{"id":"arxiv:2212.09741","title":"One Embedder, Any Task: Instruction-Finetuned Text Embeddings","date":"2022-12-19","section":"Retrieval, embeddings, benchmarks","quality":0.51946,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/","source_url":"https://arxiv.org/abs/2212.09741","text_available":true,"authors":["Hongjin Su","Weijia Shi","Jungo Kasai","Yizhong Wang"],"outgoing":[{"source":"arxiv:2212.09741","target":"acl:2022.emnlp-main.340","evidence":"id","id":"acl:2022.emnlp-main.340","title":"Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e43a06571b7e8d89abd5f20f8f520c8bc3d3eb32ffb156cb7fc9e0469fc14b0a/"},{"source":"arxiv:2212.09741","target":"arxiv:2112.07899","evidence":"id","id":"arxiv:2112.07899","title":"Large Dual Encoders Are Generalizable Retrievers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a98e608b4e34094fed75a118c0663f5f02f485081981cecc2bf3350181d61a51/"}],"incoming":[{"source":"arxiv:2511.07317","target":"arxiv:2212.09741","evidence":"title","id":"arxiv:2511.07317","title":"RLVE: Scaling Up Reinforcement Learning for Language Models with Adaptive Verifiable Environments","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c0c18cb0ae7fe76a7239f38ce819c7ac588c1184dd27c5672f868ff4e2e26605/"}],"text_parts":[{"start":0,"end":15997,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html"},{"start":15997,"end":31588,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html"},{"start":31588,"end":35862,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html"}],"contents":[{"title":"One Embedder, Any Task: Instruction-Finetuned Text Embeddings","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"Abstract","offset":65,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"1 Introduction","offset":1342,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"2 InstructOR","offset":5341,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"2.1 Embedding Architecture","offset":5915,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"2.2 Training Objective","offset":6836,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"2.3 MEDI: Multitask Embedding Data with Instructions","offset":8934,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"Data Construction","offset":9266,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"Instruction Annotation","offset":11765,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"3 Experiments","offset":13457,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"3.1 Main Results","offset":14016,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"4 Analysis and Ablations","offset":15584,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-1.html#text"},{"title":"4.1 Instructions Enable Diverse Training","offset":15997,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"4.2 Instruction Robustness","offset":17414,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"4.3 Complexity of Instructions","offset":18241,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"4.4 Model Sizes and Instruction Finetuning","offset":19312,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"4.5 Instructions Mitigate Domain Shifts","offset":20133,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"4.6 Qualititive Analysis","offset":20668,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"5 Related Work","offset":21449,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Text Embeddings","offset":21468,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Instruction Finetuning","offset":23195,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"6 Conclusion","offset":24192,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"7 Limitations","offset":24734,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Acknowledgements","offset":25779,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Appendix A Training Setups","offset":25909,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Minibatch Sampling","offset":25940,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Training","offset":26648,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Baselines","offset":27011,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Appendix B Embedding Evaluations","offset":28000,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"B.1 Massive Text Embedding Benchmark","offset":28479,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Retrieval","offset":29166,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Reranking","offset":29824,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Clustering","offset":30256,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Pair Classification","offset":30677,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Classification","offset":31173,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"STS","offset":31578,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-2.html#text"},{"title":"Summarization","offset":32048,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"},{"title":"B.2 Prompt Retrieval","offset":32611,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"},{"title":"B.3 Automatic Evaluation for Generation","offset":33549,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"},{"title":"Appendix C Full instructions","offset":34517,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"},{"title":"Appendix D Full Results","offset":34624,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"},{"title":"Instructions for reporting errors","offset":34898,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/part-3.html#text"}],"text_code_points":35862,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b162d4a59ad6b25fd180e566615cf3dc2cd3a6c505f0421e15fd036236ad2ecd/source.md"}