Welcome everyone to exchange and discuss!</p>\n","updatedAt":"2026-07-28T03:28:37.431Z","author":{"_id":"65eb0831219e61b9d4b90b22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65eb0831219e61b9d4b90b22/wtbNuCwxaaBJ0oukHrLgn.jpeg","fullname":"IceInPot","name":"IceInPot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.859928548336029},"editors":["IceInPot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/65eb0831219e61b9d4b90b22/wtbNuCwxaaBJ0oukHrLgn.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.24516","authors":[{"_id":"6a681a8673f69d5af2bec624","name":"Jiahao Xie","hidden":false},{"_id":"6a681a8673f69d5af2bec625","user":{"_id":"65eb0831219e61b9d4b90b22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65eb0831219e61b9d4b90b22/wtbNuCwxaaBJ0oukHrLgn.jpeg","isPro":false,"fullname":"IceInPot","user":"IceInPot","type":"user","name":"IceInPot"},"name":"Zhongbin Guo","status":"claimed_verified","statusLastChangedAt":"2026-07-28T08:57:58.911Z","hidden":false},{"_id":"6a681a8673f69d5af2bec626","name":"Qianle Wang","hidden":false},{"_id":"6a681a8673f69d5af2bec627","name":"Ruiqi Lu","hidden":false},{"_id":"6a681a8673f69d5af2bec628","name":"Dongling Xiao","hidden":false},{"_id":"6a681a8673f69d5af2bec629","name":"Wanxuan Sun","hidden":false},{"_id":"6a681a8673f69d5af2bec62a","name":"Cheng Yang","hidden":false}],"publishedAt":"2026-07-27T00:00:00.000Z","submittedOnDailyAt":"2026-07-28T00:00:00.000Z","title":"DecoupleMix: Decoupled Ratio Search and Convex Allocation for Scalable VLM Data Recipes","submittedOnDailyBy":{"_id":"65eb0831219e61b9d4b90b22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65eb0831219e61b9d4b90b22/wtbNuCwxaaBJ0oukHrLgn.jpeg","isPro":false,"fullname":"IceInPot","user":"IceInPot","type":"user","name":"IceInPot"},"summary":"While data curation for Vision Language Models (VLMs) is increasingly active, public practice for constructing pretraining mixtures remains largely heuristic: practitioners stack datasets that pass quality filters, set cross-domain ratios by intuition, and lack a principled, attributable criterion for admitting new data, while frontier recipes remain undisclosed. We formulate data construction as a systematic mixture-optimization problem and turn it into a reproducible engineering discipline by decoupling the mixture into two orthogonal sub-problems: inter-class ratios across capabilities and intra-class ratios within a category. For inter-class allocation, we use a single-variable iterative search; for intra-class composition, we apply a multidimensional, dataset-level assessment scoring Quality and Difficulty, and formulate selection as a constrained convex optimization with a diversity objective. The DecoupleMix framework delivers two critical capabilities: guiding what data to collect next and rendering dataset validation a controlled, attributable experiment. Experiments show our approach consistently surpasses heuristic baselines. Moreover, optimal ratios discovered on small-scale proxies transfer seamlessly to larger scales without retuning. Using 80B additional multimodal continue-pretraining tokens, our VLM is competitive with strong open-source models trained with substantially larger multimodal budgets.","upvotes":5,"discussionId":"6a681a8673f69d5af2bec62b"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"62c3f4c5b9045dabfc48d6b2","avatarUrl":"/avatars/5cb186d512f77b1e4beeb0edeeb6b2b3.svg","isPro":false,"fullname":"reinotaile","user":"reinotaile","type":"user"},{"_id":"65eb0831219e61b9d4b90b22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65eb0831219e61b9d4b90b22/wtbNuCwxaaBJ0oukHrLgn.jpeg","isPro":false,"fullname":"IceInPot","user":"IceInPot","type":"user"},{"_id":"688597b9bc432a62b41f902d","avatarUrl":"/avatars/c010093eb08319049b9b627c961e9dec.svg","isPro":false,"fullname":"jiangkunze","user":"CloudsRise","type":"user"},{"_id":"648ec0da5adc64d0d97d8b3a","avatarUrl":"/avatars/8e1fe165b8209da7b2e7aa0d4d6b4745.svg","isPro":false,"fullname":"chao","user":"ChaoGong","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.24516.md","query":{}}">
DecoupleMix: Decoupled Ratio Search and Convex Allocation for Scalable VLM Data Recipes
Abstract
While data curation for Vision Language Models (VLMs) is increasingly active, public practice for constructing pretraining mixtures remains largely heuristic: practitioners stack datasets that pass quality filters, set cross-domain ratios by intuition, and lack a principled, attributable criterion for admitting new data, while frontier recipes remain undisclosed. We formulate data construction as a systematic mixture-optimization problem and turn it into a reproducible engineering discipline by decoupling the mixture into two orthogonal sub-problems: inter-class ratios across capabilities and intra-class ratios within a category. For inter-class allocation, we use a single-variable iterative search; for intra-class composition, we apply a multidimensional, dataset-level assessment scoring Quality and Difficulty, and formulate selection as a constrained convex optimization with a diversity objective. The DecoupleMix framework delivers two critical capabilities: guiding what data to collect next and rendering dataset validation a controlled, attributable experiment. Experiments show our approach consistently surpasses heuristic baselines. Moreover, optimal ratios discovered on small-scale proxies transfer seamlessly to larger scales without retuning. Using 80B additional multimodal continue-pretraining tokens, our VLM is competitive with strong open-source models trained with substantially larger multimodal budgets.
Community
Welcome everyone to exchange and discuss!
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/2607.24516 in a model README.md to link it from this page.
Cite arxiv.org/abs/2607.24516 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2607.24516 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.