4b284b12bc4/evaluation/generation/slim.4b284b12bc4_gem_xsum_article_DOC_summary_5.json

{
  "results": [
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_precision": 0.0025292500918997793,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_precision_stderr": 0.0006843948420078455
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_recall": 0.0020532297175197265,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_recall_stderr": 0.0005556296741534733
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_fmeasure": 0.002218415772902317,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_fmeasure_stderr": 0.0005977307451849004
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_precision": 0.0004376650603065697,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_precision_stderr": 0.00024169751059179606
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_recall": 0.00040004436924525715,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_recall_stderr": 0.0002447488200268485
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_fmeasure": 0.00041371259854665804,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_fmeasure_stderr": 0.0002415623180229552
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_precision": 0.0021231909904583543,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_precision_stderr": 0.0005742362353931501
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_recall": 0.0017609197535564964,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_recall_stderr": 0.0004851127145778472
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_fmeasure": 0.0018828340816919485,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_fmeasure_stderr": 0.0005121954193145257
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_precision": 0.002170837264519722,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_precision_stderr": 0.0005838940116621144
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_recall": 0.0017952250708806817,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_recall_stderr": 0.0004910386484249093
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_fmeasure": 0.0019227239855572795,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_fmeasure_stderr": 0.0005197852399034371
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "bleu": 2.9417748605436574e-39,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "bleu_stderr": 1.644365126953672e-33
    }
  ],
  "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
  }
}