{
  "schema_version": 1,
  "adapter_key": "checkpoint-normalized-dpo",
  "validated_at": "2026-09-01",
  "accelerator": {"vendor": "NVIDIA", "model": "A100"},
  "command": [
    "auto-research", "checkpoint-post-train", "--objective", "normalized-dpo",
    "--dataset", "ultrafeedback", "--beta", "0.1", "--steps", "2",
    "--model-id", "HuggingFaceTB/SmolLM2-135M-Instruct",
    "--model-revision", "12fd25f77366fa6b3b4b768ec3050bf629380bac",
    "--dataset-revision", "292c16329d921287c4166934cac1a6ad1e13a6c5",
    "--maximum-examples", "8", "--evaluation-examples", "4",
    "--batch-size", "2", "--maximum-length", "192",
    "--mixed-precision", "bf16", "--seeds", "42,43,44"
  ],
  "dataset": {
    "name": "HuggingFaceH4/ultrafeedback_binarized",
    "revision": "292c16329d921287c4166934cac1a6ad1e13a6c5",
    "train_examples": 8,
    "test_examples": 4
  },
  "checkpoint": {
    "model_id": "HuggingFaceTB/SmolLM2-135M-Instruct",
    "revision": "12fd25f77366fa6b3b4b768ec3050bf629380bac"
  },
  "result": "passed",
  "metrics": {
    "seeds": [42, 43, 44],
    "baseline_preference_accuracy_mean": 0.0,
    "final_preference_accuracy_mean": 0.0,
    "baseline_preference_margin_mean": -1.25,
    "final_preference_margin_mean": -1.21875,
    "final_preference_margin_std": 0.03125
  },
  "provenance": {
    "commit": "583e7a439003e6d4581d2415bd678289553b0e7e",
    "artifact_path": "docs/gpu-validations/checkpoint-normalized-dpo-a100-20260901.json",
    "raw_predictions_committed": false,
    "checkpoint_committed": false
  }
}
