proj", "model.layers.89.mlp.experts.157.gate_proj", "model.layers.89.mlp.experts.158.gate_proj", "model.layers.89.mlp.experts.159.gate_proj", "model.layers.89.mlp.experts.0.up_proj", "model.layers.89.mlp.experts.1.up_proj", "model.layers.89.mlp.experts.2.up_proj", "model.layers.89.mlp.experts.3.up_proj", "model.layers.89.mlp.experts.4.up_proj", "model.layers.89.mlp.experts.5.up_proj", "model.layers.89.mlp.experts.6.up_proj", "model.layers.89.mlp.experts.7.up_proj", "model.layers.89.mlp.experts.8.up_proj", "model.layers.89.mlp.experts.9.up_proj", "model.layers.89.mlp.experts.10.up_proj", "model.layers.89.mlp.experts.11.up_proj", "model.layers.89.mlp.experts.12.up_proj", "model.layers.89.mlp.experts.13.up_proj", "model.layers.89.mlp.experts.14.up_proj", "model.layers.89.mlp.experts.15.up_proj", "model.layers.89.mlp.experts.16.up_proj", "model.layers.89.mlp.experts.17.up_proj", "model.layers.89.mlp.experts.18.up_proj", "model.layers.89.mlp.experts.19.up_proj", "model.layers.89.mlp.experts.20.up_proj", "model.layers.89.mlp.experts.21.up_proj", "model.layers.89.mlp.experts.22.up_proj", "model.layers.89.mlp.experts.23.up_proj", "model.layers.89.mlp.experts.24.up_proj", "model.layers.89.mlp.experts.25.up_proj", "model.layers.89.mlp.experts.26.up_proj", "model.layers.89.mlp.experts.27.up_proj", "model.layers.89.mlp.experts.28.up_proj", "model.layers.89.mlp.experts.29.up_proj", "model.layers.89.mlp.experts.30.up_proj", "model.layers.89.mlp.experts.31.up_proj", "model.layers.89.mlp.experts.32.up_proj", "model.layers.89.mlp.experts.33.up_proj", "model.layers.89.mlp.experts.34.up_proj", "model.layers.89.mlp.experts.35.up_proj", "model.layers.89.mlp.experts.36.up_proj", "model.layers.89.mlp.experts.37.up_proj", "model.layers.89.mlp.experts.38.up_proj", "model.layers.89.mlp.experts.39.up_proj", "model.layers.89.mlp.experts.40.up_proj", "model.layers.89.mlp.experts.41.up_proj", "model.layers.89.mlp.experts.42.up_proj", "model.layers.89.mlp.experts.43.up_proj", "model.layers.89.mlp.experts.44.up_proj", "model.layers.89.mlp.experts.45.up_proj", "model.layers.89.mlp.experts.46.up_proj", "model.layers.89.mlp.experts.47.up_proj", "model.layers.89.mlp.experts.48.up_proj", "model.layers.89.mlp.experts.49.up_proj", "model.layers.89.mlp.experts.50.up_proj", "model.layers.89.mlp.experts.51.up_proj", "model.layers.89.mlp.experts.52.up_proj", "model.layers.89.mlp.experts.53.up_proj", "model.layers.89.mlp.experts.54.up_proj", "model.layers.89.mlp.experts.55.up_proj", "model.layers.89.mlp.experts.56.up_proj", "model.layers.89.mlp.experts.57.up_proj", "model.layers.89.mlp.experts.58.up_proj", "model.layers.89.mlp.experts.59.up_proj", "model.layers.89.mlp.experts.60.up_proj", "model.layers.89.mlp.experts.61.up_proj", "model.layers.89.mlp.experts.62.up_proj", "model.layers.89.mlp.experts.63.up_proj", "model.layers.89.mlp.experts.64.up_proj", "model.layers.89.mlp.experts.65.up_proj", "model.layers.89.mlp.experts.66.up_proj", "model.layers.89.mlp.experts.67.up_proj", "model.layers.89.mlp.experts.68.up_proj", "model.layers.89.mlp.experts.69.up_proj", "model.layers.89.mlp.experts.70.up_proj", "model.layers.89.mlp.experts.71.up_proj", "model.layers.89.mlp.experts.72.up_proj", "model.layers.89.mlp.experts.73.up_proj", "model.layers.89.mlp.experts.74.up_proj", "model.layers.89.mlp.experts.75.up_proj", "model.layers.89.mlp.experts.76.up_proj", "model.layers.89.mlp.experts.77.up_proj", "model.layers.89.mlp.experts.78.up_proj", "model.layers.89.mlp.experts.79.up_proj", "model.layers.89.mlp.experts.80.up_proj", "model.layers.89.mlp.experts.81.up_proj", "model.layers.89.mlp.experts.82.up_proj", "model.layers.89.mlp.experts.83.up_proj", "model.layers.89.mlp.experts.84.up_proj", "model.layers.89.mlp.experts.85.up_proj", "model.layers.89.mlp.experts.86.up_proj", "model.layers.89.mlp.experts.87.up_proj", "model.layers.89.mlp.experts.88.up_proj", "model.layers.89.mlp.experts.89.up_proj", "model.layers.89.mlp.experts.90.up_proj", "model.layers.89.mlp.experts.91.up_proj", "model.layers.89.mlp.experts.92.up_proj", "model.layers.89.mlp.experts.93.up_proj", "model.layers.89.mlp.experts.94.up_proj", "model.layers.89.mlp.experts.95.up_proj", "model.layers.89.mlp.experts.96.up_proj", "model.layers.89.mlp.experts.97.up_proj", "model.layers.89.mlp.experts.98.up_proj", "model.layers.89.mlp.experts.99.up_proj", "model.layers.89.mlp.experts.100.up_proj", "model.layers.89.mlp.experts.101.up_proj", "model.layers.89.mlp.experts.102.up_proj", "model.layers.89.mlp.experts.103.up_proj", "model.layers.89.mlp.experts.104.up_proj", "model.layers.89.mlp.experts.105.up_proj", "model.layers.89.mlp.experts.106.up_proj", "model.layers.89.mlp.experts.107.up_proj", "model.layers.89.mlp.experts.108.up_proj", "model.layers.89.mlp.experts.109.up_proj", "model.layers.89.mlp.experts.110.up_proj", "model.layers.89.mlp.experts.111.up_proj", "model.layers.89.mlp.experts.112.up_proj", "model.layers.89.mlp.experts.113.up_proj", "model.layers.89.mlp.experts.114.up_proj", "model.layers.89.mlp.experts.115.up_proj", "model.layers.89.mlp.experts.116.up_proj", "model.layers.89.mlp.experts.117.up_proj", "model.layers.89.mlp.experts.118.up_proj", "model.layers.89.mlp.experts.119.up_proj", "model.layers.89.mlp.experts.120.up_proj", "model.layers.89.mlp.experts.121.up_proj", "model.layers.89.mlp.experts.122.up_proj", "model.layers.89.mlp.experts.123.up_proj", "model.layers.89.mlp.experts.124.up_proj", "model.layers.89.mlp.experts.125.up_proj", "model.layers.89.mlp.experts.126.up_proj", "model.layers.89.mlp.experts.127.up_proj", "model.layers.89.mlp.experts.128.up_proj", "model.layers.89.mlp.experts.129.up_proj", "model.layers.89.mlp.experts.130.up_proj", "model.layers.89.mlp.experts.131.up_proj", "model.layers.89.mlp.experts.132.up_proj", "model.layers.89.mlp.experts.133.up_proj", "model.layers.89.mlp.experts.134.up_proj", "model.layers.89.mlp.experts.135.up_proj", "model.layers.89.mlp.experts.136.up_proj", "model.layers.89.mlp.experts.137.up_proj", "model.layers.89.mlp.experts.138.up_proj", "model.layers.89.mlp.experts.139.up_proj", "model.layers.89.mlp.experts.140.up_proj", "model.layers.89.mlp.experts.141.up_proj", "model.layers.89.mlp.experts.142.up_proj", "model.layers.89.mlp.experts.143.up_proj", "model.layers.89.mlp.experts.144.up_proj", "model.layers.89.mlp.experts.145.up_proj", "model.layers.89.mlp.experts.146.up_proj", "model.layers.89.mlp.experts.147.up_proj", "model.layers.89.mlp.experts.148.up_proj", "model.layers.89.mlp.experts.149.up_proj", "model.layers.89.mlp.experts.150.up_proj", "model.layers.89.mlp.experts.151.up_proj", "model.layers.89.mlp.experts.152.up_proj", "model.layers.89.mlp.experts.153.up_proj", "model.layers.89.mlp.experts.154.up_proj", "model.layers.89.mlp.experts.155.up_proj", "model.layers.89.mlp.experts.156.up_proj", "model.layers.89.mlp.experts.157.up_proj", "model.layers.89.mlp.experts.158.up_proj", "model.layers.89.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00011942125856877206, "dbits": 2516582400 }, { "dkld": -8.455412462354817e-05, "dbits": 5033164800 }, { "dkld": -8.544474840165017e-05, "dbits": 7549747200 }, { "dkld": -8.792970329524197e-05, "dbits": 12582912000 } ] }, { "idx": 536, "layers": [ "model.layers.89.mlp.experts.0.down_proj", "model.layers.89.mlp.experts.1.down_proj", "model.layers.89.mlp.experts.2.down_proj", "model.layers.89.mlp.experts.3.down_proj", "model.layers.89.mlp.experts.4.down_proj", "model.layers.89.mlp.experts.5.down_proj", "model.layers.89.mlp.experts.6.down_proj", "model.layers.89.mlp.experts.7.down_proj", "model.layers.89.mlp.experts.8.down_proj", "model.layers.89.mlp.experts.9.down_proj", "model.layers.89.mlp.experts.10.down_proj", "model.layers.89.mlp.experts.11.down_proj", "model.layers.89.mlp.experts.12.down_proj", "model.layers.89.mlp.experts.13.down_proj", "model.layers.89.mlp.experts.14.down_proj", "model.layers.89.mlp.experts.15.down_proj", "model.layers.89.mlp.experts.16.down_proj", "model.layers.89.mlp.experts.17.down_proj", "model.layers.89.mlp.experts.18.down_proj", "model.layers.89.mlp.experts.19.down_proj", "model.layers.89.mlp.experts.20.down_proj", "model.layers.89.mlp.experts.21.down_proj", "model.layers.89.mlp.experts.22.down_proj", "model.layers.89.mlp.experts.23.down_proj", "model.layers.89.mlp.experts.24.down_proj", "model.layers.89.mlp.experts.25.down_proj", "model.layers.89.mlp.experts.26.down_proj", "model.layers.89.mlp.experts.27.down_proj", "model.layers.89.mlp.experts.28.down_proj", "model.layers.89.mlp.experts.29.down_proj", "model.layers.89.mlp.experts.30.down_proj", "model.layers.89.mlp.experts.31.down_proj", "model.layers.89.mlp.experts.32.down_proj", "model.layers.89.mlp.experts.33.down_proj", "model.layers.89.mlp.experts.34.down_proj", "model.layers.89.mlp.experts.35.down_proj", "model.layers.89.mlp.experts.36.down_proj", "model.layers.89.mlp.experts.37.down_proj", "model.layers.89.mlp.experts.38.down_proj", "model.layers.89.mlp.experts.39.down_proj", "model.layers.89.mlp.experts.40.down_proj", "model.layers.89.mlp.experts.41.down_proj", "model.layers.89.mlp.experts.42.down_proj", "model.layers.89.mlp.experts.43.down_proj", "model.layers.89.mlp.experts.44.down_proj", "model.layers.89.mlp.experts.45.down_proj", "model.layers.89.mlp.experts.46.down_proj", "model.layers.89.mlp.experts.47.down_proj", "model.layers.89.mlp.experts.48.down_proj", "model.layers.89.mlp.experts.49.down_proj", "model.layers.89.mlp.experts.50.down_proj", "model.layers.89.mlp.experts.51.down_proj", "model.layers.89.mlp.experts.52.down_proj", "model.layers.89.mlp.experts.53.down_proj", "model.layers.89.mlp.experts.54.down_proj", "model.layers.89.mlp.experts.55.down_proj", "model.layers.89.mlp.experts.56.down_proj", "model.layers.89.mlp.experts.57.down_proj", "model.layers.89.mlp.experts.58.down_proj", "model.layers.89.mlp.experts.59.down_proj", "model.layers.89.mlp.experts.60.down_proj", "model.layers.89.mlp.experts.61.down_proj", "model.layers.89.mlp.experts.62.down_proj", "model.layers.89.mlp.experts.63.down_proj", "model.layers.89.mlp.experts.64.down_proj", "model.layers.89.mlp.experts.65.down_proj", "model.layers.89.mlp.experts.66.down_proj", "model.layers.89.mlp.experts.67.down_proj", "model.layers.89.mlp.experts.68.down_proj", "model.layers.89.mlp.experts.69.down_proj", "model.layers.89.mlp.experts.70.down_proj", "model.layers.89.mlp.experts.71.down_proj", "model.layers.89.mlp.experts.72.down_proj", "model.layers.89.mlp.experts.73.down_proj", "model.layers.89.mlp.experts.74.down_proj", "model.layers.89.mlp.experts.75.down_proj", "model.layers.89.mlp.experts.76.down_proj", "model.layers.89.mlp.experts.77.down_proj", "model.layers.89.mlp.experts.78.down_proj", "model.layers.89.mlp.experts.79.down_proj", "model.layers.89.mlp.experts.80.down_proj", "model.layers.89.mlp.experts.81.down_proj", "model.layers.89.mlp.experts.82.down_proj", "model.layers.89.mlp.experts.83.down_proj", "model.layers.89.mlp.experts.84.down_proj", "model.layers.89.mlp.experts.85.down_proj", "model.layers.89.mlp.experts.86.down_proj", "model.layers.89.mlp.experts.87.down_proj", "model.layers.89.mlp.experts.88.down_proj", "model.layers.89.mlp.experts.89.down_proj", "model.layers.89.mlp.experts.90.down_proj", "model.layers.89.mlp.experts.91.down_proj", "model.layers.89.mlp.experts.92.down_proj", "model.layers.89.mlp.experts.93.down_proj", "model.layers.89.mlp.experts.94.down_proj", "model.layers.89.mlp.experts.95.down_proj", "model.layers.89.mlp.experts.96.down_proj", "model.layers.89.mlp.experts.97.down_proj", "model.layers.89.mlp.experts.98.down_proj", "model.layers.89.mlp.experts.99.down_proj", "model.layers.89.mlp.experts.100.down_proj", "model.layers.89.mlp.experts.101.down_proj", "model.layers.89.mlp.experts.102.down_proj", "model.layers.89.mlp.experts.103.down_proj", "model.layers.89.mlp.experts.104.down_proj", "model.layers.89.mlp.experts.105.down_proj", "model.layers.89.mlp.experts.106.down_proj", "model.layers.89.mlp.experts.107.down_proj", "model.layers.89.mlp.experts.108.down_proj", "model.layers.89.mlp.experts.109.down_proj", "model.layers.89.mlp.experts.110.down_proj", "model.layers.89.mlp.experts.111.down_proj", "model.layers.89.mlp.experts.112.down_proj", "model.layers.89.mlp.experts.113.down_proj", "model.layers.89.mlp.experts.114.down_proj", "model.layers.89.mlp.experts.115.down_proj", "model.layers.89.mlp.experts.116.down_proj", "model.layers.89.mlp.experts.117.down_proj", "model.layers.89.mlp.experts.118.down_proj", "model.layers.89.mlp.experts.119.down_proj", "model.layers.89.mlp.experts.120.down_proj", "model.layers.89.mlp.experts.121.down_proj", "model.layers.89.mlp.experts.122.down_proj", "model.layers.89.mlp.experts.123.down_proj", "model.layers.89.mlp.experts.124.down_proj", "model.layers.89.mlp.experts.125.down_proj", "model.layers.89.mlp.experts.126.down_proj", "model.layers.89.mlp.experts.127.down_proj", "model.layers.89.mlp.experts.128.down_proj", "model.layers.89.mlp.experts.129.down_proj", "model.layers.89.mlp.experts.130.down_proj", "model.layers.89.mlp.experts.131.down_proj", "model.layers.89.mlp.experts.132.down_proj", "model.layers.89.mlp.experts.133.down_proj", "model.layers.89.mlp.experts.134.down_proj", "model.layers.89.mlp.experts.135.down_proj", "model.layers.89.mlp.experts.136.down_proj", "model.layers.89.mlp.experts.137.down_proj", "model.layers.89.mlp.experts.138.down_proj", "model.layers.89.mlp.experts.139.down_proj", "model.layers.89.mlp.experts.140.down_proj", "model.layers.89.mlp.experts.141.down_proj", "model.layers.89.mlp.experts.142.down_proj", "model.layers.89.mlp.experts.143.down_proj", "model.layers.89.mlp.experts.144.down_proj", "model.layers.89.mlp.experts.145.down_proj", "model.layers.89.mlp.experts.146.down_proj", "model.layers.89.mlp.experts.147.down_proj", "model.layers.89.mlp.experts.148.down_proj", "model.layers.89.mlp.experts.149.down_proj", "model.layers.89.mlp.experts.150.down_proj", "model.layers.89.mlp.experts.151.down_proj", "model.layers.89.mlp.experts.152.down_proj", "model.layers.89.mlp.experts.153.down_proj", "model.layers.89.mlp.experts.154.down_proj", "model.layers.89.mlp.experts.155.down_proj", "model.layers.89.mlp.experts.156.down_proj", "model.layers.89.mlp.experts.157.down_proj", "model.layers.89.mlp.experts.158.down_proj", "model.layers.89.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -5.3233373910194226e-05, "dbits": 1258291200 }, { "dkld": -5.221348255873837e-05, "dbits": 2516582400 }, { "dkld": -5.635889247060377e-05, "dbits": 3774873600 }, { "dkld": -6.254604086280979e-05, "dbits": 6291456000 } ] }, { "idx": 537, "layers": [ "model.layers.90.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00018974440172314644, "dbits": 62914560 }, { "dkld": -0.0002443412318825777, "dbits": 125829120 }, { "dkld": -0.0002305973321199417, "dbits": 188743680 }, { "dkld": -0.0002232681959867533, "dbits": 314572800 } ] }, { "idx": 538, "layers": [ "model.layers.90.self_attn.k_proj", "model.layers.90.self_attn.v_proj" ], "candidates": [ { "dkld": -0.00017077848315238953, "dbits": 10485760 }, { "dkld": -0.00028046807274223484, "dbits": 20971520 }, { "dkld": -0.00025082286447286606, "dbits": 31457280 }, { "dkld": -0.0002247016876936042, "dbits": 52428800 } ] }, { "idx": 539, "layers": [ "model.layers.90.self_attn.o_proj" ], "candidates": [ { "dkld": -8.75694677233696e-05, "dbits": 62914560 }, { "dkld": -8.934363722801208e-05, "dbits": 125829120 }, { "dkld": -0.00010915761813522218, "dbits": 188743680 }, { "dkld": -0.00014081057161093036, "dbits": 314572800 } ] }, { "idx": 540, "layers": [ "model.layers.90.mlp.shared_experts.gate_proj", "model.layers.90.mlp.shared_experts.up_proj", "model.layers.90.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": -0.0006203852593898856, "dbits": 23592960 }, { "dkld": -0.0006362467072904138, "dbits": 47185920 }, { "dkld": -0.0005469106137752644, "dbits": 70778880 }, { "dkld": -0.0006009818986058291, "dbits": 117964800 } ] }, { "idx": 541, "layers": [ "model.layers.90.mlp.experts.0.gate_proj", "model.layers.90.mlp.experts.1.gate_proj", "model.layers.90.mlp.experts.2.gate_proj", "model.layers.90.mlp.experts.3.gate_proj", "model.layers.90.mlp.experts.4.gate_proj", "model.layers.90.mlp.experts.5.gate_proj", "model.layers.90.mlp.experts.6.gate_proj", "model.layers.90.mlp.experts.7.gate_proj", "model.layers.90.mlp.experts.8.gate_proj", "model.layers.90.mlp.experts.9.gate_proj", "model.layers.90.mlp.experts.10.gate_proj", "model.layers.90.mlp.experts.11.gate_proj", "model.layers.90.mlp.experts.12.gate_proj", "model.layers.90.mlp.experts.13.gate_proj", "model.layers.90.mlp.experts.14.gate_proj", "model.layers.90.mlp.experts.15.gate_proj", "model.layers.90.mlp.experts.16.gate_proj", "model.layers.90.mlp.experts.17.gate_proj", "model.layers.90.mlp.experts.18.gate_proj", "model.layers.90.mlp.experts.19.gate_proj", "model.layers.90.mlp.experts.20.gate_proj", "model.layers.90.mlp.experts.21.gate_proj", "model.layers.90.mlp.experts.22.gate_proj", "model.layers.90.mlp.experts.23.gate_proj", "model.layers.90.mlp.experts.24.gate_proj", "model.layers.90.mlp.experts.25.gate_proj", "model.layers.90.mlp.experts.26.gate_proj", "model.layers.90.mlp.experts.27.gate_proj", "model.layers.90.mlp.experts.28.gate_proj", "model.layers.90.mlp.experts.29.gate_proj", "model.layers.90.mlp.experts.30.gate_proj", "model.layers.90.mlp.experts.31.gate_proj", "model.layers.90.mlp.experts.32.gate_proj", "model.layers.90.mlp.experts.33.gate_proj", "model.layers.90.mlp.experts.34.gate_proj", "model.layers.90.mlp.experts.35.gate_proj", "model.layers.90.mlp.experts.36.gate_proj", "model.layers.90.mlp.experts.37.gate_proj", "model.layers.90.mlp.experts.38.gate_proj", "model.layers.90.mlp.experts.39.gate_proj", "model.layers.90.mlp.experts.40.gate_proj", "model.layers.90.mlp.experts.41.gate_proj", "model.layers.90.mlp.experts.42.gate_proj", "model.layers.90.mlp.experts.43.gate_proj", "model.layers.90.mlp.experts.44.gate_proj", "model.layers.90.mlp.experts.45.gate_proj", "model.layers.90.mlp.experts.46.gate_proj", "model.layers.90.mlp.experts.47.gate_proj", "model.layers.90.mlp.experts.48.gate_proj", "model.layers.90.mlp.experts.49.gate_proj", "model.layers.90.mlp.experts.50.gate_proj", "model.layers.90.mlp.experts.51.gate_proj", "model.layers.90.mlp.experts.52.gate_proj", "model.layers.90.mlp.experts.53.gate_proj", "model.layers.90.mlp.experts.54.gate_proj", "model.layers.90.mlp.experts.55.gate_proj", "model.layers.90.mlp.experts.56.gate_proj", "model.layers.90.mlp.experts.57.gate_proj", "model.layers.90.mlp.experts.58.gate_proj", "model.layers.90.mlp.experts.59.gate_proj", "model.layers.90.mlp.experts.60.gate_proj", "model.layers.90.mlp.experts.61.gate_proj", "model.layers.90.mlp.experts.62.gate_proj", "model.layers.90.mlp.experts.63.gate_proj", "model.layers.90.mlp.experts.64.gate_proj", "model.layers.90.mlp.experts.65.gate_proj", "model.layers.90.mlp.experts.66.gate_proj", "model.layers.90.mlp.experts.67.gate_proj", "model.layers.90.mlp.experts.68.gate_proj", "model.layers.90.mlp.experts.69.gate_proj", "model.layers.90.mlp.experts.70.gate_proj", "model.layers.90.mlp.experts.71.gate_proj", "model.layers.90.mlp.experts.72.gate_proj", "model.layers.90.mlp.experts.73.gate_proj", "model.layers.90.mlp.experts.74.gate_proj", "model.layers.90.mlp.experts.75.gate_proj", "model.layers.90.mlp.experts.76.gate_proj", "model.layers.90.mlp.experts.77.gate_proj", "model.layers.90.mlp.experts.78.gate_proj", "model.layers.90.mlp.experts.79.gate_proj", "model.layers.90.mlp.experts.80.gate_proj", "model.layers.90.mlp.experts.81.gate_proj", "model.layers.90.mlp.experts.82.gate_proj", "model.layers.90.mlp.experts.83.gate_proj", "model.layers.90.mlp.experts.84.gate_proj", "model.layers.90.mlp.experts.85.gate_proj", "model.layers.90.mlp.experts.86.gate_proj", "model.layers.90.mlp.experts.87.gate_proj", "model.layers.90.mlp.experts.88.gate_proj", "model.layers.90.mlp.experts.89.gate_proj", "model.layers.90.mlp.experts.90.gate_proj", "model.layers.90.mlp.experts.91.gate_proj", "model.layers.90.mlp.experts.92.gate_proj", "model.layers.90.mlp.experts.93.gate_proj", "model.layers.90.mlp.experts.94.gate_proj", "model.layers.90.mlp.experts.95.gate_proj", "model.layers.90.mlp.experts.96.gate_proj", "model.layers.90.mlp.experts.97.gate_proj", "model.layers.90.mlp.experts.98.gate_proj", "model.layers.90.mlp.experts.99.gate_proj", "model.layers.90.mlp.experts.100.gate_proj", "model.layers.90.mlp.experts.101.gate_proj", "model.layers.90.mlp.experts.102.gate_proj", "model.layers.90.mlp.experts.103.gate_proj", "model.layers.90.mlp.experts.104.gate_proj", "model.layers.90.mlp.experts.105.gate_proj", "model.layers.90.mlp.experts.106.gate_proj", "model.layers.90.mlp.experts.107.gate_proj", "model.layers.90.mlp.experts.108.gate_proj", "model.layers.90.mlp.experts.109.gate_proj", "model.layers.90.mlp.experts.110.gate_proj", "model.layers.90.mlp.experts.111.gate_proj", "model.layers.90.mlp.experts.112.gate_proj", "model.layers.90.mlp.experts.113.gate_proj", "model.layers.90.mlp.experts.114.gate_proj", "model.layers.90.mlp.experts.115.gate_proj", "model.layers.90.mlp.experts.116.gate_proj", "model.layers.90.mlp.experts.117.gate_proj", "model.layers.90.mlp.experts.118.gate_proj", "model.layers.90.mlp.experts.119.gate_proj", "model.layers.90.mlp.experts.120.gate_proj", "model.layers.90.mlp.experts.121.gate_proj", "model.layers.90.mlp.experts.122.gate_proj", "model.layers.90.mlp.experts.123.gate_proj", "model.layers.90.mlp.experts.124.gate_proj", "model.layers.90.mlp.experts.125.gate_proj", "model.layers.90.mlp.experts.126.gate_proj", "model.layers.90.mlp.experts.127.gate_proj", "model.layers.90.mlp.experts.128.gate_proj", "model.layers.90.mlp.experts.129.gate_proj", "model.layers.90.mlp.experts.130.gate_proj", "model.layers.90.mlp.experts.131.gate_proj", "model.layers.90.mlp.experts.132.gate_proj", "model.layers.90.mlp.experts.133.gate_proj", "model.layers.90.mlp.experts.134.gate_proj", "model.layers.90.mlp.experts.135.gate_proj", "model.layers.90.mlp.experts.136.gate_proj", "model.layers.90.mlp.experts.137.gate_proj", "model.layers.90.mlp.experts.138.gate_proj", "model.layers.90.mlp.experts.139.gate_proj", "model.layers.90.mlp.experts.140.gate_proj", "model.layers.90.mlp.experts.141.gate_proj", "model.layers.90.mlp.experts.142.gate_proj", "model.layers.90.mlp.experts.143.gate_proj", "model.layers.90.mlp.experts.144.gate_proj", "model.layers.90.mlp.experts.145.gate_proj", "model.layers.90.mlp.experts.146.gate_proj", "model.layers.90.mlp.experts.147.gate_proj", "model.layers.90.mlp.experts.148.gate_proj", "model.layers.90.mlp.experts.149.gate_proj", "model.layers.90.mlp.experts.150.gate_proj", "model.layers.90.mlp.experts.151.gate_proj", "model.layers.90.mlp.experts.152.gate_proj", "model.layers.90.mlp.experts.153.gate_proj", "model.layers.90.mlp.experts.154.gate_proj", "model.layers.90.mlp.experts.155.gate_proj", "model.layers.90.mlp.experts.156.gate_proj", "model.layers.90.mlp.experts.157.gate_proj", "model.layers.90.mlp.experts.158.gate_proj", "model.layers.90.mlp.experts.159.gate_proj", "model.layers.90.mlp.experts.0.up_proj", "model.layers.90.mlp.experts.1.up_proj", "model.layers.90.mlp.experts.2.up_proj", "model.layers.90.mlp.experts.3.up_proj", "model.layers.90.mlp.experts.4.up_proj", "model.layers.90.mlp.experts.5.up_proj", "model.layers.90.mlp.experts.6.up_proj", "model.layers.90.mlp.experts.7.up_proj", "model.layers.90.mlp.experts.8.up_proj", "model.layers.90.mlp.experts.9.up_proj", "model.layers.90.mlp.experts.10.up_proj", "model.layers.90.mlp.experts.11.up_proj", "model.layers.90.mlp.experts.12.up_proj", "model.layers.90.mlp.experts.13.up_proj", "model.layers.90.mlp.experts.14.up_proj", "model.layers.90.mlp.experts.15.up_proj", "model.layers.90.mlp.experts.16.up_proj", "model.layers.90.mlp.experts.17.up_proj", "model.layers.90.mlp.experts.18.up_proj", "model.layers.90.mlp.experts.19.up_proj", "model.layers.90.mlp.experts.20.up_proj", "model.layers.90.mlp.experts.21.up_proj", "model.layers.90.mlp.experts.22.up_proj", "model.layers.90.mlp.experts.23.up_proj", "model.layers.90.mlp.experts.24.up_proj", "model.layers.90.mlp.experts.25.up_proj", "model.layers.90.mlp.experts.26.up_proj", "model.layers.90.mlp.experts.27.up_proj", "model.layers.90.mlp.experts.28.up_proj", "model.layers.90.mlp.experts.29.up_proj", "model.layers.90.mlp.experts.30.up_proj", "model.layers.90.mlp.experts.31.up_proj", "model.layers.90.mlp.experts.32.up_proj", "model.layers.90.mlp.experts.33.up_proj", "model.layers.90.mlp.experts.34.up_proj", "model.layers.90.mlp.experts.35.up_proj", "model.layers.90.mlp.experts.36.up_proj", "model.layers.90.mlp.experts.37.up_proj", "model.layers.90.mlp.experts.38.up_proj", "model.layers.90.mlp.experts.39.up_proj", "model.layers.90.mlp.experts.40.up_proj", "model.layers.90.mlp.experts.41.up_proj", "model.layers.90.mlp.experts.42.up_proj", "model.layers.90.mlp.experts.43.up_proj", "model.layers.90.mlp.experts.44.up_proj", "model.layers.90.mlp.experts.45.up_proj", "model.layers.90.mlp.experts.46.up_proj", "model.layers.90.mlp.experts.47.up_proj", "model.layers.90.mlp.experts.48.up_proj", "model.layers.90.mlp.experts.49.up_proj", "model.layers.90.mlp.experts.50.up_proj", "model.layers.90.mlp.experts.51.up_proj", "model.layers.90.mlp.experts.52.up_proj", "model.layers.90.mlp.experts.53.up_proj", "model.layers.90.mlp.experts.54.up_proj", "model.layers.90.mlp.experts.55.up_proj", "model.layers.90.mlp.experts.56.up_proj", "model.layers.90.mlp.experts.57.up_proj", "model.layers.90.mlp.experts.58.up_proj", "model.layers.90.mlp.experts.59.up_proj", "model.layers.90.mlp.experts.60.up_proj", "model.layers.90.mlp.experts.61.up_proj", "model.layers.90.mlp.experts.62.up_proj", "model.layers.90.mlp.experts.63.up_proj", "model.layers.90.mlp.experts.64.up_proj", "model.layers.90.mlp.experts.65.up_proj", "model.layers.90.mlp.experts.66.up_proj", "model.layers.90.mlp.experts.67.up_proj", "model.layers.90.mlp.experts.68.up_proj", "model.layers.90.mlp.experts.69.up_proj", "model.layers.90.mlp.experts.70.up_proj", "model.layers.90.mlp.experts.71.up_proj", "model.layers.90.mlp.experts.72.up_proj", "model.layers.90.mlp.experts.73.up_proj", "model.layers.90.mlp.experts.74.up_proj", "model.layers.90.mlp.experts.75.up_proj", "model.layers.90.mlp.experts.76.up_proj", "model.layers.90.mlp.experts.77.up_proj", "model.layers.90.mlp.experts.78.up_proj", "model.layers.90.mlp.experts.79.up_proj", "model.layers.90.mlp.experts.80.up_proj", "model.layers.90.mlp.experts.81.up_proj", "model.layers.90.mlp.experts.82.up_proj", "model.layers.90.mlp.experts.83.up_proj", "model.layers.90.mlp.experts.84.up_proj", "model.layers.90.mlp.experts.85.up_proj", "model.layers.90.mlp.experts.86.up_proj", "model.layers.90.mlp.experts.87.up_proj", "model.layers.90.mlp.experts.88.up_proj", "model.layers.90.mlp.experts.89.up_proj", "model.layers.90.mlp.experts.90.up_proj", "model.layers.90.mlp.experts.91.up_proj", "model.layers.90.mlp.experts.92.up_proj", "model.layers.90.mlp.experts.93.up_proj", "model.layers.90.mlp.experts.94.up_proj", "model.layers.90.mlp.experts.95.up_proj", "model.layers.90.mlp.experts.96.up_proj", "model.layers.90.mlp.experts.97.up_proj", "model.layers.90.mlp.experts.98.up_proj", "model.layers.90.mlp.experts.99.up_proj", "model.layers.90.mlp.experts.100.up_proj", "model.layers.90.mlp.experts.101.up_proj", "model.layers.90.mlp.experts.102.up_proj", "model.layers.90.mlp.experts.103.up_proj", "model.layers.90.mlp.experts.104.up_proj", "model.layers.90.mlp.experts.105.up_proj", "model.layers.90.mlp.experts.106.up_proj", "model.layers.90.mlp.experts.107.up_proj", "model.layers.90.mlp.experts.108.up_proj", "model.layers.90.mlp.experts.109.up_proj", "model.layers.90.mlp.experts.110.up_proj", "model.layers.90.mlp.experts.111.up_proj", "model.layers.90.mlp.experts.112.up_proj", "model.layers.90.mlp.experts.113.up_proj", "model.layers.90.mlp.experts.114.up_proj", "model.layers.90.mlp.experts.115.up_proj", "model.layers.90.mlp.experts.116.up_proj", "model.layers.90.mlp.experts.117.up_proj", "model.layers.90.mlp.experts.118.up_proj", "model.layers.90.mlp.experts.119.up_proj", "model.layers.90.mlp.experts.120.up_proj", "model.layers.90.mlp.experts.121.up_proj", "model.layers.90.mlp.experts.122.up_proj", "model.layers.90.mlp.experts.123.up_proj", "model.layers.90.mlp.experts.124.up_proj", "model.layers.90.mlp.experts.125.up_proj", "model.layers.90.mlp.experts.126.up_proj", "model.layers.90.mlp.experts.127.up_proj", "model.layers.90.mlp.experts.128.up_proj", "model.layers.90.mlp.experts.129.up_proj", "model.layers.90.mlp.experts.130.up_proj", "model.layers.90.mlp.experts.131.up_proj", "model.layers.90.mlp.experts.132.up_proj", "model.layers.90.mlp.experts.133.up_proj", "model.layers.90.mlp.experts.134.up_proj", "model.layers.90.mlp.experts.135.up_proj", "model.layers.90.mlp.experts.136.up_proj", "model.layers.90.mlp.experts.137.up_proj", "model.layers.90.mlp.experts.138.up_proj", "model.layers.90.mlp.experts.139.up_proj", "model.layers.90.mlp.experts.140.up_proj", "model.layers.90.mlp.experts.141.up_proj", "model.layers.90.mlp.experts.142.up_proj", "model.layers.90.mlp.experts.143.up_proj", "model.layers.90.mlp.experts.144.up_proj", "model.layers.90.mlp.experts.145.up_proj", "model.layers.90.mlp.experts.146.up_proj", "model.layers.90.mlp.experts.147.up_proj", "model.layers.90.mlp.experts.148.up_proj", "model.layers.90.mlp.experts.149.up_proj", "model.layers.90.mlp.experts.150.up_proj", "model.layers.90.mlp.experts.151.up_proj", "model.layers.90.mlp.experts.152.up_proj", "model.layers.90.mlp.experts.153.up_proj", "model.layers.90.mlp.experts.154.up_proj", "model.layers.90.mlp.experts.155.up_proj", "model.layers.90.mlp.experts.156.up_proj", "model.layers.90.mlp.experts.157.up_proj", "model.layers.90.mlp.experts.158.up_proj", "model.layers.90.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -0.00011577596887946129, "dbits": 2516582400 }, { "dkld": -6.834920495749075e-05, "dbits": 5033164800 }, { "dkld": -9.785685688257217e-05, "dbits": 7549747200 }, { "dkld": -9.820088744164068e-05, "dbits": 12582912000 } ] }, { "idx": 542, "layers": [ "model.layers.90.mlp.experts.0.down_proj", "model.layers.90.mlp.experts.1.down_proj", "model.layers.90.mlp.experts.2.down_proj", "model.layers.90.mlp.experts.3.down_proj", "model.layers.90.mlp.experts.4.down_proj", "model.layers.90.mlp.experts.5.down_proj", "model.layers.90.mlp.experts.6.down_proj", "model.layers.90.mlp.experts.7.down_proj", "model.layers.90.mlp.experts.8.down_proj", "model.layers.90.mlp.experts.9.down_proj", "model.layers.90.mlp.experts.10.down_proj", "model.layers.90.mlp.experts.11.down_proj", "model.layers.90.mlp.experts.12.down_proj", "model.layers.90.mlp.experts.13.down_proj", "model.layers.90.mlp.experts.14.down_proj", "model.layers.90.mlp.experts.15.down_proj", "model.layers.90.mlp.experts.16.down_proj", "model.layers.90.mlp.experts.17.down_proj", "model.layers.90.mlp.experts.18.down_proj", "model.layers.90.mlp.experts.19.down_proj", "model.layers.90.mlp.experts.20.down_proj", "model.layers.90.mlp.experts.21.down_proj", "model.layers.90.mlp.experts.22.down_proj", "model.layers.90.mlp.experts.23.down_proj", "model.layers.90.mlp.experts.24.down_proj", "model.layers.90.mlp.experts.25.down_proj", "model.layers.90.mlp.experts.26.down_proj", "model.layers.90.mlp.experts.27.down_proj", "model.layers.90.mlp.experts.28.down_proj", "model.layers.90.mlp.experts.29.down_proj", "model.layers.90.mlp.experts.30.down_proj", "model.layers.90.mlp.experts.31.down_proj", "model.layers.90.mlp.experts.32.down_proj", "model.layers.90.mlp.experts.33.down_proj", "model.layers.90.mlp.experts.34.down_proj", "model.layers.90.mlp.experts.35.down_proj", "model.layers.90.mlp.experts.36.down_proj", "model.layers.90.mlp.experts.37.down_proj", "model.layers.90.mlp.experts.38.down_proj", "model.layers.90.mlp.experts.39.down_proj", "model.layers.90.mlp.experts.40.down_proj", "model.layers.90.mlp.experts.41.down_proj", "model.layers.90.mlp.experts.42.down_proj", "model.layers.90.mlp.experts.43.down_proj", "model.layers.90.mlp.experts.44.down_proj", "model.layers.90.mlp.experts.45.down_proj", "model.layers.90.mlp.experts.46.down_proj", "model.layers.90.mlp.experts.47.down_proj", "model.layers.90.mlp.experts.48.down_proj", "model.layers.90.mlp.experts.49.down_proj", "model.layers.90.mlp.experts.50.down_proj", "model.layers.90.mlp.experts.51.down_proj", "model.layers.90.mlp.experts.52.down_proj", "model.layers.90.mlp.experts.53.down_proj", "model.layers.90.mlp.experts.54.down_proj", "model.layers.90.mlp.experts.55.down_proj", "model.layers.90.mlp.experts.56.down_proj", "model.layers.90.mlp.experts.57.down_proj", "model.layers.90.mlp.experts.58.down_proj", "model.layers.90.mlp.experts.59.down_proj", "model.layers.90.mlp.experts.60.down_proj", "model.layers.90.mlp.experts.61.down_proj", "model.layers.90.mlp.experts.62.down_proj", "model.layers.90.mlp.experts.63.down_proj", "model.layers.90.mlp.experts.64.down_proj", "model.layers.90.mlp.experts.65.down_proj", "model.layers.90.mlp.experts.66.down_proj", "model.layers.90.mlp.experts.67.down_proj", "model.layers.90.mlp.experts.68.down_proj", "model.layers.90.mlp.experts.69.down_proj", "model.layers.90.mlp.experts.70.down_proj", "model.layers.90.mlp.experts.71.down_proj", "model.layers.90.mlp.experts.72.down_proj", "model.layers.90.mlp.experts.73.down_proj", "model.layers.90.mlp.experts.74.down_proj", "model.layers.90.mlp.experts.75.down_proj", "model.layers.90.mlp.experts.76.down_proj", "model.layers.90.mlp.experts.77.down_proj", "model.layers.90.mlp.experts.78.down_proj", "model.layers.90.mlp.experts.79.down_proj", "model.layers.90.mlp.experts.80.down_proj", "model.layers.90.mlp.experts.81.down_proj", "model.layers.90.mlp.experts.82.down_proj", "model.layers.90.mlp.experts.83.down_proj", "model.layers.90.mlp.experts.84.down_proj", "model.layers.90.mlp.experts.85.down_proj", "model.layers.90.mlp.experts.86.down_proj", "model.layers.90.mlp.experts.87.down_proj", "model.layers.90.mlp.experts.88.down_proj", "model.layers.90.mlp.experts.89.down_proj", "model.layers.90.mlp.experts.90.down_proj", "model.layers.90.mlp.experts.91.down_proj", "model.layers.90.mlp.experts.92.down_proj", "model.layers.90.mlp.experts.93.down_proj", "model.layers.90.mlp.experts.94.down_proj", "model.layers.90.mlp.experts.95.down_proj", "model.layers.90.mlp.experts.96.down_proj", "model.layers.90.mlp.experts.97.down_proj", "model.layers.90.mlp.experts.98.down_proj", "model.layers.90.mlp.experts.99.down_proj", "model.layers.90.mlp.experts.100.down_proj", "model.layers.90.mlp.experts.101.down_proj", "model.layers.90.mlp.experts.102.down_proj", "model.layers.90.mlp.experts.103.down_proj", "model.layers.90.mlp.experts.104.down_proj", "model.layers.90.mlp.experts.105.down_proj", "model.layers.90.mlp.experts.106.down_proj", "model.layers.90.mlp.experts.107.down_proj", "model.layers.90.mlp.experts.108.down_proj", "model.layers.90.mlp.experts.109.down_proj", "model.layers.90.mlp.experts.110.down_proj", "model.layers.90.mlp.experts.111.down_proj", "model.layers.90.mlp.experts.112.down_proj", "model.layers.90.mlp.experts.113.down_proj", "model.layers.90.mlp.experts.114.down_proj", "model.layers.90.mlp.experts.115.down_proj", "model.layers.90.mlp.experts.116.down_proj", "model.layers.90.mlp.experts.117.down_proj", "model.layers.90.mlp.experts.118.down_proj", "model.layers.90.mlp.experts.119.down_proj", "model.layers.90.mlp.experts.120.down_proj", "model.layers.90.mlp.experts.121.down_proj", "model.layers.90.mlp.experts.122.down_proj", "model.layers.90.mlp.experts.123.down_proj", "model.layers.90.mlp.experts.124.down_proj", "model.layers.90.mlp.experts.125.down_proj", "model.layers.90.mlp.experts.126.down_proj", "model.layers.90.mlp.experts.127.down_proj", "model.layers.90.mlp.experts.128.down_proj", "model.layers.90.mlp.experts.129.down_proj", "model.layers.90.mlp.experts.130.down_proj", "model.layers.90.mlp.experts.131.down_proj", "model.layers.90.mlp.experts.132.down_proj", "model.layers.90.mlp.experts.133.down_proj", "model.layers.90.mlp.experts.134.down_proj", "model.layers.90.mlp.experts.135.down_proj", "model.layers.90.mlp.experts.136.down_proj", "model.layers.90.mlp.experts.137.down_proj", "model.layers.90.mlp.experts.138.down_proj", "model.layers.90.mlp.experts.139.down_proj", "model.layers.90.mlp.experts.140.down_proj", "model.layers.90.mlp.experts.141.down_proj", "model.layers.90.mlp.experts.142.down_proj", "model.layers.90.mlp.experts.143.down_proj", "model.layers.90.mlp.experts.144.down_proj", "model.layers.90.mlp.experts.145.down_proj", "model.layers.90.mlp.experts.146.down_proj", "model.layers.90.mlp.experts.147.down_proj", "model.layers.90.mlp.experts.148.down_proj", "model.layers.90.mlp.experts.149.down_proj", "model.layers.90.mlp.experts.150.down_proj", "model.layers.90.mlp.experts.151.down_proj", "model.layers.90.mlp.experts.152.down_proj", "model.layers.90.mlp.experts.153.down_proj", "model.layers.90.mlp.experts.154.down_proj", "model.layers.90.mlp.experts.155.down_proj", "model.layers.90.mlp.experts.156.down_proj", "model.layers.90.mlp.experts.157.down_proj", "model.layers.90.mlp.experts.158.down_proj", "model.layers.90.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -9.934101253748773e-05, "dbits": 1258291200 }, { "dkld": -9.760893881322064e-05, "dbits": 2516582400 }, { "dkld": -9.08853486180361e-05, "dbits": 3774873600 }, { "dkld": -9.315516799689172e-05, "dbits": 6291456000 } ] }, { "idx": 543, "layers": [ "model.layers.91.self_attn.q_proj" ], "candidates": [ { "dkld": -0.00020282519981265623, "dbits": 62914560 }, { "dkld": -4.3134577572351285e-05, "dbits": 125829120 }, { "dkld": -7.594898343087075e-05, "dbits": 188743680 }, { "dkld": -3.8640201091774684e-05, "dbits": 314572800 } ] }, { "idx": 544, "layers": [ "model.layers.91.self_attn.k_proj", "model.layers.91.self_attn.v_proj" ], "candidates": [ { "dkld": -0.0007203550077974907, "dbits": 10485760 }, { "dkld": -0.0010587766766548268, "dbits": 20971520 }, { "dkld": -0.001104738563299179, "dbits": 31457280 }, { "dkld": -0.0011799175292253605, "dbits": 52428800 } ] }, { "idx": 545, "layers": [ "model.layers.91.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0001588214188814191, "dbits": 62914560 }, { "dkld": -0.00034118406474591134, "dbits": 125829120 }, { "dkld": -0.0003084273077547661, "dbits": 188743680 }, { "dkld": -0.0003397489897906808, "dbits": 314572800 } ] }, { "idx": 546, "layers": [ "model.layers.91.mlp.shared_experts.gate_proj", "model.layers.91.mlp.shared_experts.up_proj", "model.layers.91.mlp.shared_experts.down_proj" ], "candidates": [ { "dkld": 0.0007621382363140472, "dbits": 23592960 }, { "dkld": -5.8660842478283626e-05, "dbits": 47185920 }, { "dkld": 0.00012654149904846312, "dbits": 70778880 }, { "dkld": 0.0002691408619284602, "dbits": 117964800 } ] }, { "idx": 547, "layers": [ "model.layers.91.mlp.experts.0.gate_proj", "model.layers.91.mlp.experts.1.gate_proj", "model.layers.91.mlp.experts.2.gate_proj", "model.layers.91.mlp.experts.3.gate_proj", "model.layers.91.mlp.experts.4.gate_proj", "model.layers.91.mlp.experts.5.gate_proj", "model.layers.91.mlp.experts.6.gate_proj", "model.layers.91.mlp.experts.7.gate_proj", "model.layers.91.mlp.experts.8.gate_proj", "model.layers.91.mlp.experts.9.gate_proj", "model.layers.91.mlp.experts.10.gate_proj", "model.layers.91.mlp.experts.11.gate_proj", "model.layers.91.mlp.experts.12.gate_proj", "model.layers.91.mlp.experts.13.gate_proj", "model.layers.91.mlp.experts.14.gate_proj", "model.layers.91.mlp.experts.15.gate_proj", "model.layers.91.mlp.experts.16.gate_proj", "model.layers.91.mlp.experts.17.gate_proj", "model.layers.91.mlp.experts.18.gate_proj", "model.layers.91.mlp.experts.19.gate_proj", "model.layers.91.mlp.experts.20.gate_proj", "model.layers.91.mlp.experts.21.gate_proj", "model.layers.91.mlp.experts.22.gate_proj", "model.layers.91.mlp.experts.23.gate_proj", "model.layers.91.mlp.experts.24.gate_proj", "model.layers.91.mlp.experts.25.gate_proj", "model.layers.91.mlp.experts.26.gate_proj", "model.layers.91.mlp.experts.27.gate_proj", "model.layers.91.mlp.experts.28.gate_proj", "model.layers.91.mlp.experts.29.gate_proj", "model.layers.91.mlp.experts.30.gate_proj", "model.layers.91.mlp.experts.31.gate_proj", "model.layers.91.mlp.experts.32.gate_proj", "model.layers.91.mlp.experts.33.gate_proj", "model.layers.91.mlp.experts.34.gate_proj", "model.layers.91.mlp.experts.35.gate_proj", "model.layers.91.mlp.experts.36.gate_proj", "model.layers.91.mlp.experts.37.gate_proj", "model.layers.91.mlp.experts.38.gate_proj", "model.layers.91.mlp.experts.39.gate_proj", "model.layers.91.mlp.experts.40.gate_proj", "model.layers.91.mlp.experts.41.gate_proj", "model.layers.91.mlp.experts.42.gate_proj", "model.layers.91.mlp.experts.43.gate_proj", "model.layers.91.mlp.experts.44.gate_proj", "model.layers.91.mlp.experts.45.gate_proj", "model.layers.91.mlp.experts.46.gate_proj", "model.layers.91.mlp.experts.47.gate_proj", "model.layers.91.mlp.experts.48.gate_proj", "model.layers.91.mlp.experts.49.gate_proj", "model.layers.91.mlp.experts.50.gate_proj", "model.layers.91.mlp.experts.51.gate_proj", "model.layers.91.mlp.experts.52.gate_proj", "model.layers.91.mlp.experts.53.gate_proj", "model.layers.91.mlp.experts.54.gate_proj", "model.layers.91.mlp.experts.55.gate_proj", "model.layers.91.mlp.experts.56.gate_proj", "model.layers.91.mlp.experts.57.gate_proj", "model.layers.91.mlp.experts.58.gate_proj", "model.layers.91.mlp.experts.59.gate_proj", "model.layers.91.mlp.experts.60.gate_proj", "model.layers.91.mlp.experts.61.gate_proj", "model.layers.91.mlp.experts.62.gate_proj", "model.layers.91.mlp.experts.63.gate_proj", "model.layers.91.mlp.experts.64.gate_proj", "model.layers.91.mlp.experts.65.gate_proj", "model.layers.91.mlp.experts.66.gate_proj", "model.layers.91.mlp.experts.67.gate_proj", "model.layers.91.mlp.experts.68.gate_proj", "model.layers.91.mlp.experts.69.gate_proj", "model.layers.91.mlp.experts.70.gate_proj", "model.layers.91.mlp.experts.71.gate_proj", "model.layers.91.mlp.experts.72.gate_proj", "model.layers.91.mlp.experts.73.gate_proj", "model.layers.91.mlp.experts.74.gate_proj", "model.layers.91.mlp.experts.75.gate_proj", "model.layers.91.mlp.experts.76.gate_proj", "model.layers.91.mlp.experts.77.gate_proj", "model.layers.91.mlp.experts.78.gate_proj", "model.layers.91.mlp.experts.79.gate_proj", "model.layers.91.mlp.experts.80.gate_proj", "model.layers.91.mlp.experts.81.gate_proj", "model.layers.91.mlp.experts.82.gate_proj", "model.layers.91.mlp.experts.83.gate_proj", "model.layers.91.mlp.experts.84.gate_proj", "model.layers.91.mlp.experts.85.gate_proj", "model.layers.91.mlp.experts.86.gate_proj", "model.layers.91.mlp.experts.87.gate_proj", "model.layers.91.mlp.experts.88.gate_proj", "model.layers.91.mlp.experts.89.gate_proj", "model.layers.91.mlp.experts.90.gate_proj", "model.layers.91.mlp.experts.91.gate_proj", "model.layers.91.mlp.experts.92.gate_proj", "model.layers.91.mlp.experts.93.gate_proj", "model.layers.91.mlp.experts.94.gate_proj", "model.layers.91.mlp.experts.95.gate_proj", "model.layers.91.mlp.experts.96.gate_proj", "model.layers.91.mlp.experts.97.gate_proj", "model.layers.91.mlp.experts.98.gate_proj", "model.layers.91.mlp.experts.99.gate_proj", "model.layers.91.mlp.experts.100.gate_proj", "model.layers.91.mlp.experts.101.gate_proj", "model.layers.91.mlp.experts.102.gate_proj", "model.layers.91.mlp.experts.103.gate_proj", "model.layers.91.mlp.experts.104.gate_proj", "model.layers.91.mlp.experts.105.gate_proj", "model.layers.91.mlp.experts.106.gate_proj", "model.layers.91.mlp.experts.107.gate_proj", "model.layers.91.mlp.experts.108.gate_proj", "model.layers.91.mlp.experts.109.gate_proj", "model.layers.91.mlp.experts.110.gate_proj", "model.layers.91.mlp.experts.111.gate_proj", "model.layers.91.mlp.experts.112.gate_proj", "model.layers.91.mlp.experts.113.gate_proj", "model.layers.91.mlp.experts.114.gate_proj", "model.layers.91.mlp.experts.115.gate_proj", "model.layers.91.mlp.experts.116.gate_proj", "model.layers.91.mlp.experts.117.gate_proj", "model.layers.91.mlp.experts.118.gate_proj", "model.layers.91.mlp.experts.119.gate_proj", "model.layers.91.mlp.experts.120.gate_proj", "model.layers.91.mlp.experts.121.gate_proj", "model.layers.91.mlp.experts.122.gate_proj", "model.layers.91.mlp.experts.123.gate_proj", "model.layers.91.mlp.experts.124.gate_proj", "model.layers.91.mlp.experts.125.gate_proj", "model.layers.91.mlp.experts.126.gate_proj", "model.layers.91.mlp.experts.127.gate_proj", "model.layers.91.mlp.experts.128.gate_proj", "model.layers.91.mlp.experts.129.gate_proj", "model.layers.91.mlp.experts.130.gate_proj", "model.layers.91.mlp.experts.131.gate_proj", "model.layers.91.mlp.experts.132.gate_proj", "model.layers.91.mlp.experts.133.gate_proj", "model.layers.91.mlp.experts.134.gate_proj", "model.layers.91.mlp.experts.135.gate_proj", "model.layers.91.mlp.experts.136.gate_proj", "model.layers.91.mlp.experts.137.gate_proj", "model.layers.91.mlp.experts.138.gate_proj", "model.layers.91.mlp.experts.139.gate_proj", "model.layers.91.mlp.experts.140.gate_proj", "model.layers.91.mlp.experts.141.gate_proj", "model.layers.91.mlp.experts.142.gate_proj", "model.layers.91.mlp.experts.143.gate_proj", "model.layers.91.mlp.experts.144.gate_proj", "model.layers.91.mlp.experts.145.gate_proj", "model.layers.91.mlp.experts.146.gate_proj", "model.layers.91.mlp.experts.147.gate_proj", "model.layers.91.mlp.experts.148.gate_proj", "model.layers.91.mlp.experts.149.gate_proj", "model.layers.91.mlp.experts.150.gate_proj", "model.layers.91.mlp.experts.151.gate_proj", "model.layers.91.mlp.experts.152.gate_proj", "model.layers.91.mlp.experts.153.gate_proj", "model.layers.91.mlp.experts.154.gate_proj", "model.layers.91.mlp.experts.155.gate_proj", "model.layers.91.mlp.experts.156.gate_proj", "model.layers.91.mlp.experts.157.gate_proj", "model.layers.91.mlp.experts.158.gate_proj", "model.layers.91.mlp.experts.159.gate_proj", "model.layers.91.mlp.experts.0.up_proj", "model.layers.91.mlp.experts.1.up_proj", "model.layers.91.mlp.experts.2.up_proj", "model.layers.91.mlp.experts.3.up_proj", "model.layers.91.mlp.experts.4.up_proj", "model.layers.91.mlp.experts.5.up_proj", "model.layers.91.mlp.experts.6.up_proj", "model.layers.91.mlp.experts.7.up_proj", "model.layers.91.mlp.experts.8.up_proj", "model.layers.91.mlp.experts.9.up_proj", "model.layers.91.mlp.experts.10.up_proj", "model.layers.91.mlp.experts.11.up_proj", "model.layers.91.mlp.experts.12.up_proj", "model.layers.91.mlp.experts.13.up_proj", "model.layers.91.mlp.experts.14.up_proj", "model.layers.91.mlp.experts.15.up_proj", "model.layers.91.mlp.experts.16.up_proj", "model.layers.91.mlp.experts.17.up_proj", "model.layers.91.mlp.experts.18.up_proj", "model.layers.91.mlp.experts.19.up_proj", "model.layers.91.mlp.experts.20.up_proj", "model.layers.91.mlp.experts.21.up_proj", "model.layers.91.mlp.experts.22.up_proj", "model.layers.91.mlp.experts.23.up_proj", "model.layers.91.mlp.experts.24.up_proj", "model.layers.91.mlp.experts.25.up_proj", "model.layers.91.mlp.experts.26.up_proj", "model.layers.91.mlp.experts.27.up_proj", "model.layers.91.mlp.experts.28.up_proj", "model.layers.91.mlp.experts.29.up_proj", "model.layers.91.mlp.experts.30.up_proj", "model.layers.91.mlp.experts.31.up_proj", "model.layers.91.mlp.experts.32.up_proj", "model.layers.91.mlp.experts.33.up_proj", "model.layers.91.mlp.experts.34.up_proj", "model.layers.91.mlp.experts.35.up_proj", "model.layers.91.mlp.experts.36.up_proj", "model.layers.91.mlp.experts.37.up_proj", "model.layers.91.mlp.experts.38.up_proj", "model.layers.91.mlp.experts.39.up_proj", "model.layers.91.mlp.experts.40.up_proj", "model.layers.91.mlp.experts.41.up_proj", "model.layers.91.mlp.experts.42.up_proj", "model.layers.91.mlp.experts.43.up_proj", "model.layers.91.mlp.experts.44.up_proj", "model.layers.91.mlp.experts.45.up_proj", "model.layers.91.mlp.experts.46.up_proj", "model.layers.91.mlp.experts.47.up_proj", "model.layers.91.mlp.experts.48.up_proj", "model.layers.91.mlp.experts.49.up_proj", "model.layers.91.mlp.experts.50.up_proj", "model.layers.91.mlp.experts.51.up_proj", "model.layers.91.mlp.experts.52.up_proj", "model.layers.91.mlp.experts.53.up_proj", "model.layers.91.mlp.experts.54.up_proj", "model.layers.91.mlp.experts.55.up_proj", "model.layers.91.mlp.experts.56.up_proj", "model.layers.91.mlp.experts.57.up_proj", "model.layers.91.mlp.experts.58.up_proj", "model.layers.91.mlp.experts.59.up_proj", "model.layers.91.mlp.experts.60.up_proj", "model.layers.91.mlp.experts.61.up_proj", "model.layers.91.mlp.experts.62.up_proj", "model.layers.91.mlp.experts.63.up_proj", "model.layers.91.mlp.experts.64.up_proj", "model.layers.91.mlp.experts.65.up_proj", "model.layers.91.mlp.experts.66.up_proj", "model.layers.91.mlp.experts.67.up_proj", "model.layers.91.mlp.experts.68.up_proj", "model.layers.91.mlp.experts.69.up_proj", "model.layers.91.mlp.experts.70.up_proj", "model.layers.91.mlp.experts.71.up_proj", "model.layers.91.mlp.experts.72.up_proj", "model.layers.91.mlp.experts.73.up_proj", "model.layers.91.mlp.experts.74.up_proj", "model.layers.91.mlp.experts.75.up_proj", "model.layers.91.mlp.experts.76.up_proj", "model.layers.91.mlp.experts.77.up_proj", "model.layers.91.mlp.experts.78.up_proj", "model.layers.91.mlp.experts.79.up_proj", "model.layers.91.mlp.experts.80.up_proj", "model.layers.91.mlp.experts.81.up_proj", "model.layers.91.mlp.experts.82.up_proj", "model.layers.91.mlp.experts.83.up_proj", "model.layers.91.mlp.experts.84.up_proj", "model.layers.91.mlp.experts.85.up_proj", "model.layers.91.mlp.experts.86.up_proj", "model.layers.91.mlp.experts.87.up_proj", "model.layers.91.mlp.experts.88.up_proj", "model.layers.91.mlp.experts.89.up_proj", "model.layers.91.mlp.experts.90.up_proj", "model.layers.91.mlp.experts.91.up_proj", "model.layers.91.mlp.experts.92.up_proj", "model.layers.91.mlp.experts.93.up_proj", "model.layers.91.mlp.experts.94.up_proj", "model.layers.91.mlp.experts.95.up_proj", "model.layers.91.mlp.experts.96.up_proj", "model.layers.91.mlp.experts.97.up_proj", "model.layers.91.mlp.experts.98.up_proj", "model.layers.91.mlp.experts.99.up_proj", "model.layers.91.mlp.experts.100.up_proj", "model.layers.91.mlp.experts.101.up_proj", "model.layers.91.mlp.experts.102.up_proj", "model.layers.91.mlp.experts.103.up_proj", "model.layers.91.mlp.experts.104.up_proj", "model.layers.91.mlp.experts.105.up_proj", "model.layers.91.mlp.experts.106.up_proj", "model.layers.91.mlp.experts.107.up_proj", "model.layers.91.mlp.experts.108.up_proj", "model.layers.91.mlp.experts.109.up_proj", "model.layers.91.mlp.experts.110.up_proj", "model.layers.91.mlp.experts.111.up_proj", "model.layers.91.mlp.experts.112.up_proj", "model.layers.91.mlp.experts.113.up_proj", "model.layers.91.mlp.experts.114.up_proj", "model.layers.91.mlp.experts.115.up_proj", "model.layers.91.mlp.experts.116.up_proj", "model.layers.91.mlp.experts.117.up_proj", "model.layers.91.mlp.experts.118.up_proj", "model.layers.91.mlp.experts.119.up_proj", "model.layers.91.mlp.experts.120.up_proj", "model.layers.91.mlp.experts.121.up_proj", "model.layers.91.mlp.experts.122.up_proj", "model.layers.91.mlp.experts.123.up_proj", "model.layers.91.mlp.experts.124.up_proj", "model.layers.91.mlp.experts.125.up_proj", "model.layers.91.mlp.experts.126.up_proj", "model.layers.91.mlp.experts.127.up_proj", "model.layers.91.mlp.experts.128.up_proj", "model.layers.91.mlp.experts.129.up_proj", "model.layers.91.mlp.experts.130.up_proj", "model.layers.91.mlp.experts.131.up_proj", "model.layers.91.mlp.experts.132.up_proj", "model.layers.91.mlp.experts.133.up_proj", "model.layers.91.mlp.experts.134.up_proj", "model.layers.91.mlp.experts.135.up_proj", "model.layers.91.mlp.experts.136.up_proj", "model.layers.91.mlp.experts.137.up_proj", "model.layers.91.mlp.experts.138.up_proj", "model.layers.91.mlp.experts.139.up_proj", "model.layers.91.mlp.experts.140.up_proj", "model.layers.91.mlp.experts.141.up_proj", "model.layers.91.mlp.experts.142.up_proj", "model.layers.91.mlp.experts.143.up_proj", "model.layers.91.mlp.experts.144.up_proj", "model.layers.91.mlp.experts.145.up_proj", "model.layers.91.mlp.experts.146.up_proj", "model.layers.91.mlp.experts.147.up_proj", "model.layers.91.mlp.experts.148.up_proj", "model.layers.91.mlp.experts.149.up_proj", "model.layers.91.mlp.experts.150.up_proj", "model.layers.91.mlp.experts.151.up_proj", "model.layers.91.mlp.experts.152.up_proj", "model.layers.91.mlp.experts.153.up_proj", "model.layers.91.mlp.experts.154.up_proj", "model.layers.91.mlp.experts.155.up_proj", "model.layers.91.mlp.experts.156.up_proj", "model.layers.91.mlp.experts.157.up_proj", "model.layers.91.mlp.experts.158.up_proj", "model.layers.91.mlp.experts.159.up_proj" ], "candidates": [ { "dkld": -7.74919986724909e-05, "dbits": 2516582400 }, { "dkld": -8.284412324428836e-05, "dbits": 5033164800 }, { "dkld": -9.35871154069956e-05, "dbits": 7549747200 }, { "dkld": -9.8340585827833e-05, "dbits": 12582912000 } ] }, { "idx": 548, "layers": [ "model.layers.91.mlp.experts.0.down_proj", "model.layers.91.mlp.experts.1.down_proj", "model.layers.91.mlp.experts.2.down_proj", "model.layers.91.mlp.experts.3.down_proj", "model.layers.91.mlp.experts.4.down_proj", "model.layers.91.mlp.experts.5.down_proj", "model.layers.91.mlp.experts.6.down_proj", "model.layers.91.mlp.experts.7.down_proj", "model.layers.91.mlp.experts.8.down_proj", "model.layers.91.mlp.experts.9.down_proj", "model.layers.91.mlp.experts.10.down_proj", "model.layers.91.mlp.experts.11.down_proj", "model.layers.91.mlp.experts.12.down_proj", "model.layers.91.mlp.experts.13.down_proj", "model.layers.91.mlp.experts.14.down_proj", "model.layers.91.mlp.experts.15.down_proj", "model.layers.91.mlp.experts.16.down_proj", "model.layers.91.mlp.experts.17.down_proj", "model.layers.91.mlp.experts.18.down_proj", "model.layers.91.mlp.experts.19.down_proj", "model.layers.91.mlp.experts.20.down_proj", "model.layers.91.mlp.experts.21.down_proj", "model.layers.91.mlp.experts.22.down_proj", "model.layers.91.mlp.experts.23.down_proj", "model.layers.91.mlp.experts.24.down_proj", "model.layers.91.mlp.experts.25.down_proj", "model.layers.91.mlp.experts.26.down_proj", "model.layers.91.mlp.experts.27.down_proj", "model.layers.91.mlp.experts.28.down_proj", "model.layers.91.mlp.experts.29.down_proj", "model.layers.91.mlp.experts.30.down_proj", "model.layers.91.mlp.experts.31.down_proj", "model.layers.91.mlp.experts.32.down_proj", "model.layers.91.mlp.experts.33.down_proj", "model.layers.91.mlp.experts.34.down_proj", "model.layers.91.mlp.experts.35.down_proj", "model.layers.91.mlp.experts.36.down_proj", "model.layers.91.mlp.experts.37.down_proj", "model.layers.91.mlp.experts.38.down_proj", "model.layers.91.mlp.experts.39.down_proj", "model.layers.91.mlp.experts.40.down_proj", "model.layers.91.mlp.experts.41.down_proj", "model.layers.91.mlp.experts.42.down_proj", "model.layers.91.mlp.experts.43.down_proj", "model.layers.91.mlp.experts.44.down_proj", "model.layers.91.mlp.experts.45.down_proj", "model.layers.91.mlp.experts.46.down_proj", "model.layers.91.mlp.experts.47.down_proj", "model.layers.91.mlp.experts.48.down_proj", "model.layers.91.mlp.experts.49.down_proj", "model.layers.91.mlp.experts.50.down_proj", "model.layers.91.mlp.experts.51.down_proj", "model.layers.91.mlp.experts.52.down_proj", "model.layers.91.mlp.experts.53.down_proj", "model.layers.91.mlp.experts.54.down_proj", "model.layers.91.mlp.experts.55.down_proj", "model.layers.91.mlp.experts.56.down_proj", "model.layers.91.mlp.experts.57.down_proj", "model.layers.91.mlp.experts.58.down_proj", "model.layers.91.mlp.experts.59.down_proj", "model.layers.91.mlp.experts.60.down_proj", "model.layers.91.mlp.experts.61.down_proj", "model.layers.91.mlp.experts.62.down_proj", "model.layers.91.mlp.experts.63.down_proj", "model.layers.91.mlp.experts.64.down_proj", "model.layers.91.mlp.experts.65.down_proj", "model.layers.91.mlp.experts.66.down_proj", "model.layers.91.mlp.experts.67.down_proj", "model.layers.91.mlp.experts.68.down_proj", "model.layers.91.mlp.experts.69.down_proj", "model.layers.91.mlp.experts.70.down_proj", "model.layers.91.mlp.experts.71.down_proj", "model.layers.91.mlp.experts.72.down_proj", "model.layers.91.mlp.experts.73.down_proj", "model.layers.91.mlp.experts.74.down_proj", "model.layers.91.mlp.experts.75.down_proj", "model.layers.91.mlp.experts.76.down_proj", "model.layers.91.mlp.experts.77.down_proj", "model.layers.91.mlp.experts.78.down_proj", "model.layers.91.mlp.experts.79.down_proj", "model.layers.91.mlp.experts.80.down_proj", "model.layers.91.mlp.experts.81.down_proj", "model.layers.91.mlp.experts.82.down_proj", "model.layers.91.mlp.experts.83.down_proj", "model.layers.91.mlp.experts.84.down_proj", "model.layers.91.mlp.experts.85.down_proj", "model.layers.91.mlp.experts.86.down_proj", "model.layers.91.mlp.experts.87.down_proj", "model.layers.91.mlp.experts.88.down_proj", "model.layers.91.mlp.experts.89.down_proj", "model.layers.91.mlp.experts.90.down_proj", "model.layers.91.mlp.experts.91.down_proj", "model.layers.91.mlp.experts.92.down_proj", "model.layers.91.mlp.experts.93.down_proj", "model.layers.91.mlp.experts.94.down_proj", "model.layers.91.mlp.experts.95.down_proj", "model.layers.91.mlp.experts.96.down_proj", "model.layers.91.mlp.experts.97.down_proj", "model.layers.91.mlp.experts.98.down_proj", "model.layers.91.mlp.experts.99.down_proj", "model.layers.91.mlp.experts.100.down_proj", "model.layers.91.mlp.experts.101.down_proj", "model.layers.91.mlp.experts.102.down_proj", "model.layers.91.mlp.experts.103.down_proj", "model.layers.91.mlp.experts.104.down_proj", "model.layers.91.mlp.experts.105.down_proj", "model.layers.91.mlp.experts.106.down_proj", "model.layers.91.mlp.experts.107.down_proj", "model.layers.91.mlp.experts.108.down_proj", "model.layers.91.mlp.experts.109.down_proj", "model.layers.91.mlp.experts.110.down_proj", "model.layers.91.mlp.experts.111.down_proj", "model.layers.91.mlp.experts.112.down_proj", "model.layers.91.mlp.experts.113.down_proj", "model.layers.91.mlp.experts.114.down_proj", "model.layers.91.mlp.experts.115.down_proj", "model.layers.91.mlp.experts.116.down_proj", "model.layers.91.mlp.experts.117.down_proj", "model.layers.91.mlp.experts.118.down_proj", "model.layers.91.mlp.experts.119.down_proj", "model.layers.91.mlp.experts.120.down_proj", "model.layers.91.mlp.experts.121.down_proj", "model.layers.91.mlp.experts.122.down_proj", "model.layers.91.mlp.experts.123.down_proj", "model.layers.91.mlp.experts.124.down_proj", "model.layers.91.mlp.experts.125.down_proj", "model.layers.91.mlp.experts.126.down_proj", "model.layers.91.mlp.experts.127.down_proj", "model.layers.91.mlp.experts.128.down_proj", "model.layers.91.mlp.experts.129.down_proj", "model.layers.91.mlp.experts.130.down_proj", "model.layers.91.mlp.experts.131.down_proj", "model.layers.91.mlp.experts.132.down_proj", "model.layers.91.mlp.experts.133.down_proj", "model.layers.91.mlp.experts.134.down_proj", "model.layers.91.mlp.experts.135.down_proj", "model.layers.91.mlp.experts.136.down_proj", "model.layers.91.mlp.experts.137.down_proj", "model.layers.91.mlp.experts.138.down_proj", "model.layers.91.mlp.experts.139.down_proj", "model.layers.91.mlp.experts.140.down_proj", "model.layers.91.mlp.experts.141.down_proj", "model.layers.91.mlp.experts.142.down_proj", "model.layers.91.mlp.experts.143.down_proj", "model.layers.91.mlp.experts.144.down_proj", "model.layers.91.mlp.experts.145.down_proj", "model.layers.91.mlp.experts.146.down_proj", "model.layers.91.mlp.experts.147.down_proj", "model.layers.91.mlp.experts.148.down_proj", "model.layers.91.mlp.experts.149.down_proj", "model.layers.91.mlp.experts.150.down_proj", "model.layers.91.mlp.experts.151.down_proj", "model.layers.91.mlp.experts.152.down_proj", "model.layers.91.mlp.experts.153.down_proj", "model.layers.91.mlp.experts.154.down_proj", "model.layers.91.mlp.experts.155.down_proj", "model.layers.91.mlp.experts.156.down_proj", "model.layers.91.mlp.experts.157.down_proj", "model.layers.91.mlp.experts.158.down_proj", "model.layers.91.mlp.experts.159.down_proj" ], "candidates": [ { "dkld": -4.457132890821336e-05, "dbits": 1258291200 }, { "dkld": -3.0472781509166547e-05, "dbits": 2516582400 }, { "dkld": -4.8959441483031885e-05, "dbits": 3774873600 }, { "dkld": -5.1709916442641e-05, "dbits": 6291456000 } ] } ], "base_kld": 0.07912451997399331, "arch_string": "Glm4MoeForCausalLM" }