{
  "benchmark": "OpenAI HumanEval",
  "benchmark_sha256": "b796127e635a67f93fb35c04f4cb03cf06f38c8072ee7cee8833d7bee06979ef",
  "daily_assumption": "24 h continuous generation at aggregate observed decode rate; excludes prompts, tests, idle time, electricity and wear",
  "daily_tokens_continuous_24h": 1344990,
  "date_utc": "2026-10-10T12:15:23.576314+00:00",
  "eval_seconds": 245.1338,
  "host": {
    "cpu": "i386",
    "machine": "x86_64",
    "ollama_version": "0.34.0",
    "python": "3.9.10",
    "ram_bytes": 34359738368
  },
  "metadata_note": "After measurement, only the protocol.sandbox description was corrected to identify process-fork denial and the subsequent sanitized replay; measured numeric fields were unchanged.",
  "model": "qwen2.5-coder:1.5b",
  "model_digest": "sha256:29d8c98fa6b098e200069bfb88b9508dc3e85586d20cba59f8dda9a808165104",
  "output_tokens": 3816,
  "pass_rate": 0.7,
  "passed": 35,
  "peak_ollama_rss_bytes": 1886416896,
  "population_n": 164,
  "protocol": {
    "num_ctx": 2048,
    "num_predict": 256,
    "sandbox": "macOS sandbox-exec denied network, file writes and process fork; randomized completion marker; CPU 5s, wall 8s. Post-run replay additionally sanitized the child environment.",
    "temperature": 0,
    "test_timeout_s": 8
  },
  "quantization": "Ollama Q4_K_M registry artifact; model blob SHA-256 pinned",
  "run_id": "29-free-tokens-20261010-02",
  "sample_n": 50,
  "seed": 29029,
  "tokens_per_second": 15.567
}