4b284b12bc4/evaluation/generation/slim.4b284b12bc4_gem_xsum_article_DOC_summary_3.json

{
  "results": [
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_precision": 0.12156176985300436,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_precision_stderr": 0.0019916743006548566
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_recall": 0.2919486745230574,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_recall_stderr": 0.004594202930152972
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_fmeasure": 0.16788870363340208,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_fmeasure_stderr": 0.0026015869149755492
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_precision": 0.023390561110934703,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_precision_stderr": 0.0009482124699139898
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_recall": 0.05931283277905219,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_recall_stderr": 0.00253961685376344
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_fmeasure": 0.03291830334125208,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_fmeasure_stderr": 0.0013348595001679636
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_precision": 0.09410425031338661,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_precision_stderr": 0.001505615984367887
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_recall": 0.22729829891507616,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_recall_stderr": 0.0035793671647219765
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_fmeasure": 0.12998891345659275,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_fmeasure_stderr": 0.00194502365003499
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_precision": 0.09794367331761664,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_precision_stderr": 0.0016077634502078913
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_recall": 0.23679450093245694,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_recall_stderr": 0.003856459984346274
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_fmeasure": 0.13536430596241408,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_fmeasure_stderr": 0.0020963457328233175
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "bleu": 1.3876528749760366,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "bleu_stderr": 0.09352517366139018
    }
  ],
  "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": 3,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}