Language Models are Unsupervised Multitask Learners

Preprint 2019 Alec RadfordJeffrey WuRewon ChildDavid LuanDario AmodeiIlya Sutskever

Natural language processing tasks, such as question answering, machine translation, reading comprehension, and summarization, are typically approached with supervised learning on taskspecific datasets. We demonstrate that language models begin to learn these tasks without any explicit supervision when trained on a new dataset of millions of webpages called WebText... (read more)

PDF

Results from the Paper


 SOTA for Language Modelling on Text8 (using extra training data)

     Get a GitHub badge
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
LEADERBOARD
Question Answering Children's Book Test GPT-2 Accuracy-CN 93.30% # 1
Accuracy-NE 89.05% # 1
Document Summarization CNN / Daily Mail GPT-2 ROUGE-1 29.34 # 13
ROUGE-2 8.27 # 13
ROUGE-L 26.58 # 13
Language Modelling enwiki8 GPT-2x Bit per Character (BPC) 0.93 # 1
Number of params 1542M # 1
Language Modelling enwiki8 GPT-2 Bit per Character (BPC) 0.93 # 1
Number of params 1542M # 1
Language Modelling One Billion Word GPT-2 PPL 42.16 # 14
Number of params 1.54B # 1
Language Modelling Penn Treebank (Word Level) GPT-2 Test perplexity 35.76 # 1
Params 1542M # 1
Language Modelling Text8 GPT-2 Bit per Character (BPC) 0.98 # 1
Language Modelling WikiText-103 GPT-2 Medium Test perplexity 26.37 # 16
Number of params 355M # 1
Language Modelling WikiText-103 GPT-2 Small Test perplexity 37.50 # 27
Number of params 124M # 1
Language Modelling WikiText-103 GPT-2 Large Test perplexity 22.05 # 12
Number of params 774M # 1
Language Modelling WikiText-103 GPT-2 Full Test perplexity 17.48 # 7
Number of params 1542M # 1
Language Modelling WikiText-2 GPT-2 Test perplexity 18.34 # 1
Number of params 1542M # 1
Common Sense Reasoning Winograd Schema Challenge GPT-2 Score 70.70 # 1

Results from Other Papers


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
SOURCE PAPER COMPARE
Natural Language Understanding Wisconsin Sleep Cohort (WSC) GPT-2 Radford et al. (2019) Accuracy 70.7 # 3