4b284b12boscar/evaluation/generation/slim.4b284b12boscar_gem_xsum_article_DOC_summary_3.json

{
  "results": [
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_precision": 0.21408207074509233,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_precision_stderr": 0.003911771705487585
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_recall": 0.27616262040256995,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_recall_stderr": 0.00445926401302726
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge1_fmeasure": 0.22352993200471194,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge1_fmeasure_stderr": 0.0034158533172611158
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_precision": 0.04872713712902665,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_precision_stderr": 0.0021207481086453045
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_recall": 0.06294658987847104,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_recall_stderr": 0.002453274077945568
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rouge2_fmeasure": 0.0505319645754636,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rouge2_fmeasure_stderr": 0.0019929881863291095
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_precision": 0.159438518303562,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_precision_stderr": 0.003102061741719361
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_recall": 0.20607630574913904,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_recall_stderr": 0.0035265999447064503
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeL_fmeasure": 0.16637492784414312,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeL_fmeasure_stderr": 0.0027123155106531316
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_precision": 0.16373162683207845,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_precision_stderr": 0.0030891132218923796
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_recall": 0.2156279714811016,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_recall_stderr": 0.003810201919065186
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "rougeLsum_fmeasure": 0.17215854927610266,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "rougeLsum_fmeasure_stderr": 0.00276757297974389
    },
    {
      "task_name": "gem_xsum",
      "prompt_name": "article_DOC_summary",
      "bleu": 2.2845962769932853,
      "dataset_path": "GEM/xsum",
      "dataset_name": null,
      "subset": "",
      "bleu_stderr": 0.1118763427251602
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-oscar-repetitions/4b284b12boscar/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
  }
}