4b284b12bc4/evaluation/generation/slim.4b284b12bc4_GEM-wiki_lingua_en_tldr_en_5.json

{
  "results": [
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_precision": 0.008397509125114434,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0008676037544483993
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_recall": 0.013182384245950918,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0012488579545023588
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_fmeasure": 0.00885537771521418,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0008474145806800235
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_precision": 0.0018519121687661717,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0002652232008037664
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_recall": 0.003201546871291623,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.00042951400653203656
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_fmeasure": 0.0020845828252393957,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0002776449859686965
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_precision": 0.006312806712827651,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.0006339300731280119
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_recall": 0.010466170637582555,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.0010119772330203227
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_fmeasure": 0.006795329450745382,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0006430172034977336
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_precision": 0.0076894646083944945,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.000786099271409283
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_recall": 0.012227793486074013,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.0011695312324965957
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_fmeasure": 0.008112137370163754,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0007719985459243486
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "bleu": 5.133528491740168e-07,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 9.288876136024227e-07
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-c4-repetitions/4b284b12bc4/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16",
    "task_args": "",
    "num_fewshot": 5,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}