{
  "run_id": "cymraegbench-qwen3-5-9b-local-bf16-v0.1",
  "created_at": "2026-08-10T15:02:55.473302+00:00",
  "status": "completed",
  "seed": 1,
  "max_cases_per_suite": 100,
  "techiaith_revision": "47839d2147c97fd2f10a52dd36751608e5fa36bf",
  "model": {
    "id": "qwen3-5-9b-local-bf16",
    "label": "Qwen3.5 9B",
    "weights": "Qwen/Qwen3.5-9B",
    "weight_access": "open"
  },
  "inference": {
    "engine": "vLLM 0.26.0",
    "precision": "BF16",
    "accelerator": "NVIDIA RTX 4090",
    "max_model_length": 4096,
    "max_parallel_sequences": 4,
    "thinking": false,
    "final_min_output_tokens": 2048,
    "final_repetition_penalty": 1.05
  },
  "status_counts": {
    "completed": 2053,
    "refusal": 0,
    "invalid": 0,
    "error": 0
  },
  "retry_policy": "Rows with status=error were retried in place; successful rows were not rerun. One length-limited CCC response succeeded after raising the output ceiling and adding a light repetition penalty.",
  "formula": {
    "overall_score": "50% macro mean of 9 objective suites + 50% macro mean of 9 Techiaith suites",
    "translation": "Corpus BLEU, reported separately"
  }
}
