2b855b14bc4/evaluation/generation/slim.2b855b14bc4_GEM-wiki_lingua_en_tldr_en_1.json

{
  "results": [
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_precision": 0.2207038496805443,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0032659748628433537
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_recall": 0.251663946725595,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0029657894607177486
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_fmeasure": 0.19417305007072105,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0020131275910174776
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_precision": 0.05709462181330353,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0019186803623788273
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_recall": 0.06012367373911754,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.0015152413452514553
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_fmeasure": 0.04589509928417269,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0010508101322780883
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_precision": 0.1704632379450581,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.002734203705081264
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_recall": 0.19213302081797887,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.002299044580273186
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_fmeasure": 0.146599519801128,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0014355367566210007
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_precision": 0.20652656672059935,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.003131015188053134
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_recall": 0.23433499607892588,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.0027691250912208654
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_fmeasure": 0.1807008379626855,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0018700130592040445
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "bleu": 2.1926147683168558,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.05138575208910714
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-2b8-55b-c4-repetitions/2b855b14bc4/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16",
    "task_args": "",
    "num_fewshot": 1,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}