{
  "adapter_checkpoint": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/idr1_55b6476_r1/train/checkpoint_0000256.pt",
  "adapter_checkpoint_sha256": "df3c264d426941fef8ba9c10a90fe9fab304ec2864738209a4d79f9f81e0c473",
  "adapter_metadata": {
    "arm": "baseline",
    "backbone_layout": "multimodal-language-model",
    "data": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/idr1_data_ad790b0_r1/train.jsonl",
    "data_seed": 2026080814,
    "data_sha256": "6df3204573ce807db1b5057bce709189366b6674e38e5224ee3d17a3e6f0ac6c",
    "loaded_model_root": "/tmp/sa305415/745652/model",
    "lora_alpha": 16.0,
    "lora_layers": 4,
    "lora_projection_count": 31,
    "lora_rank": 8,
    "model_loader": "multimodal",
    "model_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
    "model_root": "/lustre/fs1/home/sa305415/shohin/artifacts/external/qwen3.5-9b-c202236",
    "seed": 2026080815,
    "selected_rows": 9655,
    "trainable_parameters": 2704896,
    "unfreeze_layers": 0,
    "update": 256,
    "warm_start_checkpoint": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/qwen3.5-9b-c202236/b1_u256_dddb36f/checkpoint_0000256.pt",
    "warm_start_data_sha256": "2461d6f70b44a142854d56c24e1fb42d600065e5788a2c4e055ba47b12696549",
    "warm_start_sha256": "854a7cc44fbc2b54418f4e5bd09b7efeed0da44fc9ce217b0bb6b1997b722971",
    "warm_start_update": 256,
    "workspace_architecture_sha256": null,
    "workspace_config": null
  },
  "assessor_fields_visible_to_model": false,
  "batch_size": 2,
  "candidates_output": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/idr1_55b6476_r1/eval/holdout/candidates.jsonl",
  "candidates_sha256": "152a93003a5c155e41300b6956ae3887be68ca23dd82e038a1f89fe9d39495cc",
  "data": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/idr1_data_ad790b0_r1/holdout.jsonl",
  "data_report": "/lustre/fs1/home/sa305415/shohin/artifacts/product_reasoning/idr1_data_ad790b0_r1/report.json",
  "data_report_sha256": "8e4b2817d2c96036ff8c2d1606b7a9a37e08fc7176ae7ffe42344041b03ac80e",
  "data_sha256": "df347d5990748cdca73c3a546e46e1eda861043d90347e5765549ffc3fe836f2",
  "elapsed_seconds": 989.8435881081969,
  "external_candidate_text_visible": false,
  "frozen_floors": {
    "bbh_logic": 327,
    "both_wrong": 100,
    "math500": 255,
    "mbpp": 24,
    "overall": 618
  },
  "gate": {
    "code_within_one_answer_of_vcr1": true,
    "math_within_frozen_floor": false,
    "science_within_frozen_floor": true,
    "solves_frozen_both_wrong_floor": true,
    "within_0_02_of_vcr1_overall": true
  },
  "gate_pass": false,
  "generated_tokens": 24165,
  "generated_tokens_per_second": 24.412947954923354,
  "generation_mode": "greedy",
  "internal_draft_visible": true,
  "loaded_model_root": "/tmp/sa305415/745655/model",
  "max_new_tokens": 768,
  "max_token_exhausted": 3,
  "metrics": {
    "bbh_logic": {
      "base_accuracy": 0.3712,
      "base_correct": 232,
      "both_wrong": 276,
      "both_wrong_repair_rate": 0.2536231884057971,
      "both_wrong_repaired": 70,
      "expert_accuracy": 0.4912,
      "expert_correct": 307,
      "generated_accuracy": 0.5616,
      "generated_correct": 351,
      "oracle_accuracy": 0.5584,
      "oracle_correct": 349,
      "total": 625
    },
    "math500": {
      "base_accuracy": 0.14492753623188406,
      "base_correct": 90,
      "both_wrong": 444,
      "both_wrong_repair_rate": 0.25225225225225223,
      "both_wrong_repaired": 112,
      "expert_accuracy": 0.23027375201288244,
      "expert_correct": 143,
      "generated_accuracy": 0.3993558776167472,
      "generated_correct": 248,
      "oracle_accuracy": 0.28502415458937197,
      "oracle_correct": 177,
      "total": 621
    },
    "mbpp": {
      "base_accuracy": 0.6666666666666666,
      "base_correct": 22,
      "both_wrong": 7,
      "both_wrong_repair_rate": 0.14285714285714285,
      "both_wrong_repaired": 1,
      "expert_accuracy": 0.6363636363636364,
      "expert_correct": 21,
      "generated_accuracy": 0.7878787878787878,
      "generated_correct": 26,
      "oracle_accuracy": 0.7878787878787878,
      "oracle_correct": 26,
      "total": 33
    },
    "overall": {
      "base_accuracy": 0.2689601250977326,
      "base_correct": 344,
      "both_wrong": 727,
      "both_wrong_repair_rate": 0.2517193947730399,
      "both_wrong_repaired": 183,
      "expert_accuracy": 0.36825645035183735,
      "expert_correct": 471,
      "generated_accuracy": 0.48866301798279904,
      "generated_correct": 625,
      "oracle_accuracy": 0.43158717748240816,
      "oracle_correct": 552,
      "total": 1279
    }
  },
  "model_loader": "multimodal",
  "model_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
  "model_root": "/lustre/fs1/home/sa305415/shohin/artifacts/external/qwen3.5-9b-c202236",
  "peak_gpu_memory_bytes": 20176390144,
  "runtime_fields": [
    "question"
  ],
  "schema": "shohin-idr1-revision-evaluation-v1",
  "seed": 2026080816,
  "split": "holdout",
  "status": "complete"
}