{
  "schema_version": 1,
  "adapter_key": "mad-rl-qwen-choice",
  "validated_at": "2026-09-29",
  "accelerator": {
    "vendor": "NVIDIA",
    "model": "A100-SXM4-80GB"
  },
  "command": [
    "python", "scripts/mad_rl_qwen_choice.py",
    "--checkpoint", "<public-Qwen3-4B-Instruct-2507-snapshot>",
    "--output", "<local-result.json>",
    "--seed", "42", "--warmup-steps", "15", "--steps", "8",
    "--group-size", "16", "--eval-samples", "16", "--divergence", "jsd"
  ],
  "dataset": {
    "name": "public synthetic five-choice topic suite",
    "source": "src/auto_research/post_training/mad_rl.py:TOPICS",
    "revision": "4c3051124c219e0ceb0bea7cee63cf123e52bb33",
    "split": "first 8 topics train; next 2 validation; final 2 test"
  },
  "checkpoint": {
    "model_id": "Qwen/Qwen3-4B-Instruct-2507",
    "revision": "cdbee75f17c01a7cc42f958dc650907174af0554",
    "local_weights_committed": false
  },
  "result": "passed",
  "metrics": {
    "seed": 42,
    "warmup_updates": 15,
    "rl_updates": 8,
    "group_size": 16,
    "target_distribution": [0.0, 0.0, 0.3333333333333333, 0.3333333333333333, 0.3333333333333333],
    "baseline_validation_counts": [16, 0, 16, 0, 0],
    "final_validation_counts": [16, 0, 16, 0, 0],
    "baseline_validation_jsd_valid_only": 0.41272565263942185,
    "final_validation_jsd_valid_only": 0.41272565263942185,
    "final_test_jsd_valid_only": 0.6398253243279979,
    "adapter_parameter_delta_l2": 0.09822092745005068,
    "validation_improved": false,
    "formal_paper_result_reproduced": false,
    "diagnostic_only": true,
    "tokenizer_compatibility_recheck": {
      "code_commit": "cdb53fe5a914505f097e0798107de368740a6921",
      "seed": 42,
      "warmup_updates": 2,
      "rl_updates": 2,
      "group_size": 4,
      "validation_jsd_before": 0.41272565263942185,
      "validation_jsd_after": 0.41272565263942185,
      "test_jsd_after": 0.44827065307054925,
      "adapter_parameter_delta_l2": 0.029025848443696434,
      "result": "passed"
    }
  },
  "provenance": {
    "commit": "4c3051124c219e0ceb0bea7cee63cf123e52bb33",
    "artifact_path": "docs/gpu-validations/mad-rl-qwen-choice-a100-20260929.json",
    "raw_predictions_committed": false,
    "checkpoint_committed": false
  }
}
