\n","updatedAt":"2024-06-09T06:11:58.242Z","author":{"_id":"6186ddf6a7717cb375090c01","avatarUrl":"/avatars/716b6a7d1094c8036b2a8a7b9063e8aa.svg","fullname":"Julien BLANCHON","name":"blanchon","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":142}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5297850370407104},"editors":["blanchon"],"editorAvatarUrls":["/avatars/716b6a7d1094c8036b2a8a7b9063e8aa.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"1905.10650","authors":[{"_id":"6551a6caf3c79cdaf3c9ce92","name":"Paul Michel","hidden":false},{"_id":"6551a6caf3c79cdaf3c9ce93","name":"Omer Levy","hidden":false},{"_id":"6551a6caf3c79cdaf3c9ce94","name":"Graham Neubig","hidden":false}],"publishedAt":"2019-05-25T18:27:28.000Z","title":"Are Sixteen Heads Really Better than One?","summary":"Attention is a powerful and ubiquitous mechanism for allowing neural models\nto focus on particular salient pieces of information by taking their weighted\naverage when making predictions. In particular, multi-headed attention is a\ndriving force behind many recent state-of-the-art NLP models such as\nTransformer-based MT models and BERT. These models apply multiple attention\nmechanisms in parallel, with each attention \"head\" potentially focusing on\ndifferent parts of the input, which makes it possible to express sophisticated\nfunctions beyond the simple weighted average. In this paper we make the\nsurprising observation that even if models have been trained using multiple\nheads, in practice, a large percentage of attention heads can be removed at\ntest time without significantly impacting performance. In fact, some layers can\neven be reduced to a single head. We further examine greedy algorithms for\npruning down models, and the potential speed, memory efficiency, and accuracy\nimprovements obtainable therefrom. Finally, we analyze the results with respect\nto which parts of the model are more reliant on having multiple heads, and\nprovide precursory evidence that training dynamics play a role in the gains\nprovided by multi-head attention.","upvotes":2,"discussionId":"6551a6cbf3c79cdaf3c9cea8","ai_summary":"Many NLP models can reduce their number of attention heads without significant performance degradation, offering potential improvements in speed and memory efficiency.","ai_keywords":["attention","multi-headed attention","Transformer-based MT models","BERT","pruning","speed","memory efficiency","accuracy","training dynamics"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6538119803519fddb4a17e10","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6538119803519fddb4a17e10/ffJMkdx-rM7VvLTCM6ri_.jpeg","isPro":false,"fullname":"samusenps","user":"samusenps","type":"user"},{"_id":"655ac762cb17ec19ef82719b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655ac762cb17ec19ef82719b/1kDncYrGLYS_2SR8cNdAL.png","isPro":false,"fullname":"Welcome to matlok","user":"matlok","type":"user"}],"acceptLanguages":["*"]}">
Many NLP models can reduce their number of attention heads without significant performance degradation, offering potential improvements in speed and memory efficiency.
AI-generated summary
Attention is a powerful and ubiquitous mechanism for allowing neural models
to focus on particular salient pieces of information by taking their weighted
average when making predictions. In particular, multi-headed attention is a
driving force behind many recent state-of-the-art NLP models such as
Transformer-based MT models and BERT. These models apply multiple attention
mechanisms in parallel, with each attention "head" potentially focusing on
different parts of the input, which makes it possible to express sophisticated
functions beyond the simple weighted average. In this paper we make the
surprising observation that even if models have been trained using multiple
heads, in practice, a large percentage of attention heads can be removed at
test time without significantly impacting performance. In fact, some layers can
even be reduced to a single head. We further examine greedy algorithms for
pruning down models, and the potential speed, memory efficiency, and accuracy
improvements obtainable therefrom. Finally, we analyze the results with respect
to which parts of the model are more reliant on having multiple heads, and
provide precursory evidence that training dynamics play a role in the gains
provided by multi-head attention.