Try the live demo: <a href=\"https://huggingface.co/spaces/xiaomi-research/milmmt-46-translation\">https://huggingface.co/spaces/xiaomi-research/milmmt-46-translation</a></p>\n","updatedAt":"2026-08-12T04:28:11.035Z","author":{"_id":"64db17edd68a6ddcc7b3ffd9","avatarUrl":"/avatars/61628600921e4641c1ddd034f3837e9a.svg","fullname":"Pengzhi Gao","name":"gpengzhi","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.907363772392273},"editors":["gpengzhi"],"editorAvatarUrls":["/avatars/61628600921e4641c1ddd034f3837e9a.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.10812","authors":[{"_id":"6a7bdc9d1653ef87c6af1bd6","name":"Chris Han","hidden":false},{"_id":"6a7bdc9d1653ef87c6af1bd7","name":"Pengzhi Gao","hidden":false},{"_id":"6a7bdc9d1653ef87c6af1bd8","name":"Pei Fu","hidden":false},{"_id":"6a7bdc9d1653ef87c6af1bd9","name":"Jian Luan","hidden":false}],"publishedAt":"2026-08-11T00:00:00.000Z","submittedOnDailyAt":"2026-08-12T00:00:00.000Z","title":"Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation","submittedOnDailyBy":{"_id":"64db17edd68a6ddcc7b3ffd9","avatarUrl":"/avatars/61628600921e4641c1ddd034f3837e9a.svg","isPro":false,"fullname":"Pengzhi Gao","user":"gpengzhi","type":"user","name":"gpengzhi"},"summary":"We study reference-free post-training for multilingual machine translation with open large language models. Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO) with a reward that averages two reference-free quality estimation models and is gated by language identification. We then linearly interpolate the supervised fine-tuning (SFT) and reinforcement learning (RL) model checkpoints to obtain MiLMMT-46-v1.0. Across 46 languages, the resulting models consistently improve translation quality over their SFT counterparts, outperform strong recent open baselines, including Seed-X, HY-MT2, and TranslateGemma, and achieve leading reference-free scores against evaluated proprietary systems such as Google Translate, Gemini 3 Pro, and GPT-5. We further investigate on-policy distillation and find that it reaches, but does not surpass, the quality frontier achieved by RL with checkpoint interpolation. We release the models and code to facilitate future research.","upvotes":5,"discussionId":"6a7bdc9d1653ef87c6af1bda","githubRepo":"https://github.com/xiaomi-research/gemmax","githubRepoAddedBy":"user","ai_summary":"Open multilingual translation models are improved via group relative policy optimization with reference-free quality rewards and checkpoint interpolation, surpassing strong open and proprietary baselines.","ai_keywords":["Group Relative Policy Optimization","quality estimation","language identification","supervised fine-tuning","reinforcement learning","checkpoint interpolation","on-policy distillation","multilingual machine translation"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":68,"organization":{"_id":"6821ba7e5a7efab94a235406","name":"xiaomi-research","fullname":"Xiaomi Research","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/673735e4373ad40af7f81ea1/DR4m0bz2Du1l0Z8Txg351.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64db17edd68a6ddcc7b3ffd9","avatarUrl":"/avatars/61628600921e4641c1ddd034f3837e9a.svg","isPro":false,"fullname":"Pengzhi Gao","user":"gpengzhi","type":"user"},{"_id":"69bcb85876fe7a5ea0a52581","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/NMYcwRgtzWxIzkoaAgSOh.jpeg","isPro":false,"fullname":"井上 美月","user":"jrobinson9","type":"user"},{"_id":"6a5db0ce059560f3ce0455d7","avatarUrl":"/avatars/71f29595ada24ae57c0d8b2004a2ebf2.svg","isPro":false,"fullname":"Cathy Payne","user":"cathypayne9","type":"user"},{"_id":"6694e62da0a0fcbfa89f5077","avatarUrl":"/avatars/87004096308a290f4c7f7a55086120fc.svg","isPro":false,"fullname":"HAN","user":"HAN1111111","type":"user"},{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","isPro":false,"fullname":"Urro","user":"urroxyz","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6821ba7e5a7efab94a235406","name":"xiaomi-research","fullname":"Xiaomi Research","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/673735e4373ad40af7f81ea1/DR4m0bz2Du1l0Z8Txg351.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.10812.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.10812 in a dataset README.md to link it from this page.