{"results": [{"task_name": "mlsum_es", "prompt_name": "palm_prompt", "bleu": 3.916653973278921, "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": 0.13320174843239593}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_precision": 0.238006133260283, "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": 0.002592201534618347}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_recall": 0.28440217742758356, "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": 0.0032686854143649984}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge1_fmeasure": 0.23777639529079925, "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": 0.002353644348181871}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_precision": 0.07002703238460953, "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": 0.0019467819130961877}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_recall": 0.08785375976968864, "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": 0.002327977791822037}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rouge2_fmeasure": 0.07113316160241126, "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": 0.0018688817237639986}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_precision": 0.18083602836591542, "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": 0.002222442145519105}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_recall": 0.21460635266053055, "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": 0.0026818176382247703}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeL_fmeasure": 0.17988536856242504, "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": 0.0020214064930605237}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_precision": 0.18813961119405384, "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": 0.00226738528856828}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_recall": 0.2256886585459008, "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": 0.0028543730103714284}, {"task_name": "mlsum_es", "prompt_name": "palm_prompt", "rougeLsum_fmeasure": 0.18777680808178215, "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": 0.0020733063902455624}], "config": {"model": "hf-causal", "model_args": "pretrained=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/176bt0/p31lossseq_global_step498,use_accelerate=True,tokenizer=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/176bt0/p31lossseq_global_step498,dtype=bfloat16", "num_fewshot": 0, "batch_size": 4, "device": "cuda", "use_cache": false, "limit": 3000, "bootstrap_iters": 10, "seed": 1234}}