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

{"results": [{"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "bleu": 3.612948631805906, "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": 0.09706767527109322}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_precision": 0.25720877609362947, "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": 0.002866204775543915}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_recall": 0.21570826045711883, "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": 0.0028219768467931316}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge1_fmeasure": 0.2132867654263301, "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": 0.0023414288877090226}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_precision": 0.0656763315976497, "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": 0.002091174894878516}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_recall": 0.058883909534972835, "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": 0.0018684387267495434}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rouge2_fmeasure": 0.056007499633099506, "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": 0.001671996707075678}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_precision": 0.20296195866113786, "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": 0.002463373191290898}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_recall": 0.16800778229835897, "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": 0.0023113725152130527}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeL_fmeasure": 0.16637065130970433, "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": 0.0019401940534334063}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_precision": 0.20768190658213875, "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": 0.0024996543423916463}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_recall": 0.17213528120005406, "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": 0.0023628667195956534}, {"task_name": "mlsum_es", "prompt_name": "layman_summ_es", "rougeLsum_fmeasure": 0.1703003950966778, "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": 0.0019655927463575074}], "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}}