eval_results.json

[
  {
    "task": "hellaswag",
    "benchmark": "HellaSwag",
    "metric": "acc_norm",
    "score": 0.25851424019119695,
    "shots": 0,
    "runtime_sec": 156.87,
    "status": "success"
  },
  {
    "task": "piqa",
    "benchmark": "PIQA",
    "metric": "acc_norm",
    "score": 0.5495103373231773,
    "shots": 0,
    "runtime_sec": 27.06,
    "status": "success"
  },
  {
    "task": "winogrande",
    "benchmark": "WinoGrande",
    "metric": "acc",
    "score": 0.5027624309392266,
    "shots": 0,
    "runtime_sec": 20.36,
    "status": "success"
  },
  {
    "task": "boolq",
    "benchmark": "BoolQ",
    "metric": "acc",
    "score": 0.37920489296636084,
    "shots": 0,
    "runtime_sec": 47.74,
    "status": "success"
  },
  {
    "task": "arc_easy",
    "benchmark": "ARC-Easy",
    "metric": "acc_norm",
    "score": 0.30387205387205385,
    "shots": 0,
    "runtime_sec": 35.14,
    "status": "success"
  },
  {
    "task": "arc_challenge",
    "benchmark": "ARC-Challenge",
    "metric": "acc_norm",
    "score": 0.23208191126279865,
    "shots": 0,
    "runtime_sec": 26.46,
    "status": "success"
  },
  {
    "task": "openbookqa",
    "benchmark": "OpenBookQA",
    "metric": "acc_norm",
    "score": 0.258,
    "shots": 0,
    "runtime_sec": 20.55,
    "status": "success"
  },
  {
    "task": "commonsense_qa",
    "benchmark": "CommonsenseQA",
    "metric": "acc",
    "score": 0.21048321048321048,
    "shots": 0,
    "runtime_sec": 24.0,
    "status": "success"
  },
  {
    "task": "lambada_openai",
    "benchmark": "LAMBADA",
    "metric": "acc",
    "score": 0.11003299049097613,
    "shots": 0,
    "runtime_sec": 65.51,
    "status": "success"
  },
  {
    "task": "blimp",
    "benchmark": "BLiMP",
    "metric": "acc",
    "score": 0.6532537313432836,
    "shots": 0,
    "runtime_sec": 330.74,
    "status": "success"
  },
  {
    "task": "mmlu",
    "benchmark": "MMLU",
    "metric": "acc",
    "score": 0.24626121635094717,
    "shots": 0,
    "runtime_sec": 269.48,
    "status": "success"
  },
  {
    "task": "wikitext",
    "benchmark": "WikiText-2",
    "metric": "word_perplexity",
    "score": 217.79068002976928,
    "shots": 0,
    "runtime_sec": 38.93,
    "status": "success"
  },
  {
    "task": "wikitext",
    "benchmark": "WikiText-2",
    "metric": "byte_perplexity",
    "score": 2.7366864042695744,
    "shots": 0,
    "runtime_sec": 36.83,
    "status": "success"
  },
  {
    "task": "sciq",
    "benchmark": "SciQ",
    "metric": "acc_norm",
    "score": 0.425,
    "shots": 0,
    "runtime_sec": 41.89,
    "status": "success"
  },
  {
    "task": "copa",
    "benchmark": "COPA",
    "metric": "acc",
    "score": 0.52,
    "shots": 0,
    "runtime_sec": 16.87,
    "status": "success"
  },
  {
    "task": "race",
    "benchmark": "RACE",
    "metric": "acc",
    "score": 0.23636363636363636,
    "shots": 0,
    "runtime_sec": 116.74,
    "status": "success"
  },
  {
    "task": "swag",
    "benchmark": "SWAG",
    "metric": "acc_norm",
    "score": 0.32810156952914127,
    "shots": 0,
    "runtime_sec": 138.44,
    "status": "success"
  },
  {
    "task": "truthfulqa_mc2",
    "benchmark": "TruthfulQA MC2",
    "metric": "acc",
    "score": 0.48090628877613545,
    "shots": 0,
    "runtime_sec": 48.45,
    "status": "success"
  }
]