{
  "analysis_sha256": "a53a03e535dc92cbc2217c939f818851c3468684ba93eee698c536f09b773375",
  "architecture_profile": {
    "adapter_id": "qwen3-vl-moe-v1",
    "audio_present": false,
    "config_model_type": "qwen3_vl_moe",
    "dense": false,
    "mtp_declared": false,
    "notes": [
      "Thin convert: Qwen3-VL-30B-A3B-Instruct MoE via MLX-VLM convert path.",
      "Primary product runtime is AX Engine; MLX-VLM is the vision-compatible runtime.",
      "Vision tower remains protected at BF16; development evidence only; no MTP claim.",
      "MoE experts quantize as fused switch modules with uniform per-group precision; artifacts are development evidence until certified."
    ],
    "optimization_scope": "text-path",
    "product_family": "qwen3-vl",
    "support_level": "supported",
    "support_tier": "convertible",
    "text_layer_count": 48,
    "vision_present": true
  },
  "assignments": [
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.0253125,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.045000000000000005,
        "long_context_loss": 0.016875,
        "mtp_acceptance_loss": 0.0045000000000000005,
        "output_kl": 0.05625,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.036562500000000005,
        "token_disagreement": 0.030937500000000003
      },
      "module_path": "model.language_model.embed_tokens",
      "outlier_strategy": "none",
      "parameters": 311164928,
      "predicted_loss": 0.0307125,
      "reason": "protected embedding policy",
      "role": "embedding",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.embed_tokens.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.0.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.0.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.0.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.0.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.0.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.0.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.0.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.0.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.0.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.0.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.0.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.0.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.0.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.0.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.1.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.1.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.1.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.1.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.1.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.1.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.1.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.1.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.1.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.1.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.1.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.1.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.1.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.1.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.2.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.2.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.2.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.2.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.2.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.2.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.2.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.2.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.2.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.2.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.2.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.2.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.2.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.2.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.3.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.3.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.3.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.3.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.3.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.3.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.3.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.3.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.3.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.3.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.3.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.3.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.3.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.3.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.3.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.3.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.3.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.3.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.4.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.4.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.4.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.4.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.4.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.4.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.4.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.4.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.4.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.4.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.4.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.4.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.4.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.4.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.5.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.5.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.5.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.5.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.5.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.5.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.5.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.5.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.5.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.5.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.5.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.5.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.5.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.5.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.6.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.6.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.6.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.6.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.6.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.6.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.6.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.6.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.6.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.6.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.6.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.6.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.6.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.6.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.7.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.7.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.7.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.7.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.7.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.7.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.7.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.7.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.7.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.7.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.7.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.7.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.7.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.7.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.10.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.10.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.10.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.10.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.10.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.10.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.10.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.10.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.10.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.10.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.10.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.10.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.10.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.10.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.11.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.11.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.11.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.11.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.11.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.11.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.11.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.11.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.11.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.11.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.11.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.11.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.11.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.11.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.7.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.7.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.7.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.7.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.8.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.8.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.8.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.8.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.8.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.8.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.8.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.8.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.8.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.8.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.8.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.8.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.8.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.8.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.9.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.9.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.9.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.9.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.9.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.9.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.9.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.9.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.9.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.9.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.9.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.9.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.9.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.9.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.11.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.11.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.11.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.11.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.12.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.12.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.12.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.12.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.12.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.12.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.12.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.12.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.12.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.12.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.12.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.12.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.12.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.12.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.13.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.13.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.13.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.13.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.13.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.13.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.13.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.13.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.13.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.13.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.13.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.13.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.13.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.13.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.14.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.14.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.14.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.14.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.14.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.14.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.14.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.14.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.14.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.14.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.14.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.14.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.14.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.14.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.15.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.15.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.15.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.15.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.15.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.15.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.15.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.15.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.15.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.15.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.15.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.15.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.15.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.15.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.15.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.15.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.15.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.15.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.16.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.16.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.16.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.16.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.16.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.16.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.16.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.16.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.16.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.16.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.16.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.16.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.16.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.16.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.17.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.17.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.17.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.17.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.17.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.17.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.17.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.17.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.17.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.17.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.17.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.17.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.17.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.17.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.18.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.18.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.18.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.18.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.18.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.18.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.18.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.18.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.18.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.18.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.18.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.18.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.18.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.18.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.19.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.19.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.19.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.19.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.19.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.19.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.19.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.19.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.19.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.19.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.19.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.19.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.19.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.19.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.19.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.19.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.19.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.19.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.20.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.20.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.20.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.20.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.20.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.20.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.20.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.20.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.20.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.20.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.20.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.20.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.20.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.20.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.21.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.21.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.21.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.21.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.21.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.21.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.21.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.21.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.21.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.21.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.21.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.21.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.21.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.21.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.22.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.22.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.22.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.22.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.22.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.22.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.22.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.22.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.22.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.22.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.22.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.22.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.22.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.22.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.23.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.23.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.23.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.23.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.23.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.23.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.23.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.23.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.23.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.23.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.23.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.23.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.23.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.23.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.23.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.23.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.23.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.23.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.24.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.24.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.24.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.24.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.24.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.24.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.24.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.24.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.24.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.24.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.24.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.24.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.24.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.24.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.25.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.25.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.25.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.25.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.25.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.25.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.25.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.25.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.25.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.25.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.25.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.25.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.25.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.25.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.26.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.input_layernorm.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.26.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.26.mlp.experts.down_proj"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.26.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.26.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.26.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.26.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.26.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.26.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.26.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.26.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.26.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.26.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.26.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.27.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.27.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.27.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.27.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.27.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.27.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.27.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.27.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.27.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.27.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.27.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.27.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.27.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.27.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.self_attn.v_proj.weight"
    },
    {
      "bits": 4,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.12091525958289964,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.2149604614807105,
        "long_context_loss": 0.08061017305526641,
        "mtp_acceptance_loss": 0.021496046148071046,
        "output_kl": 0.2687005768508881,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.17465537495307726,
        "token_disagreement": 0.14778531726798846
      },
      "module_path": "model.language_model.layers.27.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.1467105149605849,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 5.0
      },
      "tensor": "model.language_model.layers.27.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.27.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.27.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.28.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.28.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.28.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.28.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.28.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.28.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.28.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.28.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.28.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.28.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.28.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.28.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.28.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.28.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.29.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.29.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.29.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.29.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.29.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.29.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.29.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.29.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.29.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.29.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.29.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.29.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.29.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.29.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.30.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.30.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.30.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.30.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.30.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.30.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.30.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.30.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.30.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.30.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.30.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.30.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.30.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.30.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.31.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.31.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.31.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.31.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.31.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.31.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.31.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.31.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.31.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.31.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.31.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.31.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.31.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.31.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.self_attn.v_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.31.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.31.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.31.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.31.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.32.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.32.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.32.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.32.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.32.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.32.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.32.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.32.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.32.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.32.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.32.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.32.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.32.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.32.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.33.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.33.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.33.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.33.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.33.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.33.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.33.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.33.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.33.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.33.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.33.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.33.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.33.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.33.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.34.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.34.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.34.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.34.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.34.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.34.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.34.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.34.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.34.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.34.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.34.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.34.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.34.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.34.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.35.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.35.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.35.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.35.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.35.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.35.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.35.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.35.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.35.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.35.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.35.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.35.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.35.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.35.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.self_attn.v_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.35.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.35.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.35.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.35.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.36.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.36.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.36.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.36.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.36.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.36.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.36.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.36.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.36.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.36.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.36.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.36.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.36.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.36.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.37.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.37.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.37.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.37.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.37.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.37.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.37.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.37.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.37.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.37.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.37.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.37.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.37.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.37.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.38.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.38.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.38.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.38.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.38.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.38.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.38.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.38.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.38.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.38.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.38.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.38.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.38.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.38.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.39.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.39.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.39.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.39.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.39.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.39.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.39.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.39.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.39.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.39.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.39.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.39.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.39.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.39.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.self_attn.v_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.39.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.39.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.39.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.39.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.40.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.40.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.40.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.40.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.40.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.40.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.40.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.40.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.40.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.40.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.40.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.40.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.40.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.40.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.41.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.41.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.41.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.41.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.41.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.41.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.41.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.41.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.41.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.41.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.41.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.41.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.41.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.41.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.42.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.42.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.42.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.42.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.42.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.42.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.42.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.42.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.42.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.42.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.42.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.42.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.42.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.self_attn.q_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.42.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.43.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.43.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.43.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.43.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.43.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.43.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.43.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.43.self_attn.k_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.43.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.43.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.43.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.43.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.43.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.43.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.43.self_attn.v_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.43.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.43.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.43.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.43.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.44.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.44.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.44.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.44.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.44.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.44.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.44.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.44.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.44.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.self_attn.k_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.44.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.44.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.44.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.44.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.44.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.44.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.44.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.45.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.45.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.45.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.45.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.45.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.030937500000000003,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.05500000000000001,
        "long_context_loss": 0.020625,
        "mtp_acceptance_loss": 0.0055000000000000005,
        "output_kl": 0.06875,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.044687500000000005,
        "token_disagreement": 0.037812500000000006
      },
      "module_path": "model.language_model.layers.45.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.0375375,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.45.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.45.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.45.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.self_attn.k_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.45.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.45.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.45.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.45.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.self_attn.q_norm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.45.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.45.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.45.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.46.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.46.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.46.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.46.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.46.mlp.experts.gate_up_proj"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.021876116042958818,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.03889087296526012,
        "long_context_loss": 0.014584077361972545,
        "mtp_acceptance_loss": 0.003889087296526012,
        "output_kl": 0.04861359120657515,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.03159883428427385,
        "token_disagreement": 0.026737475163616336
      },
      "module_path": "model.language_model.layers.46.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.02654302079879003,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.46.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.46.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.46.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.self_attn.k_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.46.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.46.self_attn.k_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.057275649276110355,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.10182337649086284,
        "long_context_loss": 0.15909902576697318,
        "mtp_acceptance_loss": 0.010182337649086284,
        "output_kl": 0.12727922061357855,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.08273149339882606,
        "token_disagreement": 0.0700035713374682
      },
      "module_path": "model.language_model.layers.46.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.07916767522164586,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.46.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.self_attn.q_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.46.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.46.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.46.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.46.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.47.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.47.input_layernorm.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.021876116042958818,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.03889087296526012,
        "long_context_loss": 0.014584077361972545,
        "mtp_acceptance_loss": 0.003889087296526012,
        "output_kl": 0.04861359120657515,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.03159883428427385,
        "token_disagreement": 0.026737475163616336
      },
      "module_path": "model.language_model.layers.47.mlp.gate",
      "outlier_strategy": "none",
      "parameters": 262144,
      "predicted_loss": 0.02654302079879003,
      "reason": "protected router policy",
      "role": "router",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.47.mlp.gate.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.47.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.47.post_attention_layernorm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.47.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.47.self_attn.k_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.47.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.47.self_attn.k_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.47.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.47.self_attn.o_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.layers.47.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 128,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.47.self_attn.q_norm.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.47.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 8388608,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.47.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.02025,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.036,
        "long_context_loss": 0.056249999999999994,
        "mtp_acceptance_loss": 0.0036,
        "output_kl": 0.045,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.029249999999999998,
        "token_disagreement": 0.02475
      },
      "module_path": "model.language_model.layers.47.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 1048576,
      "predicted_loss": 0.027989999999999998,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 8.5
      },
      "tensor": "model.language_model.layers.47.self_attn.v_proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "lm_head",
      "outlier_strategy": "none",
      "parameters": 311164928,
      "predicted_loss": 0.0,
      "reason": "protected lm_head policy",
      "role": "lm_head",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "lm_head.weight"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.47.mlp.experts.down_proj",
      "outlier_strategy": "none",
      "parameters": 201326592,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.47.mlp.experts.down_proj"
    },
    {
      "bits": 6,
      "group_size": 64,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.04275,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.07600000000000001,
        "long_context_loss": 0.028499999999999998,
        "mtp_acceptance_loss": 0.0076,
        "output_kl": 0.095,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.061750000000000006,
        "token_disagreement": 0.052250000000000005
      },
      "module_path": "model.language_model.layers.47.mlp.experts.gate_up_proj",
      "outlier_strategy": "none",
      "parameters": 402653184,
      "predicted_loss": 0.05187,
      "reason": "selected by marginal quality gain per storage bit",
      "role": "expert",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 7,
        "storage_bpw": 6.5
      },
      "tensor": "model.language_model.layers.47.mlp.experts.gate_up_proj"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.language_model.norm",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "norm",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.norm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.0.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.0.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.1.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.1.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.10.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.10.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.11.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.11.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.12.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.12.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.13.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.13.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.14.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.14.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.15.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.15.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.16.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.16.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.17.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.17.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.18.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.18.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.19.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.19.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.2.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.2.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.20.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.20.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.21.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.21.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.22.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.22.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.23.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.23.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.24.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.24.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.25.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.25.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.26.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.26.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.3.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.3.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.4.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.4.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.5.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.5.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.6.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.6.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.7.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.7.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.8.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.8.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.attn.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.attn.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.attn.proj",
      "outlier_strategy": "none",
      "parameters": 1327104,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.attn.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.attn.qkv.bias",
      "outlier_strategy": "none",
      "parameters": 3456,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.attn.qkv.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.attn.qkv",
      "outlier_strategy": "none",
      "parameters": 3981312,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.attn.qkv.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.mlp.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4304,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.mlp.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.mlp.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.mlp.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.mlp.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.mlp.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.mlp.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 4958208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.mlp.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.norm1.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.norm1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.norm1",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.norm1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.norm2.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.norm2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.blocks.9.norm2",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.blocks.9.norm2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 21233664,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 9437184,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.norm.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.norm.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.0.norm",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.0.norm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 21233664,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 9437184,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.norm.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.norm.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.1.norm",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.1.norm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 21233664,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 9437184,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.norm.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.norm.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.deepstack_merger_list.2.norm",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.deepstack_merger_list.2.norm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.linear_fc1.bias",
      "outlier_strategy": "none",
      "parameters": 4608,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.linear_fc1.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.linear_fc1",
      "outlier_strategy": "none",
      "parameters": 21233664,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.linear_fc1.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.linear_fc2.bias",
      "outlier_strategy": "none",
      "parameters": 2048,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.linear_fc2.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.linear_fc2",
      "outlier_strategy": "none",
      "parameters": 9437184,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.linear_fc2.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.norm.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.norm.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.merger.norm",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.merger.norm.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.patch_embed.proj.bias",
      "outlier_strategy": "none",
      "parameters": 1152,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.patch_embed.proj.bias"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.patch_embed.proj",
      "outlier_strategy": "none",
      "parameters": 1769472,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.patch_embed.proj.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 0.0,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.0,
        "token_disagreement": 0.0
      },
      "module_path": "model.visual.pos_embed",
      "outlier_strategy": "none",
      "parameters": 2654208,
      "predicted_loss": 0.0,
      "reason": "protected vision policy",
      "role": "vision",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.visual.pos_embed.weight"
    }
  ],
  "calibration": null,
  "candidate_bits": [
    4,
    6,
    8,
    16
  ],
  "candidate_group_sizes": [
    32,
    64
  ],
  "constraints": {
    "effective_bpw_limit": 6.0,
    "lm_head_min_bits": 16,
    "max_model_size_ratio_to_uniform4": 1.1,
    "minimum_mtp_acceptance_retention": 0.95,
    "minimum_mtp_speedup": 1.2,
    "minimum_quality_retention": 0.98
  },
  "cost_model": "abstract-bpw",
  "created_at": "2026-08-11T20:23:50.223182Z",
  "effective_bpw": 5.999997506594146,
  "evidence_kind": "architecture_prior",
  "global_validation_required": true,
  "group_size": 32,
  "hardware": {
    "name": "ax-engine-apple-silicon-affine-dwq-v3",
    "runtime": "ax-engine",
    "supported_bits": [
      2,
      3,
      4,
      6,
      8,
      16
    ],
    "supported_group_sizes": [
      32,
      64,
      128
    ],
    "supported_methods": [
      "affine",
      "awq",
      "dwq",
      "gptq",
      "gptq-act",
      "bf16"
    ]
  },
  "kernel_latency_host_id": null,
  "kernel_latency_sha256": null,
  "kv_cache": null,
  "method_near_ties": [],
  "method_near_ties_omitted": 0,
  "mtp": {
    "candidate_bits": [
      8,
      16
    ],
    "min_bits": 8,
    "mode": "protected",
    "optimize_for_acceptance": true,
    "preserve_external_sidecar": true,
    "protect_norms": true,
    "protect_output_head": true
  },
  "mtp_distribution": {},
  "nominal_bpw": 5.233981043154364,
  "objective": {
    "cosine_distance": 0.05,
    "decode_latency_cost": 0.05,
    "hidden_state_error": 0.12,
    "long_context_loss": 0.08,
    "mtp_acceptance_loss": 0.15,
    "output_kl": 0.25,
    "peak_memory_cost": 0.04,
    "prefill_latency_cost": 0.03,
    "task_loss_delta": 0.15,
    "token_disagreement": 0.08
  },
  "primary_runtime": "ax-engine",
  "profile": "general",
  "quantizer": "axquant",
  "random_seed": 0,
  "schema_version": "axquant.plan.v1",
  "software_versions": {
    "ax_engine": null,
    "axquant": "1.6.2",
    "mlx": "0.32.0",
    "mlx_lm": "0.31.3",
    "pydantic": "2.13.4",
    "python": "3.13.14",
    "safetensors": "0.8.0"
  },
  "source_model": {
    "architecture": "Qwen3VLMoeForConditionalGeneration",
    "format": "mlx",
    "local_path": "/Volumes/Ext4T/models/Qwen3-VL-30B-A3B-Instruct-bf16",
    "model_id": "Qwen/Qwen3-VL-30B-A3B-Instruct",
    "revision": "9c4b90e1e4ba969fd3b5378b57d966d725f1b86c"
  },
  "status": "planned",
  "target_bpw": 6.0,
  "target_class": "6bit",
  "target_mode": "low-memory",
  "warnings": [
    "This report contains architecture priors, not calibration measurements.",
    "Conversion planning requires --allow-unmeasured for this report.",
    "Multi-group architecture priors weakly prefer smaller group sizes; this is development evidence only (AXQ-028).",
    "Plan uses non-release architecture_prior evidence and requires complete-model validation.",
    "convert ladder: prior"
  ],
  "weight_distribution": {
    "4bit": {
      "fraction": 0.5313286097594376,
      "parameters": 16508780544
    },
    "6bit": {
      "fraction": 0.42978085907561214,
      "parameters": 13353615360
    },
    "8bit": {
      "fraction": 0.011533381122033016,
      "parameters": 358350848
    },
    "bf16": {
      "fraction": 0.027357150042917246,
      "parameters": 850007280
    }
  }
}