{
  "results": {
    "arc_easy": {
      "acc": 0.43223905723905726,
      "acc_stderr": 0.010165130379698762,
      "acc_norm": 0.3867845117845118,
      "acc_norm_stderr": 0.009993308355370987
    },
    "boolq": {
      "acc": 0.6091743119266055,
      "acc_stderr": 0.00853404406545651
    },
    "lambada_openai": {
      "ppl": 74.2398707720553,
      "ppl_stderr": 2.903833914079987,
      "acc": 0.2604308169998059,
      "acc_stderr": 0.0061143129377882795
    },
    "openbookqa": {
      "acc": 0.144,
      "acc_stderr": 0.015716934945725767,
      "acc_norm": 0.278,
      "acc_norm_stderr": 0.020055833888070914
    },
    "piqa": {
      "acc": 0.5908596300326442,
      "acc_stderr": 0.011471593460443328,
      "acc_norm": 0.5870511425462459,
      "acc_norm_stderr": 0.011487658725079094
    },
    "winogrande": {
      "acc": 0.5224940805051302,
      "acc_stderr": 0.014038257824059874
    }
  },
  "versions": {
    "arc_easy": 0,
    "boolq": 1,
    "lambada_openai": 0,
    "openbookqa": 0,
    "piqa": 0,
    "winogrande": 0
  },
  "config": {
    "model": "hf-causal-experimental",
    "model_args": "pretrained=BEE-spoke-data/smol_llama-101M-GQA,revision=main,trust_remote_code=True,dtype='float'",
    "num_fewshot": 0,
    "batch_size": "8",
    "batch_sizes": [],
    "device": "cuda",
    "no_cache": false,
    "limit": null,
    "bootstrap_iters": 100000,
    "description_dict": {}
  }
}