{
  "schema_version": 1,
  "adapter_key": "grpo-dropout",
  "validated_at": "2026-10-10",
  "accelerator": {
    "vendor": "NVIDIA",
    "model": "A100-SXM4-80GB"
  },
  "command": [
    "python",
    "scripts/run_oct10_post_training.py",
    "--objective",
    "grpo-dropout",
    "--student-path",
    "checkpoints/student",
    "--student-id",
    "Qwen/Qwen3-4B-Instruct-2507",
    "--student-revision",
    "cdbee75f17c01a7cc42f958dc650907174af0554",
    "--teacher-path",
    "checkpoints/teacher",
    "--teacher-id",
    "Siye01/Qwen3-4B-Inst-MOPD",
    "--teacher-revision",
    "b61a9f00b77302b6e947af1215b7cb994b884c4a",
    "--train-path",
    "data/gsm8k/train.jsonl",
    "--validation-path",
    "data/gsm8k/test.jsonl",
    "--dataset-revision",
    "3101c7d5072418e28b9008a6636bde82a006892c",
    "--output-dir",
    "runs/oct10/current/grpo-dropout-42",
    "--seed",
    "42",
    "--steps",
    "10",
    "--max-tokens",
    "512",
    "--group",
    "4",
    "--validation-examples",
    "2"
  ],
  "dataset": {
    "id": "openai/grade-school-math",
    "revision": "3101c7d5072418e28b9008a6636bde82a006892c"
  },
  "checkpoint": {
    "student": "Qwen/Qwen3-4B-Instruct-2507",
    "student_revision": "cdbee75f17c01a7cc42f958dc650907174af0554",
    "teacher": "Siye01/Qwen3-4B-Inst-MOPD",
    "teacher_revision": "b61a9f00b77302b6e947af1215b7cb994b884c4a"
  },
  "result": "passed",
  "metrics": {
    "adapter_parameter_delta_l2": 0.028784091661948032,
    "teacher_adapter_parameter_delta_l2": 0,
    "steps_with_nonzero_student_gradient": 1,
    "validation_examples": 2,
    "validation_exact_match": 1,
    "max_training_reward_variance": 0.1875,
    "min_retained_rollouts": 4
  },
  "evaluation_scope": "short real-checkpoint runtime validation; not paper benchmark reproduction",
  "diagnostic_only": true,
  "provenance": {
    "commit": "ea3652c0",
    "artifact_path": "docs/gpu-validations/grpo-dropout-a100-20261010.json",
    "metrics_artifact": "docs/post-training/2610.11854-grpo-dropout/metrics/checkpoint-a100-seed42.json"
  },
  "command_path_policy": "Original checkpoint locations sanitized to checkpoints/student and checkpoints/teacher; bind the public immutable revisions before replay.",
  "additional_runs": [
    {
      "seed": 42,
      "command": [
        "python",
        "scripts/run_oct10_post_training.py",
        "--objective",
        "grpo-dropout",
        "--student-path",
        "checkpoints/student",
        "--student-id",
        "Qwen/Qwen3-4B-Instruct-2507",
        "--student-revision",
        "cdbee75f17c01a7cc42f958dc650907174af0554",
        "--teacher-path",
        "checkpoints/teacher",
        "--teacher-id",
        "Siye01/Qwen3-4B-Inst-MOPD",
        "--teacher-revision",
        "b61a9f00b77302b6e947af1215b7cb994b884c4a",
        "--train-path",
        "data/gsm8k/train.jsonl",
        "--validation-path",
        "data/gsm8k/test.jsonl",
        "--dataset-revision",
        "3101c7d5072418e28b9008a6636bde82a006892c",
        "--output-dir",
        "runs/oct10/current/grpo-dropout-42-12",
        "--seed",
        "42",
        "--steps",
        "12",
        "--max-tokens",
        "512",
        "--group",
        "4",
        "--validation-examples",
        "2"
      ],
      "metrics": {
        "adapter_parameter_delta_l2": 0.03463352418412891,
        "teacher_adapter_parameter_delta_l2": 0,
        "steps_with_nonzero_student_gradient": 1,
        "validation_examples": 2,
        "validation_exact_match": 1,
        "max_training_reward_variance": 0.1875,
        "min_retained_rollouts": 4,
        "steps_with_filtering": 0
      },
      "artifact_path": "docs/post-training/2610.11854-grpo-dropout/metrics/checkpoint-a100-seed42-12.json",
      "commit": "ea3652c0",
      "diagnostic_only": true
    },
    {
      "seed": 43,
      "command": [
        "python",
        "scripts/run_oct10_post_training.py",
        "--objective",
        "grpo-dropout",
        "--student-path",
        "checkpoints/student",
        "--student-id",
        "Qwen/Qwen3-4B-Instruct-2507",
        "--student-revision",
        "cdbee75f17c01a7cc42f958dc650907174af0554",
        "--teacher-path",
        "checkpoints/teacher",
        "--teacher-id",
        "Siye01/Qwen3-4B-Inst-MOPD",
        "--teacher-revision",
        "b61a9f00b77302b6e947af1215b7cb994b884c4a",
        "--train-path",
        "data/gsm8k/train.jsonl",
        "--validation-path",
        "data/gsm8k/test.jsonl",
        "--dataset-revision",
        "3101c7d5072418e28b9008a6636bde82a006892c",
        "--output-dir",
        "runs/oct10/current/grpo-dropout-43-12",
        "--seed",
        "43",
        "--steps",
        "12",
        "--max-tokens",
        "512",
        "--group",
        "4",
        "--validation-examples",
        "2"
      ],
      "metrics": {
        "adapter_parameter_delta_l2": 0.000019026863524479133,
        "teacher_adapter_parameter_delta_l2": 0,
        "steps_with_nonzero_student_gradient": 0,
        "validation_examples": 2,
        "validation_exact_match": 1,
        "max_training_reward_variance": 0.1875,
        "min_retained_rollouts": 1,
        "steps_with_filtering": 1
      },
      "artifact_path": "docs/post-training/2610.11854-grpo-dropout/metrics/checkpoint-a100-seed43-12.json",
      "commit": "e1585f77",
      "diagnostic_only": true
    }
  ]
}