Multilingual text embedding models are commonly adapted using a single training objective across diverse tasks, despite different tasks requiring fundamentally different optimization strategies. We introduce Task-Conditional Flow Matching (TCFM), a multilingual embedding adaptation framework that selectively applies Flow Matching to translation tasks while optimizing retrieval, classification, and pair-classification tasks with objectives better aligned to their learning dynamics. TCFM further combines teacher-guided representation preservation with a three-stage curriculum to enable stable adaptation. Evaluated on the Indic Massive Text Embedding Benchmark, TCFM establishes a new state-of-the-art, consistently improving embedding quality across a diverse set of multilingual tasks and generalizing across embedding model families. We will publicly release the codebase and datasets upon acceptance of the paper.</p>\n","updatedAt":"2026-08-07T09:02:36.194Z","author":{"_id":"6a478f51d7b115876732b3bb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a478f51d7b115876732b3bb/nzbb9QpTTXivJBshyOs5u.jpeg","fullname":"Tirth Bhatt","name":"Tirth021","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.855907142162323},"editors":["Tirth021"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6a478f51d7b115876732b3bb/nzbb9QpTTXivJBshyOs5u.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.05785","authors":[{"_id":"6a757038e1228e04b32382fd","user":{"_id":"6a478f51d7b115876732b3bb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a478f51d7b115876732b3bb/nzbb9QpTTXivJBshyOs5u.jpeg","isPro":false,"fullname":"Tirth Bhatt","user":"Tirth021","type":"user","name":"Tirth021"},"name":"Tirth Bhatt","status":"claimed_verified","statusLastChangedAt":"2026-08-07T08:45:04.570Z","hidden":false},{"_id":"6a757038e1228e04b32382fe","name":"Naren Kumar S","hidden":false},{"_id":"6a757038e1228e04b32382ff","name":"Mayank Singh","hidden":false}],"publishedAt":"2026-08-06T00:00:00.000Z","submittedOnDailyAt":"2026-08-07T00:00:00.000Z","title":"Task-Conditional Flow Matching for Balanced Multilingual Text Embedding Adaptation","submittedOnDailyBy":{"_id":"6a478f51d7b115876732b3bb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a478f51d7b115876732b3bb/nzbb9QpTTXivJBshyOs5u.jpeg","isPro":false,"fullname":"Tirth Bhatt","user":"Tirth021","type":"user","name":"Tirth021"},"summary":"Multilingual text embedding models are commonly adapted using a single training objective across diverse tasks, despite different tasks requiring fundamentally different optimization strategies. We introduce Task-Conditional Flow Matching (TCFM), a multilingual embedding adaptation framework that selectively applies Flow Matching to translation tasks while optimizing retrieval, classification, and pair-classification tasks with objectives better aligned to their learning dynamics. TCFM further combines teacher-guided representation preservation with a three-stage curriculum to enable stable adaptation. Evaluated on the Indic Massive Text Embedding Benchmark, TCFM establishes a new state-of-the-art, consistently improving embedding quality across a diverse set of multilingual tasks and generalizing across embedding model families. We will publicly release the codebase and datasets upon acceptance of the paper.","upvotes":0,"discussionId":"6a757038e1228e04b3238300","organization":{"_id":"667eb54cc9fc5e32c079544d","name":"LingoIITGN","fullname":"Lingo Research Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/667b8f8ba271fc5a8e6929de/8xB-4Az0x50XC4PS3ZIbL.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"organization":{"_id":"667eb54cc9fc5e32c079544d","name":"LingoIITGN","fullname":"Lingo Research Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/667b8f8ba271fc5a8e6929de/8xB-4Az0x50XC4PS3ZIbL.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.05785.md","query":{}}">
Task-Conditional Flow Matching for Balanced Multilingual Text Embedding Adaptation
Abstract
Multilingual text embedding models are commonly adapted using a single training objective across diverse tasks, despite different tasks requiring fundamentally different optimization strategies. We introduce Task-Conditional Flow Matching (TCFM), a multilingual embedding adaptation framework that selectively applies Flow Matching to translation tasks while optimizing retrieval, classification, and pair-classification tasks with objectives better aligned to their learning dynamics. TCFM further combines teacher-guided representation preservation with a three-stage curriculum to enable stable adaptation. Evaluated on the Indic Massive Text Embedding Benchmark, TCFM establishes a new state-of-the-art, consistently improving embedding quality across a diverse set of multilingual tasks and generalizing across embedding model families. We will publicly release the codebase and datasets upon acceptance of the paper.
Community
Multilingual text embedding models are commonly adapted using a single training objective across diverse tasks, despite different tasks requiring fundamentally different optimization strategies. We introduce Task-Conditional Flow Matching (TCFM), a multilingual embedding adaptation framework that selectively applies Flow Matching to translation tasks while optimizing retrieval, classification, and pair-classification tasks with objectives better aligned to their learning dynamics. TCFM further combines teacher-guided representation preservation with a three-stage curriculum to enable stable adaptation. Evaluated on the Indic Massive Text Embedding Benchmark, TCFM establishes a new state-of-the-art, consistently improving embedding quality across a diverse set of multilingual tasks and generalizing across embedding model families. We will publicly release the codebase and datasets upon acceptance of the paper.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.05785 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.05785 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.05785 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.