evaluation_val/evaluation_mlsum/agg.limited=3000.model=tr13f-6b3-ml-t0-lmtoks341b-t0toks4b2-xp3capmixnewcodelonglossseq.task=mlsum_es.templates=palm_prompt.fewshot=0.batchsize=16.seed=1234.timestamp=2022-09-06T18:04:39.json

{"results": [{"task_name": "mlsum_es", "prompt_name": "palm_prompt", "bleu": 4.815053212697339, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "bleu_stderr": 0.22929921240192092}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_precision": 0.2664214534878677, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_precision_stderr": 0.002736901307721048}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_recall": 0.259042211572509, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_recall_stderr": 0.003023227343717013}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_fmeasure": 0.24155464763055362, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_fmeasure_stderr": 0.0024021553413102275}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_precision": 0.07596613017420617, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_precision_stderr": 0.002093675494291177}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_recall": 0.07803116812507413, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_recall_stderr": 0.002214980401457295}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_fmeasure": 0.07072194087888328, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_fmeasure_stderr": 0.0019187225902360238}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_precision": 0.20732883923716353, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_precision_stderr": 0.002355023098360087}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_recall": 0.20013530891135448, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_recall_stderr": 0.002545877916378003}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_fmeasure": 0.18681843812599136, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_fmeasure_stderr": 0.0020613027767420913}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_precision": 0.21204351337105004, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_precision_stderr": 0.0024061908027064216}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_recall": 0.2042193199739302, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_recall_stderr": 0.0025788791923903134}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_fmeasure": 0.19073355304638137, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "e3c60771-5e99-49b1-b477-c2b69f645d59", "prompt_jinja": "I will first show a news article and then provide a summary of it in Spanish:\nArticle: {{text}}\n ===\nSummary: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_fmeasure_stderr": 0.0020838845388973935}], "config": {"model": "hf-causal", "model_args": "pretrained=/gpfsscratch/rech/six/commun/experiments/muennighoff/bloomckpt/6b3t0/tr13f-6b3-ml-t0-lmtoks341b-t0toks4b2-xp3capmixnewcodelonglossseq,use_accelerate=True,tokenizer=/gpfsscratch/rech/six/commun/experiments/muennighoff/bloomckpt/6b3t0/tr13f-6b3-ml-t0-lmtoks341b-t0toks4b2-xp3capmixnewcodelonglossseq,dtype=float16", "num_fewshot": 0, "batch_size": 16, "device": "cuda", "use_cache": false, "limit": 3000, "bootstrap_iters": 10, "seed": 1234}}