evaluation/benchmark_provenance.json

{
  "filename": "humaneval-test.parquet",
  "sha256": "3bea46a554c67c5d2915e2db7d97d7733f09140e7a1776428c1b20b5ae5d124b",
  "task": "magicoder",
  "evaluation_spec": {
    "benchmark": "HumanEval",
    "dataset": "<LOCAL_PATH>/humaneval-test.parquet",
    "samples": 164,
    "primary_metric": "pass@1",
    "max_new_tokens": 512
  },
  "provenance_note": "GSM8K/IFEval local inputs were reconstructed from historical scored outputs; HumanEval is local test parquet."
}