4b284b12bc4seed2/evaluation/generation/slim.4b284b12bc4seed2_gem_xsum_article_DOC_summary_0.json

{
  "results": [
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_precision": 0.08488770249277669,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_precision_stderr": 0.001509209602333853
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_recall": 0.19813172755834577,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_recall_stderr": 0.0032474572521749194
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_fmeasure": 0.11690451064634964,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_fmeasure_stderr": 0.001967787449479756
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_precision": 0.008246438330541757,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_precision_stderr": 0.0005824935746807674
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_recall": 0.020228282417241533,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_recall_stderr": 0.0014323430166151545
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_fmeasure": 0.01153407231838246,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_fmeasure_stderr": 0.0008078640211965759
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_precision": 0.07026327535010932,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_precision_stderr": 0.0011182356154930892
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_recall": 0.16554056958926258,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_recall_stderr": 0.002477178559945266
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_fmeasure": 0.09705401381411084,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_fmeasure_stderr": 0.0014626510822599958
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_precision": 0.0700549633031646,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_precision_stderr": 0.0011794948365267052
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_recall": 0.1654358655028746,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_recall_stderr": 0.002682749737328667
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_fmeasure": 0.09684924566454005,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_fmeasure_stderr": 0.001564778913537185
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "bleu": 0.4630139659504365,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "bleu_stderr": 0.07092608567266476
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-oscarseeds/4b284b12bc4seed2/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
  }
}