4b284b12bc4/evaluation/generation/slim.4b284b12bc4_gem_xsum_article_DOC_summary_0.json

{
  "results": [
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_precision": 0.08057623396604993,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_precision_stderr": 0.0018427190672612415
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_recall": 0.1903611222785915,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_recall_stderr": 0.004119319495363413
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_fmeasure": 0.11144536072130062,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_fmeasure_stderr": 0.002449613618907836
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_precision": 0.012439638488265747,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_precision_stderr": 0.0007674450472725088
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_recall": 0.029923546174765742,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_recall_stderr": 0.0018325311907880977
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_fmeasure": 0.01730052045113504,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_fmeasure_stderr": 0.0010550283614850532
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_precision": 0.06836163772266587,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_precision_stderr": 0.001418824283099982
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_recall": 0.16290273051352785,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_recall_stderr": 0.0032456160924942976
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_fmeasure": 0.09482469984719238,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_fmeasure_stderr": 0.001891115330305303
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_precision": 0.06720724706425169,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_precision_stderr": 0.001458820899362241
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_recall": 0.16056752379802697,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_recall_stderr": 0.003380030301216589
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_fmeasure": 0.09329416183941738,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_fmeasure_stderr": 0.0019565438808381327
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "bleu": 0.711911214189282,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "bleu_stderr": 0.062271095680873176
    }
  ],
  "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": 0,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}