Hugging Face Daily Papers · · 4 min read

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA</p>\n","updatedAt":"2026-08-11T02:47:11.312Z","author":{"_id":"64303a4484f3ed1ce62a2c5a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/_pWlU7d5IwrIhEnk0_yFc.jpeg","fullname":"Andrew Chen","name":"anchen1011","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":7,"isUserFollowing":false,"primaryOrg":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/0L3sMuvL2JGls3zUKOKYU.jpeg","fullname":"Mind Lab","name":"mindlab-research","type":"org","isHf":false,"details":"agent model, continual learning","plan":"team"}}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8780060410499573},"editors":["anchen1011"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/_pWlU7d5IwrIhEnk0_yFc.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.09819","authors":[{"_id":"6a7a8c99019ce76dc7b3a9e3","name":"Mind Lab","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9e5","name":"Vin Bo","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9e6","name":"Asher Cai","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9e7","name":"Jingwei Cao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9e8","name":"Song Cao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9e9","name":"Vic Cao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ea","name":"Amelia Chen","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9eb","name":"Andrew Chen","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ec","name":"Kaijie Chen","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ed","name":"Cleon Cheng","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ee","name":"Steven Chiang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ef","name":"Kaixuan Fan","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f0","name":"Hera Feng","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f1","name":"Huan Feng","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f2","name":"Arthur Fu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f3","name":"Jun Gao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f4","name":"Pyke Han","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f5","user":{"_id":"668775b5c82095d2543b47a3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/gPgYQAj_WOqTtehiWIa-6.jpeg","isPro":false,"fullname":"NolanHo","user":"NolanHo","type":"user","name":"NolanHo"},"name":"Nolan Ho","status":"claimed_verified","statusLastChangedAt":"2026-08-11T08:45:04.454Z","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f6","name":"Ori Hong","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f7","name":"Hailee Hou","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f8","name":"Piers Hua","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9f9","name":"Charles Huang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9fa","name":"Miles Jiang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9fb","name":"Nora Jiang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9fc","name":"Yuyi Jiang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9fd","name":"Qiuyu Jin","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9fe","name":"Fancy Kong","hidden":false},{"_id":"6a7a8c99019ce76dc7b3a9ff","name":"Kuss Koo","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa00","name":"Jaron Lee","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa01","name":"Andrew Lei","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa02","name":"Alexy Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa03","name":"Dawn Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa04","name":"Lucian Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa05","name":"Ray Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa06","name":"Ricardo Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa07","name":"Smith Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa08","name":"Theo Li","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa09","name":"Allen Lin","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0a","name":"Elliot Lin","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0b","name":"Fan Lin","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0c","name":"Chen Ling","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0d","name":"Kairus Liu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0e","name":"Kieran Liu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa0f","name":"Logan Liu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa10","name":"Neo Liu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa11","name":"Xiang Liu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa12","name":"Yuxin Lu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa13","name":"Maeve Luo","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa14","name":"Pony Ma","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa15","name":"Verity Niu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa16","name":"Cole Qiao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa17","name":"Guian Qiu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa18","name":"Vince Qu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa19","name":"Sentry","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1a","name":"Niko Song","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1b","name":"Vincent Wang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1c","name":"Bo Wu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1d","name":"Rio Yang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1e","name":"Evelyn Ye","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa1f","name":"Fiona Ye","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa20","name":"Ina Ye","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa21","name":"Regis Ye","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa22","user":{"_id":"6240c0c9d35be2c16ccfbe41","avatarUrl":"/avatars/d30d6c029a89939e7802f590480c5492.svg","isPro":true,"fullname":"Josh Ying","user":"zfying","type":"user","name":"zfying"},"name":"Josh Ying","status":"claimed_verified","statusLastChangedAt":"2026-08-11T08:45:04.463Z","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa23","name":"Atlas Zeng","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa24","name":"Danney Zeng","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa25","name":"Salmon Zhan","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa26","user":{"_id":"66ed14074e66bc3801792a15","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66ed14074e66bc3801792a15/t3gZsX64LI_Y7j-65LNc5.jpeg","isPro":false,"fullname":"Anya Zhang","user":"ShiyaZ","type":"user","name":"ShiyaZ"},"name":"Anya Zhang","status":"claimed_verified","statusLastChangedAt":"2026-08-11T08:45:04.471Z","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa27","user":{"_id":"64bce15bafd1e46c5504ad38","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64bce15bafd1e46c5504ad38/vkEjiu-mIagKlrXzDH75o.png","isPro":false,"fullname":"Di Zhang","user":"di-zhang-fdu","type":"user","name":"di-zhang-fdu"},"name":"Di Zhang","status":"claimed_verified","statusLastChangedAt":"2026-08-11T08:45:04.479Z","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa28","name":"Mia Zhang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa29","name":"Sueky Zhang","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2a","name":"Wei Zhao","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2b","name":"Ada Zhou","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2c","name":"Adrian Zhou","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2d","name":"Yuhua Zhou","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2e","name":"Juno Zhu","hidden":false},{"_id":"6a7a8c99019ce76dc7b3aa2f","name":"Murphy Zhuang","hidden":false}],"publishedAt":"2026-08-10T00:00:00.000Z","submittedOnDailyAt":"2026-08-11T00:00:00.000Z","title":"Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA","submittedOnDailyBy":{"_id":"64303a4484f3ed1ce62a2c5a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/_pWlU7d5IwrIhEnk0_yFc.jpeg","isPro":false,"fullname":"Andrew Chen","user":"anchen1011","type":"user","name":"anchen1011"},"summary":"Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned HCP contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.","upvotes":157,"discussionId":"6a7a8c99019ce76dc7b3aa30","ai_summary":"Macaron-V1 is an open agent-model family that uses a Mixture-of-LoRA architecture and recursive self-improvement to enable continual learning and collaboration across specialized tasks.","ai_keywords":["Mixture-of-LoRA","LoRA adapters","Model-Harness Co-design","recursive self-improvement","MindForge","LongStraw","sparse MoE","GenUI","continual learning"],"ai_summary_model":"thinkingmachines/Inkling-Small","organization":{"_id":"69d05e49e99e437a3b18bebc","name":"mindlab-research","fullname":"Mind Lab","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/0L3sMuvL2JGls3zUKOKYU.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64303a4484f3ed1ce62a2c5a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/_pWlU7d5IwrIhEnk0_yFc.jpeg","isPro":false,"fullname":"Andrew Chen","user":"anchen1011","type":"user"},{"_id":"677aa4d51d4f6bcb2e217c93","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/BPjftZ9rVhaGJfEBRDmLb.png","isPro":false,"fullname":"Xiaoteng Ma","user":"xtma","type":"user"},{"_id":"6621e57973c423da59461b65","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6621e57973c423da59461b65/dbbMP6c7rGA7um97_OKGY.jpeg","isPro":false,"fullname":"Atlas Zeng","user":"Menghuan1918","type":"user"},{"_id":"65aa20bbb68db4f26ebc12a5","avatarUrl":"/avatars/9b4692d2db69b42297306eba15a8bf0d.svg","isPro":false,"fullname":"guoshuai han","user":"0Han0","type":"user"},{"_id":"668775b5c82095d2543b47a3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/gPgYQAj_WOqTtehiWIa-6.jpeg","isPro":false,"fullname":"NolanHo","user":"NolanHo","type":"user"},{"_id":"6677dd120f480511595f489d","avatarUrl":"/avatars/23787ea6190b811d1b9608489173c716.svg","isPro":false,"fullname":"bsqaq","user":"bsqaq","type":"user"},{"_id":"65d4a5788fa4ad07bc29a48b","avatarUrl":"/avatars/ee70eaae8e19ab0472ff72eb3992fb2f.svg","isPro":false,"fullname":"y","user":"rioy","type":"user"},{"_id":"62e217e52e4cab6e39dd54c2","avatarUrl":"/avatars/18eb5199171711585e928906d7ee75bc.svg","isPro":false,"fullname":"JuneGao","user":"JuneGao","type":"user"},{"_id":"6533bbc824173495ad0513aa","avatarUrl":"/avatars/01ddbe00e26cea1e27c3fdcb96423814.svg","isPro":false,"fullname":"chengyuan","user":"chengyuan0522","type":"user"},{"_id":"68a9c660f5f4b7a2a6cb1824","avatarUrl":"/avatars/63be535e1912c463438b7ce0c318db71.svg","isPro":false,"fullname":"caosong","user":"sant1an","type":"user"},{"_id":"6577e03b3ceeb2f078f66512","avatarUrl":"/avatars/27fe1cff6bc8e6e98966359d7ca4adc4.svg","isPro":false,"fullname":"Xiang Lei","user":"Lei00764","type":"user"},{"_id":"6a0421fdafb8cb93b94cb932","avatarUrl":"/avatars/f23232a3cf1c04ef066c995f501966e1.svg","isPro":false,"fullname":"11","user":"heraaaaaaa","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"organization":{"_id":"69d05e49e99e437a3b18bebc","name":"mindlab-research","fullname":"Mind Lab","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64303a4484f3ed1ce62a2c5a/0L3sMuvL2JGls3zUKOKYU.jpeg"},"query":{}}">
Papers
arxiv:2608.09819

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Published on Aug 10
· Submitted by
Andrew Chen
on Aug 11
#2 Paper of the day
Authors:
,

Abstract

Macaron-V1 is an open agent-model family that uses a Mixture-of-LoRA architecture and recursive self-improvement to enable continual learning and collaboration across specialized tasks.

Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned HCP contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.

Community

Paper submitter about 16 hours ago

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Models citing this paper

No model linking this paper

Cite arxiv.org/abs/2608.09819 in a model README.md to link it from this page.

Datasets citing this paper

No dataset linking this paper

Cite arxiv.org/abs/2608.09819 in a dataset README.md to link it from this page.

Spaces citing this paper

No Space linking this paper

Cite arxiv.org/abs/2608.09819 in a Space README.md to link it from this page.

Collections including this paper

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.

More from Hugging Face Daily Papers