lm1-8b7-178b-c4-repetitions
/
8b7178b88b
/evaluation
/generation
/slim.8b7178b88b_GEM-wiki_lingua_en_tldr_en_1.json
{ | |
"results": [ | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge1_precision": 0.20521455786196224, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge1_precision_stderr": 0.0021566423356602616 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge1_recall": 0.3450088538250452, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge1_recall_stderr": 0.0028141198924903532 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge1_fmeasure": 0.2384414177661812, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge1_fmeasure_stderr": 0.001967738213001087 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge2_precision": 0.0541891609093382, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge2_precision_stderr": 0.0010252409370391552 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge2_recall": 0.09354429527016248, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge2_recall_stderr": 0.0017970539715243055 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rouge2_fmeasure": 0.06328901574064953, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rouge2_fmeasure_stderr": 0.0011184668519141068 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeL_precision": 0.14442950228734072, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeL_precision_stderr": 0.0014388633343097553 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeL_recall": 0.25157906179953965, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeL_recall_stderr": 0.002266706627328072 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeL_fmeasure": 0.16967595559198553, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeL_fmeasure_stderr": 0.0013571592618223686 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeLsum_precision": 0.19302045441232318, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeLsum_precision_stderr": 0.002028535492752175 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeLsum_recall": 0.3251189492217731, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeLsum_recall_stderr": 0.0026790268923513527 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"rougeLsum_fmeasure": 0.22437259777229618, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"rougeLsum_fmeasure_stderr": 0.0018534765898883354 | |
}, | |
{ | |
"task_name": "GEM/wiki_lingua_en", | |
"prompt_name": "tldr_en", | |
"bleu": 3.3568815479070158, | |
"dataset_path": "GEM/wiki_lingua", | |
"dataset_name": "en", | |
"subset": null, | |
"bleu_stderr": 0.062408281042864173 | |
} | |
], | |
"config": { | |
"model": "hf-causal", | |
"model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-8b7-178b-c4-repetitions/8b7178b88b/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16", | |
"task_args": "", | |
"num_fewshot": 1, | |
"batch_size": 16, | |
"device": "cuda", | |
"use_cache": false, | |
"limit": 3000, | |
"bootstrap_iters": 10, | |
"seed": 1234 | |
} | |
} |