{
  "schema_version": 2,
  "purpose": "engineering smoke; not an official ScienceQA benchmark",
  "date": "2026-08-11",
  "hardware": "NVIDIA A30 24GB",
  "cuda": "12.8",
  "model": "HuggingFaceTB/SmolVLM2-256M-Video-Instruct",
  "revision": "067788b187b95ebe7b2e040b3e4299e342e5b8fd",
  "parameters": 256484928,
  "weights_updated": false,
  "seed": 42,
  "examples_per_split": 2,
  "data_note": "Two genuine ScienceQA image-text examples arranged as a temporary validation/test engineering fixture; not the official benchmark split.",
  "champion_id": "g0-t0",
  "trials": [
    {
      "trial_id": "g0-t0",
      "generation": 0,
      "prompt_style": "direct",
      "use_hint": true,
      "image_size": "native",
      "max_new_tokens": 16,
      "accuracy": 0.5,
      "parse_rate": 1.0,
      "latency_seconds_per_example": 0.5385349411517382,
      "peak_gpu_memory_mb": 852.56982421875
    },
    {
      "trial_id": "g1-t1",
      "generation": 1,
      "prompt_style": "context-first",
      "use_hint": true,
      "image_size": "native",
      "max_new_tokens": 16,
      "accuracy": 0.5,
      "parse_rate": 1.0,
      "latency_seconds_per_example": 0.2657866347581148,
      "peak_gpu_memory_mb": 853.50830078125
    },
    {
      "trial_id": "g1-t2",
      "generation": 1,
      "prompt_style": "elimination",
      "use_hint": true,
      "image_size": "native",
      "max_new_tokens": 16,
      "accuracy": 0.5,
      "parse_rate": 1.0,
      "latency_seconds_per_example": 0.3302196003496647,
      "peak_gpu_memory_mb": 853.50830078125
    },
    {
      "trial_id": "g2-t1",
      "generation": 2,
      "prompt_style": "direct",
      "use_hint": true,
      "image_size": 384,
      "max_new_tokens": 16,
      "accuracy": 0.5,
      "parse_rate": 1.0,
      "latency_seconds_per_example": 0.265604592859745,
      "peak_gpu_memory_mb": 853.06982421875
    },
    {
      "trial_id": "g2-t2",
      "generation": 2,
      "prompt_style": "direct",
      "use_hint": true,
      "image_size": "native",
      "max_new_tokens": 4,
      "accuracy": 0.5,
      "parse_rate": 1.0,
      "latency_seconds_per_example": 0.2658186387270689,
      "peak_gpu_memory_mb": 853.06982421875
    }
  ],
  "isolated_test": {
    "baseline_accuracy": 0.5,
    "champion_accuracy": 0.5,
    "parse_rate": 1.0,
    "latency_seconds_per_example": 0.3255016915500164,
    "peak_gpu_memory_mb": 853.06982421875
  },
  "manifest_ref": "experiments:a30-vlm-checkpoint-smoke-20260811",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/experiments/a30-vlm-checkpoint-smoke-20260811.json",
    "historical_migration": "historical-metrics-v2-2026-08-09",
    "original_code_commit": "not recorded",
    "dataset_fingerprint": "not recorded in historical artifact"
  }
}