eval_results.json

[
  {
    "task": "hellaswag",
    "benchmark": "HellaSwag",
    "metric": "acc_norm",
    "score": 0.2894841665006971,
    "shots": 0,
    "runtime_sec": 217.62,
    "status": "success"
  },
  {
    "task": "piqa",
    "benchmark": "PIQA",
    "metric": "acc_norm",
    "score": 0.5854189336235038,
    "shots": 0,
    "runtime_sec": 34.14,
    "status": "success"
  },
  {
    "task": "winogrande",
    "benchmark": "WinoGrande",
    "metric": "acc",
    "score": 0.5153906866614049,
    "shots": 0,
    "runtime_sec": 22.26,
    "status": "success"
  },
  {
    "task": "boolq",
    "benchmark": "BoolQ",
    "metric": "acc",
    "score": 0.618960244648318,
    "shots": 0,
    "runtime_sec": 53.58,
    "status": "success"
  },
  {
    "task": "arc_easy",
    "benchmark": "ARC-Easy",
    "metric": "acc_norm",
    "score": 0.3661616161616162,
    "shots": 0,
    "runtime_sec": 42.41,
    "status": "success"
  },
  {
    "task": "arc_challenge",
    "benchmark": "ARC-Challenge",
    "metric": "acc_norm",
    "score": 0.2431740614334471,
    "shots": 0,
    "runtime_sec": 31.16,
    "status": "success"
  },
  {
    "task": "openbookqa",
    "benchmark": "OpenBookQA",
    "metric": "acc_norm",
    "score": 0.294,
    "shots": 0,
    "runtime_sec": 24.04,
    "status": "success"
  },
  {
    "task": "commonsense_qa",
    "benchmark": "CommonsenseQA",
    "metric": "acc",
    "score": 0.19737919737919737,
    "shots": 0,
    "runtime_sec": 27.13,
    "status": "success"
  },
  {
    "task": "lambada_openai",
    "benchmark": "LAMBADA",
    "metric": "acc",
    "score": 0.22026004269357655,
    "shots": 0,
    "runtime_sec": 74.36,
    "status": "success"
  },
  {
    "task": "blimp",
    "benchmark": "BLiMP",
    "metric": "acc",
    "score": 0.7671641791044775,
    "shots": 0,
    "runtime_sec": 361.7,
    "status": "success"
  },
  {
    "task": "mmlu",
    "benchmark": "MMLU",
    "metric": "acc",
    "score": 0.2324455205811138,
    "shots": 0,
    "runtime_sec": 292.98,
    "status": "success"
  },
  {
    "task": "wikitext",
    "benchmark": "WikiText-2",
    "metric": "word_perplexity",
    "score": 46.05417779102866,
    "shots": 0,
    "runtime_sec": 32.41,
    "status": "success"
  },
  {
    "task": "wikitext",
    "benchmark": "WikiText-2",
    "metric": "byte_perplexity",
    "score": 2.0466313554082944,
    "shots": 0,
    "runtime_sec": 29.23,
    "status": "success"
  },
  {
    "task": "sciq",
    "benchmark": "SciQ",
    "metric": "acc_norm",
    "score": 0.559,
    "shots": 0,
    "runtime_sec": 43.24,
    "status": "success"
  },
  {
    "task": "copa",
    "benchmark": "COPA",
    "metric": "acc",
    "score": 0.59,
    "shots": 0,
    "runtime_sec": 19.09,
    "status": "success"
  },
  {
    "task": "race",
    "benchmark": "RACE",
    "metric": "acc",
    "score": 0.27177033492822966,
    "shots": 0,
    "runtime_sec": 94.94,
    "status": "success"
  },
  {
    "task": "swag",
    "benchmark": "SWAG",
    "metric": "acc_norm",
    "score": 0.4432170348895331,
    "shots": 0,
    "runtime_sec": 168.16,
    "status": "success"
  },
  {
    "task": "truthfulqa_mc2",
    "benchmark": "TruthfulQA MC2",
    "metric": "acc",
    "score": 0.44275007868091637,
    "shots": 0,
    "runtime_sec": 62.13,
    "status": "success"
  }
]