{"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/190501969","title":"Poly-encoders: Transformer Architectures and Pre-training Strategies for Fast and Accurate Multi-sentence Scoring","arxiv_id":"1905.01969","date":"2019-04-22","proceeding":null,"authors":["Samuel Humeau","Kurt Shuster","Marie-Anne Lachaux","Jason Weston"],"abstract":"The use of deep pre-trained bidirectional transformers has led to remarkable progress in a number of applications (Devlin et al., 2018). For tasks that make pairwise comparisons between sequences, matching a given input with a corresponding label, two approaches are common: Cross-encoders performing full self-attention over the pair and Bi-encoders encoding the pair separately. The former often performs better, but is too slow for practical use. In this work, we develop a new transformer architecture, the Poly-encoder, that learns global rather than token level self-attention features. We perform a detailed comparison of all three approaches, including what pre-training and fine-tuning strategies work best. We show our models achieve state-of-the-art results on three existing tasks; that Poly-encoders are faster than Cross-encoders and more accurate than Bi-encoders; and that the best results are obtained by pre-training on large datasets similar to the downstream tasks.","url_abs":"https://arxiv.org/abs/1905.01969v4","url_pdf":"https://arxiv.org/pdf/1905.01969v4.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":"190501969","repo_url":"https://github.com/Alexey-Borisov/3_course_diary","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"190501969","repo_url":"https://github.com/chijames/Poly-Encoder","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"190501969","repo_url":"https://github.com/csong27/collision-bert","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"190501969","repo_url":"https://github.com/fangrouli/Document-embedding-generation-models","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"190501969","repo_url":"https://github.com/i2r-simmc/i2r-simmc-2020","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"190501969","repo_url":"https://github.com/llStringll/Poly-encoders","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"190501969","repo_url":"https://github.com/sfzhou5678/PolyEncoder","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"conversational-response-selection","task_name":"Conversational Response Selection"},{"task_slug":"sentence","task_name":"Sentence"}],"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":"relu","method_name":"ReLU"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/conversational-response-selection-on-dstc7","task":"Conversational Response Selection","dataset":"DSTC7 Ubuntu","model":"Bi-encoder (v2)","rank_in_archive_order":2,"of":5,"metrics":{"1-of-100 Accuracy":"70.9%"},"uses_additional_data":false},{"leaderboard":"/sota/conversational-response-selection-on-dstc7","task":"Conversational Response Selection","dataset":"DSTC7 Ubuntu","model":"Bi-encoder","rank_in_archive_order":3,"of":5,"metrics":{"1-of-100 Accuracy":"66.3%"},"uses_additional_data":false},{"leaderboard":"/sota/conversational-response-selection-on-douban-1","task":"Conversational Response Selection","dataset":"Douban","model":"Poly-encoder","rank_in_archive_order":9,"of":16,"metrics":{"MAP":"0.608","MRR":"0.650","P@1":"0.475","R10@1":"0.299","R10@2":"0.494","R10@5":"0.822"},"uses_additional_data":false},{"leaderboard":"/sota/conversational-response-selection-on-rrs-1","task":"Conversational Response Selection","dataset":"RRS Ranking Test","model":"Poly-encoder","rank_in_archive_order":1,"of":4,"metrics":{"NDCG@3":"0.679","NDCG@5":"0.765"},"uses_additional_data":false},{"leaderboard":"/sota/conversational-response-selection-on-ubuntu-1","task":"Conversational Response Selection","dataset":"Ubuntu Dialogue (v1, Ranking)","model":"Poly-encoder","rank_in_archive_order":8,"of":25,"metrics":{"R10@1":"0.882","R10@2":"0.949","R10@5":"0.990"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1905.01969","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}