4b284b42bc4seed3/evaluation/generation/agg.4b284b42bc4seed3_gem_xsum_article_DOC_summary_3.json

{"results": [{"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge1_precision": 0.1296338781704996, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_precision_stderr": 0.0021646982503126387}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge1_recall": 0.3021070098062078, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_recall_stderr": 0.004505088335872746}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge1_fmeasure": 0.17680602744097074, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_fmeasure_stderr": 0.002631980246229425}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge2_precision": 0.028309377178479218, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_precision_stderr": 0.0011164748559469404}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge2_recall": 0.06851024845447631, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_recall_stderr": 0.0026775539108549653}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rouge2_fmeasure": 0.03904911931370361, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_fmeasure_stderr": 0.0014830525218956757}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeL_precision": 0.10464417400111554, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_precision_stderr": 0.0017426786877401284}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeL_recall": 0.2455380323796214, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_recall_stderr": 0.003722345532650257}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeL_fmeasure": 0.14303319341608928, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_fmeasure_stderr": 0.0021220160049417853}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeLsum_precision": 0.10170185371430301, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_precision_stderr": 0.001760814642264659}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeLsum_recall": 0.23909447163984915, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_recall_stderr": 0.0037550549904903216}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "rougeLsum_fmeasure": 0.13901962932381862, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_fmeasure_stderr": 0.002140626268901333}, {"task_name": "gem_xsum", "prompt_name": "article_DOC_summary", "bleu": 1.6145690136458937, "fixed_answer_choice_list": null, "dataset_path": "GEM/xsum", "dataset_name": null, "subset": "", "prompt_id": "a8d4ecfa-c944-44d5-878c-04fd5db59e64", "prompt_jinja": "Article: {{document}}\n\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "bleu_stderr": 0.08805787248504932}], "config": {"model": "hf-causal", "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-c4seeds/4b284b42bc4seed3/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}}