evaluation_bloomz-7b1-p3/evaluation_val/agg.limited=3000.model=p31lossseqglobal_step1000.task=mlsum_es.templates=layman_summ_es.fewshot=0.batchsize=16.seed=1234.timestamp=2022-09-09T23:48:40.json

{"results": [{"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "bleu": 2.6830705121606706, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "bleu_stderr": 0.14257713719805254}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_precision": 0.19994210865731296, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_precision_stderr": 0.0026547454621461738}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_recall": 0.22882499765155356, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_recall_stderr": 0.0030837265632016487}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_fmeasure": 0.19665942356583802, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_fmeasure_stderr": 0.00245186970283176}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_precision": 0.05189155779128239, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_precision_stderr": 0.0016867882237885771}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_recall": 0.06074745104675877, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_recall_stderr": 0.0018807564961523813}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_fmeasure": 0.0515077211753521, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_fmeasure_stderr": 0.0015699354248315028}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_precision": 0.15286714092332523, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_precision_stderr": 0.0021813478665272707}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_recall": 0.17277746632777954, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_recall_stderr": 0.002444939812221139}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_fmeasure": 0.14897381009906005, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_fmeasure_stderr": 0.0019744371501868186}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_precision": 0.15829244217518917, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_precision_stderr": 0.002241835408305656}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_recall": 0.17998021570910885, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_recall_stderr": 0.002565278973604084}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_fmeasure": 0.15454418245332874, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "6fc70031-95ab-40fa-9cc7-e6eda42a4833", "prompt_jinja": "My college roommate asked me what this Spanish article meant:\n {{text}}\nSo I recapped it in layman''s terms in Spanish: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_fmeasure_stderr": 0.0020344070829227297}], "config": {"model": "hf-causal", "model_args": "pretrained=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/6b3t0/p31lossseqglobal_step1000,use_accelerate=True,tokenizer=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/6b3t0/p31lossseqglobal_step1000,dtype=float16", "num_fewshot": 0, "batch_size": 16, "device": "cuda", "use_cache": false, "limit": 3000, "bootstrap_iters": 10, "seed": 1234}}