{
  "analysis_sha256": "459134574eb7d7386ec6b79c66edf86f4a5af51f2e48829fe5c7e22565434059",
  "architecture_profile": {
    "adapter_id": "qwen38-dense-v1",
    "audio_present": false,
    "config_model_type": "qwen3_5",
    "dense": true,
    "mtp_declared": false,
    "notes": [
      "Qwen 3.8 dense (e.g. Qwen3.8-27B) shares Qwen 3.5/3.6 hybrid dense tensor conventions; language path quantizes, vision stays BF16-protected.",
      "AXQ 4/6-bit ± MTP packs are checkpoint Tier 1 certified; MTP packs also have scoped MTP acceleration certification (MTP-P evidence) on AX Engine 6.16.1. Not the Qwen 3.6 campaign track and not the Super-class 2.4T MoE stream path.",
      "Cloudflare/clef is eligible as a Qwen3.8-27B decision-model fine-tune (no MTP tensors; vision stays BF16-protected)."
    ],
    "optimization_scope": "text-path",
    "product_family": "qwen3.8",
    "support_level": "supported",
    "support_tier": "convertible",
    "text_layer_count": 64,
    "vision_present": true
  },
  "assignments": [
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_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": 1271398400,
      "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": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.017898640398784485,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.031819805153394644,
        "long_context_loss": 0.01193242693252299,
        "mtp_acceptance_loss": 0.0031819805153394643,
        "output_kl": 0.0397747564417433,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.025853591687133148,
        "token_disagreement": 0.021876116042958818
      },
      "module_path": "model.language_model.embed_tokens",
      "outlier_strategy": "none",
      "parameters": 1271398400,
      "predicted_loss": 0.021717017017191843,
      "reason": "protected embedding policy",
      "role": "embedding",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 2,
        "storage_bpw": 9.0
      },
      "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": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.0.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.in_proj_a.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.0.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.in_proj_b.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.0.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.0.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.0.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.0.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.0.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.1.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.1.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.in_proj_a.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.1.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.1.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.1.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.1.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.1.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.2.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.2.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.2.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.2.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.2.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.3.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.3.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.3.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.3.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.3.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.4.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.4.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.4.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.4.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.4.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.5.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.5.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.5.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.5.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.5.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.6.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.6.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.6.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.6.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.6.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.7.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.7.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.7.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.7.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.7.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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.8.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.8.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.8.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.8.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.8.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.9.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.9.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.mlp.down_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": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.10.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.in_proj_a.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.10.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.10.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.10.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.10.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.10.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.11.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.11.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.11.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.11.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.11.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.11.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.11.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.11.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.12.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.12.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.in_proj_a.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.12.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.12.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.12.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.12.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.12.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.13.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.13.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.in_proj_a.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.13.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.13.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.13.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.13.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.13.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.14.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.dt_bias"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.14.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.14.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.14.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.14.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.14.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.15.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.15.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.15.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.15.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.15.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.15.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.15.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.15.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.16.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.in_proj_b.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.9.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.9.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.9.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.16.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.16.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.16.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.16.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.17.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.17.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.17.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.17.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.17.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.18.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.18.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.18.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.18.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.18.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.19.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.19.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.19.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.19.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.19.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.19.self_attn.q_proj.weight"
    },
    {
      "bits": 8,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.014318912319027589,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.02545584412271571,
        "long_context_loss": 0.039774756441743296,
        "mtp_acceptance_loss": 0.002545584412271571,
        "output_kl": 0.03181980515339464,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.020682873349706515,
        "token_disagreement": 0.01750089283436705
      },
      "module_path": "model.language_model.layers.19.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.019791918805411465,
      "reason": "fused-module unit upgraded by measured marginal quality per storage bit",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 9.0
      },
      "tensor": "model.language_model.layers.19.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.20.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.20.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.20.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.20.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.20.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.21.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.21.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.21.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.21.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.21.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.22.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.22.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.22.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.22.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.22.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.23.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.23.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.23.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.23.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.23.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.24.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.24.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.24.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.24.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.24.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.25.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.25.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.25.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.25.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.25.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.26.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.26.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.26.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.26.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.26.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.27.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.27.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.27.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.27.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.27.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.28.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.28.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.28.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.28.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.28.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.29.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.29.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.29.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.29.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.29.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.30.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.30.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.30.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.30.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.30.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.31.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.31.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.31.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.31.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.31.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.32.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.32.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.32.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.32.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.32.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.33.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.33.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.33.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.33.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.33.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.34.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.34.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.34.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.34.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.34.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.35.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.35.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.35.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.35.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.35.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.36.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.36.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.36.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.36.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.36.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.37.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.37.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.37.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.37.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.37.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.38.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.38.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.38.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.38.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.38.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.39.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.39.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.39.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.39.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.39.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.40.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.40.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.40.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.40.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.40.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.41.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.41.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.41.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.41.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.41.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.42.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.42.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.42.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.42.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.42.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.43.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.43.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.43.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.43.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "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": 256,
      "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": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.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.43.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.43.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.44.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.44.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.44.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.44.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.44.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.45.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.45.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.45.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.45.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.45.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.46.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.dt_bias"
    },
    {
      "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.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.in_proj_a.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.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.in_proj_b.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.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.in_proj_qkv.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.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.linear_attn.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.linear_attn.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.46.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.46.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.46.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.46.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.46.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.47.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.47.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.47.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.47.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.47.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.47.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.47.mlp.up_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.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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": 256,
      "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": 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.47.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.47.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.47.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "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": 256,
      "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": 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.47.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.47.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.47.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "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": "model.language_model.layers.48.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.48.input_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.48.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.48.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.48.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.dt_bias"
    },
    {
      "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.48.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.in_proj_a.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.48.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.in_proj_b.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.48.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.in_proj_qkv.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.48.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.48.linear_attn.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.48.linear_attn.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.48.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.48.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.48.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.48.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.48.mlp.up_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.48.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.48.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.49.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.49.input_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.49.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.49.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.49.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.dt_bias"
    },
    {
      "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.49.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.in_proj_a.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.49.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.in_proj_b.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.49.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.in_proj_qkv.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.49.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.49.linear_attn.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.49.linear_attn.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.49.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.49.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.49.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.49.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.49.mlp.up_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.49.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.49.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.50.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.50.input_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.50.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.50.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.50.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.dt_bias"
    },
    {
      "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.50.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.in_proj_a.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.50.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.in_proj_b.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.50.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.in_proj_qkv.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.50.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.50.linear_attn.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.50.linear_attn.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.50.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.50.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.50.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.50.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.50.mlp.up_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.50.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.50.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.51.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.51.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.51.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.51.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.51.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.mlp.up_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.51.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.51.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.51.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.51.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.51.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.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.51.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.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.51.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.51.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.51.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.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.51.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.51.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.52.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.52.input_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.52.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.52.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.52.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.dt_bias"
    },
    {
      "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.52.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.in_proj_a.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.52.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.in_proj_b.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.52.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.in_proj_qkv.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.52.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.52.linear_attn.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.52.linear_attn.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.52.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.52.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.52.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.52.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.52.mlp.up_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.52.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.52.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.53.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.53.input_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.53.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.53.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.53.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.dt_bias"
    },
    {
      "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.53.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.in_proj_a.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.53.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.in_proj_b.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.53.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.in_proj_qkv.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.53.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.53.linear_attn.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.53.linear_attn.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.53.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.53.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.53.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.53.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.53.mlp.up_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.53.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.53.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.54.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.54.input_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.54.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.54.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.54.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.dt_bias"
    },
    {
      "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.54.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.in_proj_a.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.54.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.in_proj_b.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.54.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.in_proj_qkv.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.54.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.54.linear_attn.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.54.linear_attn.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.54.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.54.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.54.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.54.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.54.mlp.up_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.54.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.54.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.55.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.55.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.55.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.55.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.55.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.mlp.up_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.55.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.55.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.55.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.55.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.55.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.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.55.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.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.55.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.55.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.55.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.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.55.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.55.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.56.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.56.input_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.56.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.56.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.56.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.dt_bias"
    },
    {
      "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.56.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.in_proj_a.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.56.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.in_proj_b.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.56.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.in_proj_qkv.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.56.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.56.linear_attn.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.56.linear_attn.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.56.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.56.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.56.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.56.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.56.mlp.up_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.56.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.56.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.57.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.57.input_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.57.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.57.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.57.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.dt_bias"
    },
    {
      "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.57.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.in_proj_a.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.57.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.in_proj_b.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.57.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.in_proj_qkv.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.57.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.57.linear_attn.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.57.linear_attn.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.57.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.57.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.57.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.57.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.57.mlp.up_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.57.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.57.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.58.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.58.input_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.58.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.58.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.58.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.dt_bias"
    },
    {
      "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.58.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.in_proj_a.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.58.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.in_proj_b.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.58.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.in_proj_qkv.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.58.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.58.linear_attn.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.58.linear_attn.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.58.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.58.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.58.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.58.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.58.mlp.up_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.58.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.58.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.59.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.59.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.59.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.59.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.59.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.mlp.up_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.59.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.59.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.59.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.59.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.59.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.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.59.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.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.59.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.59.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.59.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.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.59.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.59.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.60.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.60.input_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.60.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.60.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.60.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.dt_bias"
    },
    {
      "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.60.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.in_proj_a.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.60.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.in_proj_b.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.60.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.in_proj_qkv.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.60.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.60.linear_attn.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.60.linear_attn.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.60.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.60.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.60.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.60.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.60.mlp.up_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.60.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.60.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.61.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.61.input_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.61.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.61.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.61.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.dt_bias"
    },
    {
      "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.61.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.in_proj_a.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.61.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.in_proj_b.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.61.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.in_proj_qkv.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.61.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.61.linear_attn.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.61.linear_attn.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.61.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.61.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.61.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.61.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.61.mlp.up_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.61.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.61.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.62.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.62.input_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.62.linear_attn.A_log",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.A_log"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.62.linear_attn.conv1d",
      "outlier_strategy": "none",
      "parameters": 40960,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.conv1d.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.62.linear_attn.dt_bias",
      "outlier_strategy": "none",
      "parameters": 48,
      "predicted_loss": 0.0,
      "reason": "non-quantizable tensor preserved",
      "role": "attention",
      "scale_strategy": "none",
      "strategy_metadata": {
        "selected_from_candidates": 1,
        "storage_bpw": 16.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.dt_bias"
    },
    {
      "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.62.linear_attn.in_proj_a",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.in_proj_a.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.62.linear_attn.in_proj_b",
      "outlier_strategy": "none",
      "parameters": 245760,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.in_proj_b.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.62.linear_attn.in_proj_qkv",
      "outlier_strategy": "none",
      "parameters": 52428800,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.in_proj_qkv.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.62.linear_attn.in_proj_z",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.in_proj_z.weight"
    },
    {
      "bits": 16,
      "group_size": null,
      "method": "bf16",
      "metrics": {
        "cosine_distance": 0.0,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.0,
        "long_context_loss": 0.0,
        "mtp_acceptance_loss": 0.0,
        "output_kl": 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.62.linear_attn.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.62.linear_attn.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.62.linear_attn.out_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.linear_attn.out_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.62.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.62.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.62.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.62.mlp.up_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.62.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.62.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.63.input_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.63.input_layernorm.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.63.mlp.down_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.mlp.down_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.63.mlp.gate_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.mlp.gate_proj.weight"
    },
    {
      "bits": 6,
      "group_size": 32,
      "method": "affine",
      "metrics": {
        "cosine_distance": 0.03579728079756897,
        "decode_latency_cost": 0.0,
        "hidden_state_error": 0.06363961030678929,
        "long_context_loss": 0.02386485386504598,
        "mtp_acceptance_loss": 0.006363961030678929,
        "output_kl": 0.0795495128834866,
        "peak_memory_cost": 0.0,
        "prefill_latency_cost": 0.0,
        "task_loss_delta": 0.051707183374266295,
        "token_disagreement": 0.043752232085917636
      },
      "module_path": "model.language_model.layers.63.mlp.up_proj",
      "outlier_strategy": "none",
      "parameters": 89128960,
      "predicted_loss": 0.04343403403438369,
      "reason": "minimum common fused-module signature",
      "role": "mlp",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.mlp.up_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.63.post_attention_layernorm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.63.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.63.self_attn.k_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.63.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.63.self_attn.k_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.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.63.self_attn.o_proj",
      "outlier_strategy": "none",
      "parameters": 31457280,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.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.63.self_attn.q_norm",
      "outlier_strategy": "none",
      "parameters": 256,
      "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.63.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.63.self_attn.q_proj",
      "outlier_strategy": "none",
      "parameters": 62914560,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.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.63.self_attn.v_proj",
      "outlier_strategy": "none",
      "parameters": 5242880,
      "predicted_loss": 0.07916767522164586,
      "reason": "minimum common fused-module signature",
      "role": "attention",
      "scale_strategy": "group-affine",
      "strategy_metadata": {
        "selected_from_candidates": 3,
        "storage_bpw": 7.0
      },
      "tensor": "model.language_model.layers.63.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.norm",
      "outlier_strategy": "none",
      "parameters": 5120,
      "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.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": 5120,
      "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": 23592960,
      "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": [
    6,
    8,
    16
  ],
  "candidate_group_sizes": [
    32
  ],
  "constraints": {
    "effective_bpw_limit": 7.67,
    "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": "2026-10-24T06:07:55.076428Z",
  "effective_bpw": 7.66999757079141,
  "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",
      "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": 6.733410635558831,
  "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.v2",
  "software_versions": {
    "ax_engine": "7.9.0",
    "axquant": "2.3.0",
    "mlx": "0.32.4",
    "mlx_lm": "0.32.2",
    "pydantic": "2.13.4",
    "python": "3.12.13",
    "safetensors": "0.8.0"
  },
  "source_model": {
    "architecture": "Qwen3_5ForConditionalGeneration",
    "format": "mlx",
    "local_path": null,
    "model_id": "Cloudflare/clef",
    "revision": "0b331204bb13fbd2ca93a64df1956f5b55478ce5"
  },
  "status": "planned",
  "target_bpw": 7.67,
  "target_class": "8bit",
  "target_mode": "low-memory",
  "warnings": [
    "This report contains architecture priors, not calibration measurements.",
    "Conversion planning requires --allow-unmeasured for this report.",
    "Plan uses non-release architecture_prior evidence and requires complete-model validation.",
    "Allocation units are fused switch modules (and singleton tensors). This is the experimental trunk-mix path, not the default tensor-then-harmonize planner.",
    "target BPW raised from 6.8000 to 7.6700 to satisfy protection floors (policy minimum 7.6637)",
    "convert ladder: prior"
  ],
  "weight_distribution": {
    "6bit": {
      "fraction": 0.8869469412902637,
      "parameters": 24263966720
    },
    "8bit": {
      "fraction": 0.049639993942316614,
      "parameters": 1357987840
    },
    "bf16": {
      "fraction": 0.06341306476741969,
      "parameters": 1734774000
    }
  }
}