{
  "generated_at": "2026-08-03T21:27:04",
  "seed": 42,
  "source": "https://github.com/deepseek-ai/DeepSpec/tree/main/eval_datasets",
  "input_format": "{\"turns\": [\"<prompt>\", ...]} por línea",
  "multi_turn": "turns unidas con \\n (mt-bench/arena-hard-v2)",
  "sampling": "shuffle con RNG(seed) + recorte; None = dataset completo",
  "total_prompts": 1474,
  "per_dataset": {
    "gsm8k": {
      "domain": "math",
      "available": 1319,
      "sampled": 200
    },
    "math500": {
      "domain": "math",
      "available": 500,
      "sampled": 200
    },
    "aime25": {
      "domain": "math",
      "available": 30,
      "sampled": 30
    },
    "humaneval": {
      "domain": "code",
      "available": 164,
      "sampled": 164
    },
    "mbpp": {
      "domain": "code",
      "available": 257,
      "sampled": 200
    },
    "livecodebench": {
      "domain": "code",
      "available": 1055,
      "sampled": 200
    },
    "mt-bench": {
      "domain": "chat",
      "available": 80,
      "sampled": 80
    },
    "alpaca": {
      "domain": "chat",
      "available": 52002,
      "sampled": 200
    },
    "arena-hard-v2": {
      "domain": "chat",
      "available": 750,
      "sampled": 200
    }
  },
  "output": "experiments/prompts/f1-sample.jsonl"
}