{
 "weight_format": "affine",
 "profile": "JANG_2L_GS64_ProjLayerBits_Ggs64-Dgs32-Ugs64_Attn8g64_Tok8g64_NoMTP_AWQ_DiagImatrix_QAT_GPTQ",
 "source_model": {
  "repository": "deepseek-ai/DeepSeek-V4-Flash-0731",
  "release_identity": "DeepSeek-V4-Flash-0731",
  "revision": "9e165c30e2704aec5d9d593cce3eebd58bbef1cb",
  "source_path": "/Users/eric/sources/DeepSeek-V4-Flash-0731-9e165c30e270",
  "release_lock_sha256": "d8f218816eb9917c50f1773948002c670c8ebb76b862abc04790d15ca0e20d53",
  "config_sha256": "6c8f3d2d3b48707541b88f32f22ef3f0f8a6b57d8523281e2b8d3cdb0ae9a023",
  "weight_index_sha256": "98efab455cf08dfbbbaaba6f570e1bf10bf927d2b4c3c453a59c2f6f0e3be92b"
 },
 "source_revision": "9e165c30e2704aec5d9d593cce3eebd58bbef1cb",
 "critical_f32_preserved": true,
 "awq": {
  "enabled": true,
  "method": "normalized_ffn_input_fold",
  "scope": "routed/shared w1+w3 plus router, inverse ffn_norm",
  "alpha": 0.25,
  "scale_clip": [
   0.5,
   2.0
  ],
  "calibrated_layers": [
   0,
   1,
   2,
   3,
   4,
   5,
   6,
   7,
   8,
   9,
   10,
   11,
   12,
   13,
   14,
   15,
   16,
   17,
   18,
   19,
   20,
   21,
   22,
   23,
   24,
   25,
   26,
   27,
   28,
   29,
   30,
   31,
   32,
   33,
   34,
   35,
   36,
   37,
   38,
   39,
   40,
   41,
   42
  ],
  "calibration_sha256": "22efb070d5357ee937bbe67140b710dbc6d6dceaf99d550b82ebbae94e602c11",
  "folded_tensor_counts": {
   "ffn_norm": 43,
   "router": 43,
   "routed_w1_w3": 22016,
   "shared_w1_w3": 86
  },
  "runtime_sidecar_required": false
 },
 "imatrix": {
  "enabled": true,
  "method": "diagonal_activation_second_moment_fold",
  "scope": "routed w2 columns plus inverse routed w3 rows",
  "alpha": 0.25,
  "scale_clip": [
   0.5,
   2.0
  ],
  "calibrated_layers": [
   0,
   1,
   2,
   3,
   4,
   5,
   6,
   7,
   8,
   9,
   10,
   11,
   12,
   13,
   14,
   15,
   16,
   17,
   18,
   19,
   20,
   21,
   22,
   23,
   24,
   25,
   26,
   27,
   28,
   29,
   30,
   31,
   32,
   33,
   34,
   35,
   36,
   37,
   38,
   39,
   40,
   41,
   42
  ],
  "calibration_sha256": "22efb070d5357ee937bbe67140b710dbc6d6dceaf99d550b82ebbae94e602c11",
  "folded_tensor_counts": {
   "routed_w2": 11008,
   "routed_w3_inverse": 11008
  },
  "codec": "mlx_affine",
  "runtime_sidecar_required": false
 },
 "dsv4_runtime_requirements": {
  "limited_swiglu_tq_patch": false,
  "generic_mlx_sinks": false,
  "native_cache_schema": "deepseek_v4_v9",
  "generic_turboquant_kv": false,
  "long_ctx_default": true,
  "pool_quant_default": true,
  "max_num_seqs": 1
 },
 "affine_bits": {
  "routed_expert": 2,
  "attention": 8,
  "shared_expert": 8,
  "embed_tokens": 8,
  "lm_head": 8,
  "mtp_routed_expert": 2,
  "mtp_non_routed_matmul": 8,
  "norms_router_hc": 16
 },
 "affine_group_size": {
  "routed_expert": 64,
  "attention": 64,
  "shared_expert": 64,
  "embed_tokens": 64,
  "lm_head": 64,
  "mtp_routed_expert": 64,
  "mtp_non_routed_matmul": 64
 },
 "quantization": {
  "method": "affine",
  "top_level_default": {
   "bits": 2,
   "group_size": 64,
   "mode": "affine"
  },
  "routed_experts": {
   "bits": 2,
   "codec": "affine",
   "group_size": 64,
   "bit_plan": {
    "default_bits": 2,
    "codec": "affine",
    "group_size": 64,
    "routed_layer_bits": {},
    "routed_projection_bits": {},
    "mtp_routed_bits": 2,
    "mtp_routed_projection_bits": {},
    "routed_down_layer_bits": {},
    "routed_projection_layer_bits": {
     "w1": {
      "5": 3,
      "14": 3,
      "30": 3,
      "34": 3,
      "37": 3,
      "42": 3
     }
    },
    "routed_projection_group_sizes": {
     "w1": 64,
     "w2": 32,
     "w3": 64
    },
    "routed_projection_layer_group_sizes": {}
   }
  },
  "non_routed": {
   "bits": 8,
   "codec": "affine",
   "group_size": 64
  },
  "token_bookends": {
   "bits": 8,
   "codec": "affine",
   "group_size": 64,
   "override": true
  },
  "attention": {
   "bits": 8,
   "codec": "affine",
   "group_size": 64,
   "override": true
  },
  "critical_control_tensors": "source-f32",
  "override_count": 13056,
  "group_totals": {
   "8b_g64": 512,
   "2b_g64": 20480,
   "2b_g32": 11008,
   "3b_g64": 1536
  }
 },
 "routed_layer_bits": {},
 "routed_projection_bits": {},
 "routed_down_layer_bits": {},
 "routed_projection_layer_bits": {
  "w1": {
   "5": 3,
   "14": 3,
   "30": 3,
   "34": 3,
   "37": 3,
   "42": 3
  }
 },
 "routed_projection_group_sizes": {
  "w1": 64,
  "w2": 32,
  "w3": 64
 },
 "routed_projection_layer_group_sizes": {},
 "cache": {
  "schema": "deepseek_v4_v9",
  "components": [
   "swa",
   "csa",
   "hca",
   "compressor",
   "indexer"
  ],
  "sliding_window": 128,
  "compress_ratios": [
   0,
   0,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   0,
   0,
   0
  ],
  "generic_turboquant_kv": false,
  "pool_quant_default": true,
  "mtp_activation_requires_draft_cache": true
 },
 "mtp": {
  "preserved": false,
  "runtime_self_spec_enabled": false,
  "mode": "dropped",
  "num_nextn_predict_layers": 0,
  "activation_requires": "separate MTP drafter, draft cache, accept/reject verifier, and DSV4 SWA+CSA/HSA composite-cache-safe rollback"
 },
 "dspark": {
  "preserved": false,
  "runtime_enabled": false,
  "mode": "dropped",
  "stage_count": 3,
  "stage_ids": [
   0,
   1,
   2
  ],
  "inference_n_mtp_layers": 3,
  "block_size": 5,
  "noise_token_id": 128799,
  "target_main_layers": [
   40,
   41,
   42
  ],
  "markov_rank": 256,
  "activation_requires": [
   "main hidden-state capture at target layers",
   "five-token DSpark draft block",
   "Markov and confidence heads",
   "SWA+CSA+HCA composite-cache atomic rollback"
  ]
 },
 "runtime": {
  "bundle_has_mtp": false,
  "mtp_layers": 0,
  "mtp_mode": "dropped"
 },
 "source_config": {
  "n_routed_experts": 256,
  "num_experts_per_tok": 6,
  "num_hidden_layers": 43,
  "num_nextn_predict_layers": 1,
  "inference_n_mtp_layers": 3,
  "sliding_window": 128,
  "compress_ratios": [
   0,
   0,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   128,
   4,
   0,
   0,
   0
  ],
  "hc_mult": 4,
  "hc_sinkhorn_iters": 20,
  "swiglu_limit": 10.0,
  "routed_scaling_factor": 1.5
 },
 "model_family": "deepseek_v4",
 "chat": {
  "encoder": "encoding_dsv4",
  "encoder_fn": "encode_messages",
  "chat_template_source": "official_python_encoder",
  "has_tokenizer_chat_template": false,
  "bos_token": "<\uff5cbegin\u2581of\u2581sentence\uff5c>",
  "eos_token": "<\uff5cend\u2581of\u2581sentence\uff5c>",
  "bos_token_id": 0,
  "eos_token_id": 1,
  "role_tokens": {
   "user": "<\uff5cUser\uff5c>",
   "assistant": "<\uff5cAssistant\uff5c>",
   "latest_reminder": "<\uff5clatest_reminder\uff5c>"
  },
  "reasoning": {
   "supported": true,
   "modes": [
    "chat",
    "thinking"
   ],
   "default_mode": "thinking",
   "default_effort": "low",
   "thinking_start": "<think>",
   "thinking_end": "</think>",
   "reasoning_effort_levels": [
    "low",
    "high",
    "max"
   ],
   "drop_earlier_reasoning": true
  },
  "tool_calling": {
   "supported": true,
   "parser": "dsml",
   "dsml_token": "\uff5cDSML\uff5c",
   "tool_calls_block": "tool_calls",
   "invoke_block": "invoke",
   "parameter_block": "parameter",
   "tool_output_tag": "tool_result"
  },
  "sampling_defaults": {
   "_from_model_config": true,
   "bos_token_id": 0,
   "eos_token_id": 1,
   "do_sample": true,
   "temperature": 0.6,
   "top_p": 0.95,
   "transformers_version": "4.46.3",
   "top_k": 0,
   "temperature_note": "coding default override 2026-08-03; vendor card documents 1.0"
  }
 },
 "qat": {
  "enabled": true,
  "method": "gptq_error_compensated_rounding",
  "scope": "all routed expert w1/w2/w3 codes; scales/biases min-max f16 (byte-identical grid to the non-QAT sibling); non-routed and critical controls unchanged",
  "objective": "per-expert routing-weighted output reconstruction; Hessian H = sum_k w_k x x^T per expert (f64 factorization, escalating damping from 0.01), blocked GPTQ column sweep on the fixed min-max grid; BRECQ sequencing (w1/w3 first, w2 on inputs derived through the quantized w1/w3 and DSV4 limited SwiGLU)",
  "guard": "per-expert best-of vs min-max codes; experts with <24 routed calibration rows always keep min-max codes",
  "calibration": "11113-row agentic-coding X1 capture through 94.995GiB NR8 keeper on vMLX Python 1.6.22, pool quant on, three prompts crossing the 2052-token indexer boundary",
  "codec": "unchanged stock MLX affine gather_qmm",
  "per_layer_recon_improvement": {
   "0": {
    "w1": 0.987,
    "w2": 0.945,
    "w3": 0.987
   },
   "1": {
    "w1": 0.982,
    "w2": 0.846,
    "w3": 0.982
   },
   "2": {
    "w1": 0.983,
    "w2": 0.943,
    "w3": 0.983
   },
   "3": {
    "w1": 0.977,
    "w2": 0.939,
    "w3": 0.977
   },
   "4": {
    "w1": 0.977,
    "w2": 0.877,
    "w3": 0.977
   },
   "5": {
    "w1": 0.979,
    "w2": 0.939,
    "w3": 0.973
   },
   "6": {
    "w1": 0.974,
    "w2": 0.907,
    "w3": 0.974
   },
   "7": {
    "w1": 0.969,
    "w2": 0.931,
    "w3": 0.969
   },
   "8": {
    "w1": 0.97,
    "w2": 0.931,
    "w3": 0.97
   },
   "9": {
    "w1": 0.973,
    "w2": 0.844,
    "w3": 0.973
   },
   "10": {
    "w1": 0.971,
    "w2": 0.919,
    "w3": 0.971
   },
   "11": {
    "w1": 0.969,
    "w2": 0.926,
    "w3": 0.969
   },
   "12": {
    "w1": 0.969,
    "w2": 0.92,
    "w3": 0.969
   },
   "13": {
    "w1": 0.963,
    "w2": 0.806,
    "w3": 0.963
   },
   "14": {
    "w1": 0.968,
    "w2": 0.926,
    "w3": 0.961
   },
   "15": {
    "w1": 0.969,
    "w2": 0.926,
    "w3": 0.969
   },
   "16": {
    "w1": 0.969,
    "w2": 0.926,
    "w3": 0.969
   },
   "17": {
    "w1": 0.966,
    "w2": 0.928,
    "w3": 0.965
   },
   "18": {
    "w1": 0.97,
    "w2": 0.924,
    "w3": 0.97
   },
   "19": {
    "w1": 0.976,
    "w2": 0.935,
    "w3": 0.976
   },
   "20": {
    "w1": 0.971,
    "w2": 0.944,
    "w3": 0.97
   },
   "21": {
    "w1": 0.971,
    "w2": 0.936,
    "w3": 0.971
   },
   "22": {
    "w1": 0.972,
    "w2": 0.935,
    "w3": 0.971
   },
   "23": {
    "w1": 0.972,
    "w2": 0.923,
    "w3": 0.971
   },
   "24": {
    "w1": 0.969,
    "w2": 0.944,
    "w3": 0.968
   },
   "25": {
    "w1": 0.969,
    "w2": 0.927,
    "w3": 0.969
   },
   "26": {
    "w1": 0.969,
    "w2": 0.983,
    "w3": 0.969
   },
   "27": {
    "w1": 0.97,
    "w2": 0.958,
    "w3": 0.97
   },
   "28": {
    "w1": 0.97,
    "w2": 0.927,
    "w3": 0.97
   },
   "29": {
    "w1": 0.972,
    "w2": 0.999,
    "w3": 0.972
   },
   "30": {
    "w1": 0.974,
    "w2": 0.025,
    "w3": 0.966
   },
   "31": {
    "w1": 0.973,
    "w2": 0.996,
    "w3": 0.973
   },
   "32": {
    "w1": 0.973,
    "w2": 0.993,
    "w3": 0.973
   },
   "33": {
    "w1": 0.974,
    "w2": 0.994,
    "w3": 0.973
   },
   "34": {
    "w1": 0.978,
    "w2": 0.998,
    "w3": 0.971
   },
   "35": {
    "w1": 0.974,
    "w2": 0.998,
    "w3": 0.974
   },
   "36": {
    "w1": 0.969,
    "w2": 0.998,
    "w3": 0.969
   },
   "37": {
    "w1": 0.979,
    "w2": 0.997,
    "w3": 0.972
   },
   "38": {
    "w1": 0.969,
    "w2": 0.998,
    "w3": 0.969
   },
   "39": {
    "w1": 0.973,
    "w2": 0.99,
    "w3": 0.973
   },
   "40": {
    "w1": 0.975,
    "w2": 0.998,
    "w3": 0.974
   },
   "41": {
    "w1": 0.972,
    "w2": 0.206,
    "w3": 0.972
   },
   "42": {
    "w1": 0.982,
    "w2": 0.997,
    "w3": 0.978
   }
  },
  "date": "2026-08-03"
 }
}