{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/when-attention-meets-fast-recurrence-training","title":"When Attention Meets Fast Recurrence: Training Language Models with Reduced Compute","arxiv_id":"2102.12459","date":"2021-02-24","proceeding":"EMNLP 2021 11","authors":["Tao Lei"],"abstract":"Large language models have become increasingly difficult to train because of the growing computation time and cost. In this work, we present SRU++, a highly-efficient architecture that combines fast recurrence and attention for sequence modeling. SRU++ exhibits strong modeling capacity and training efficiency. On standard language modeling tasks such as Enwik8, Wiki-103 and Billion Word datasets, our model obtains better bits-per-character and perplexity while using 3x-10x less training cost compared to top-performing Transformer models. For instance, our model achieves a state-of-the-art result on the Enwik8 dataset using 1.6 days of training on an 8-GPU machine. We further demonstrate that SRU++ requires minimal attention for near state-of-the-art performance. Our results suggest jointly leveraging fast recurrence with little attention as a promising direction for accelerating model training and inference.","url_abs":"https://arxiv.org/abs/2102.12459v3","url_pdf":"https://arxiv.org/pdf/2102.12459v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"when-attention-meets-fast-recurrence-training","repo_url":"https://github.com/asappresearch/sru","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":null,"task_name":"GPU"},{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"machine-translation","task_name":"Machine Translation"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"sru-1","method_name":"SRU++"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[{"slug":"sru-1","name":"SRU++","full_name":"SRU++"}],"results":[{"leaderboard":"/sota/language-modelling-on-one-billion-word","task":"Language Modelling","dataset":"One Billion Word","model":"SRU++ Large","rank_in_archive_order":9,"of":27,"metrics":{"Number of params":"465M","PPL":"23.5"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-one-billion-word","task":"Language Modelling","dataset":"One Billion Word","model":"SRU++","rank_in_archive_order":14,"of":27,"metrics":{"Number of params":"328M","PPL":"25.1"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-wikitext-103","task":"Language Modelling","dataset":"WikiText-103","model":"SRU++ Large","rank_in_archive_order":21,"of":89,"metrics":{"Number of params":"234M","Test perplexity":"17.1","Validation perplexity":"16.4"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-wikitext-103","task":"Language Modelling","dataset":"WikiText-103","model":"SRU++ Base","rank_in_archive_order":33,"of":89,"metrics":{"Number of params":"148M","Test perplexity":"18.3","Validation perplexity":"17.5"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-enwiki8","task":"Language Modelling","dataset":"enwik8","model":"SRU++ Large","rank_in_archive_order":5,"of":42,"metrics":{"Bit per Character (BPC)":"0.95","Number of params":"195M"},"uses_additional_data":false},{"leaderboard":"/sota/language-modelling-on-enwiki8","task":"Language Modelling","dataset":"enwik8","model":"SRU++ Base","rank_in_archive_order":10,"of":42,"metrics":{"Bit per Character (BPC)":"0.97","Number of params":"108M"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2102.12459","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}