Hugging Face Science team releases
\n
\n\n\n\n### 🤗 Join us! \n\nWe are actively hiring for both full-time and internships. Check out [hf.co/jobs](https://hf.co/jobs)\n\n","html":"\n
\n\n \n
\n \n \n
\n JUN
\n \n \n \n
\n MAY
\n \n \n \n
\n APR
\n \n \n \n
\n \n MAR
\n \n \n \n
\n \n \n
\n NOV
\n \n \n
\n 🤏 SmolLM2
\n \n \n 🤓 SmolVLM
\n \n \n \n
\n DEC
\n \n \n \n
\n JAN 2025
\n \n \n \n
\n \n FEB
\n \n \n \n
\n \n \n
\n OCT
\n \n \n \n
\n SEP
\n \n \n \n
\n AUG
\n \n \n \n
\n \n JUL
\n \n \n
\n 🛟 Ring attention
\n \n \n 🦾 LeRobot TeleOps
\n \n \n 🥇 Win AIMO
\n \n \n 🐶 Docmatix
\n \n \n 🤏 SmolLM
\n \n \n \n
\n \n \n
\n MAR
\n \n \n
\n 🪁 Zephyr Gemma
\n \n \n 🪐 Cosmopedia
\n \n \n \n
\n APR
\n \n \n \n
\n MAY
\n \n \n \n
\n \n JUN
\n \n \n \n
\n \n \n
\n FEB
\n \n \n \n
\n JAN 2024
\n \n
\n 🔥 Warming up
\n Hugging Face Research
\nThe science team at Hugging Face is dedicated to advancing machine learning research in ways that maximize value for the whole community.
\n🛠️ Tooling & Infrastructure
\nThe foundation of ML research is tooling and infrastructure and we are working on a range of tools such as datatrove, nanotron, TRL, LeRobot, and lighteval.
\n📑 Datasets
\nHigh quality datasets are the powerhouse of LLMs and require special care and skills to build. We focus on building high-quality datasets such as no-robots, FineWeb, The Stack, and FineVideo.
\n🤖 Open Models
\nThe datatsets and training recipes of most state-of-the-art models are not released. We build cutting-edge models and release the full training pipeline as well fostering more innovation and reproducibility, such as Zephyr, StarCoder2, or SmolLM3.
\n🌸 Collaborations
\nResearch and collaboration go hand in hand. That's why we like to organize and participate in large open collaborations such as BigScience and BigCode, as well as lots of smaller partnerships such as Leaderboards on the Hub.
\n⚙️ Infrastructre
\nThe research team is organized in small teams with typically <4 people and the science cluster consists of 96 x 8xH100 nodes as well as an auto-scalable CPU cluster for dataset processing. In this setup, even a small research team can build and push out impactful artifacts.
\n📖 Educational material
\nBesides writing tech reports of research projects we also like to write more educational content to help newcomers get started to the field or practitioners. We built for example the alignment handbook, the evaluation guidebook, the pretraining tutorial, or the FineWeb blog.
\nRelease Timeline
\nThis is the release timeline so far and follow the links by clicking on the elements:
\n\n\n\n
\n\n\n\n\n
\n\n \n
\n \n \n
\n JUN
\n \n \n \n
\n MAY
\n \n \n \n
\n APR
\n \n \n \n
\n \n MAR
\n \n \n \n
\n \n \n
\n NOV
\n \n \n
\n 🤏 SmolLM2
\n \n \n 🤓 SmolVLM
\n \n \n \n
\n DEC
\n \n \n \n
\n JAN 2025
\n \n \n \n
\n \n FEB
\n \n \n \n
\n \n \n
\n OCT
\n \n \n \n
\n SEP
\n \n \n \n
\n AUG
\n \n \n \n
\n \n JUL
\n \n \n
\n 🛟 Ring attention
\n \n \n 🦾 LeRobot TeleOps
\n \n \n 🥇 Win AIMO
\n \n \n 🐶 Docmatix
\n \n \n 🤏 SmolLM
\n \n \n \n
\n \n \n
\n MAR
\n \n \n
\n 🪁 Zephyr Gemma
\n \n \n 🪐 Cosmopedia
\n \n \n \n
\n APR
\n \n \n \n
\n MAY
\n \n \n \n
\n \n JUN
\n \n \n \n
\n \n \n
\n FEB
\n \n \n \n
\n JAN 2024
\n \n
\n 🔥 Warming up
\n 🤗 Join us!
\nWe are actively hiring for both full-time and internships. Check out hf.co/jobs
\n","classNames":"hf-sanitized hf-sanitized-WI_5inlnOEZ6XfHAq5WwI"},"users":[{"_id":"5e48005437cb5b49818287a5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5e48005437cb5b49818287a5/4uCXGGui-9QifAT4qelxU.png","isPro":false,"fullname":"Leandro von Werra","user":"lvwerra","type":"user"},{"_id":"61c141342aac764ce1654e43","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61c141342aac764ce1654e43/81AwoT5IQ_Xdw0OVw7TKu.jpeg","isPro":false,"fullname":"Loubna Ben Allal","user":"loubnabnl","type":"user"},{"_id":"5df7e9e5da6d0311fd3d53f9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1583857746553-5df7e9e5da6d0311fd3d53f9.jpeg","isPro":true,"fullname":"Thomas Wolf","user":"thomwolf","type":"user"},{"_id":"629f3b18ee05727ce328ccbe","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1669189789447-629f3b18ee05727ce328ccbe.jpeg","isPro":false,"fullname":"Kashif Rasul","user":"kashif","type":"user"},{"_id":"651e96991b97c9f33d26bde6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/651e96991b97c9f33d26bde6/-Bqs6qrmz0yCfwtB2e-6q.jpeg","isPro":true,"fullname":"Elie Bakouch","user":"eliebak","type":"user"},{"_id":"5f0c746619cb630495b814fd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1594651707950-noauth.jpeg","isPro":true,"fullname":"Lewis Tunstall","user":"lewtun","type":"user"},{"_id":"602e6dee60e3dd96631c906e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1613655355830-noauth.png","isPro":false,"fullname":"Anton Lozhkov","user":"anton-l","type":"user"},{"_id":"631ce4b244503b72277fc89f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1677431596830-631ce4b244503b72277fc89f.jpeg","isPro":true,"fullname":"Quentin Gallouédec","user":"qgallouedec","type":"user"},{"_id":"6202a599216215a22221dea9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1644340617257-noauth.png","isPro":false,"fullname":"Clémentine Fourrier","user":"clefourrier","type":"user"},{"_id":"63691c3eda9b693c2730b2a2","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63691c3eda9b693c2730b2a2/hBtKpgo3_9003MWCGkw5d.png","isPro":false,"fullname":"Brigitte Tousignant","user":"BrigitteTousi","type":"user"},{"_id":"65d66b494bbd0d92b641cdbb","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d66b494bbd0d92b641cdbb/6-7dm7B-JxcoS1QlCPdMN.jpeg","isPro":false,"fullname":"Andres Marafioti","user":"andito","type":"user"},{"_id":"6200d0a443eb0913fa2df7cc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1644220542819-noauth.jpeg","isPro":true,"fullname":"Edward Beeching","user":"edbeeching","type":"user"},{"_id":"5dd96eb166059660ed1ee413","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5dd96eb166059660ed1ee413/NQtzmrDdbG0H8qkZvRyGk.jpeg","isPro":true,"fullname":"Julien Chaumond","user":"julien-c","type":"user"},{"_id":"62543749b777cd32720675c2","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62543749b777cd32720675c2/EF_KRZO4hTo8TWXOtvc-n.png","isPro":false,"fullname":"Irene Solaiman","user":"irenesolaiman","type":"user"},{"_id":"5ff8c9f4b2035d9a81a859f7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1652134289581-5ff8c9f4b2035d9a81a859f7.jpeg","isPro":false,"fullname":"Nouamane Tazi","user":"nouamanetazi","type":"user"},{"_id":"608aabf24955d2bfc3cd99c6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/608aabf24955d2bfc3cd99c6/-YxmtpzEmf3NKOTktODRP.jpeg","isPro":true,"fullname":"Aritra Roy Gosthipaty","user":"ariG23498","type":"user"},{"_id":"60c757ea5f9a76ab3f844f12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1626214544196-60c757ea5f9a76ab3f844f12.png","isPro":false,"fullname":"Margaret Mitchell","user":"meg","type":"user"},{"_id":"63e0eea7af523c37e5a77966","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1678663263366-63e0eea7af523c37e5a77966.jpeg","isPro":true,"fullname":"Nathan Habib","user":"SaylorTwift","type":"user"}],"userCount":18,"collections":[{"slug":"science/releases-6823627145cf63aec03686ae","title":"🔥 Releases","description":"Hugging Face Science team releases","gating":false,"lastUpdated":"2025-09-09T12:30:00.312Z","owner":{"_id":"672926c86c4d0ec128d7496c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5e48005437cb5b49818287a5/oges-i7Dd9Fs0FDrh3bEs.png","fullname":"Hugging Face Science","name":"science","type":"org","isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":287,"isUserFollowing":false},"items":[{"_id":"68c01db551949bb6f385b87f","position":0,"type":"collection","id":"68bd02d20928419c1dc12296","slug":"HuggingFaceFW/finepdfs-68bd02d20928419c1dc12296","title":"📄 FinePDFs","description":"","lastUpdated":"2026-01-09T22:18:36.716Z","numberItems":82,"owner":{"_id":"657882cd90df9d85fb6be8df","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62596f9e1c0a084224b93e00/1NVyAAh-WfULT4i_LWZNB.png","fullname":"FineData","name":"HuggingFaceFW","type":"org","isHf":false,"isHfAdmin":false,"isMod":false,"plan":"team","followerCount":1383,"isUserFollowing":false},"theme":"green","shareUrl":"https://hf.co/collections/HuggingFaceFW/finepdfs","upvotes":27,"isUpvotedByUser":false},{"_id":"682362cd8c856061acb55b69","position":1,"type":"collection","id":"67ab6b5e84bf8aaa60cb17c7","slug":"HuggingFaceTB/smolvlm2-smallest-video-lm-ever-67ab6b5e84bf8aaa60cb17c7","title":"SmolVLM2 📺 Smallest video LM ever 🤏🏻","description":"","lastUpdated":"2025-05-05T16:18:41.847Z","numberItems":11,"owner":{"_id":"65116f902408819fa5329c6f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/651e96991b97c9f33d26bde6/e4VK7uW5sTeCYupD0s_ob.png","fullname":"Hugging Face Smol Models Research","name":"HuggingFaceTB","type":"org","isHf":false,"isHfAdmin":false,"isMod":false,"plan":"team","followerCount":3354,"isUserFollowing":false},"theme":"purple","shareUrl":"https://hf.co/collections/HuggingFaceTB/smolvlm2-smallest-video-lm-ever","upvotes":106,"isUpvotedByUser":false},{"_id":"682362da4ebeba21adf551a8","position":2,"type":"space","author":"nanotron","authorData":{"_id":"65af989ede38fbe922da644e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5df7e9e5da6d0311fd3d53f9/qAagSltOINhPaSgZe7roz.png","fullname":"Nanotron Research","name":"nanotron","type":"org","isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":835,"isUserFollowing":false},"colorFrom":"yellow","colorTo":"purple","createdAt":"2024-06-18T17:12:28.000Z","emoji":"🌌","id":"nanotron/ultrascale-playbook","lastModified":"2025-08-21T12:03:51.000Z","likes":3698,"pinned":true,"private":false,"sdk":"static","repoType":"space","runtime":{"stage":"RUNNING","hardware":{"current":null,"requested":null},"storage":null,"replicas":{"requested":1,"current":1}},"shortDescription":"The ultimate guide to training LLM on large GPU Clusters","title":"The Ultra-Scale Playbook","isLikedByUser":false,"ai_short_description":"Read the Ultra‑Scale Playbook on training LLMs","ai_category":"Other","trendingScore":12,"tags":["static","region:us"],"featured":false},{"_id":"682362ec402b31dec0cf9196","position":3,"type":"dataset","author":"open-r1","downloads":9708,"gated":false,"id":"open-r1/OpenR1-Math-220k","lastModified":"2025-02-18T11:45:27.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":450258,"libraries":["datasets","dask","polars","mlcroissant"],"formats":["parquet"],"modalities":["text"]},"private":false,"repoType":"dataset","likes":713,"isLikedByUser":false,"isBenchmark":false}],"position":0,"theme":"orange","private":false,"shareUrl":"https://hf.co/collections/science/releases","upvotes":2,"isUpvotedByUser":false},{"slug":"science/papers-68236d6ce46091bb768ae820","title":"📝 Papers","description":"Hugging Face Science team papers","gating":false,"lastUpdated":"2025-09-12T16:07:41.752Z","owner":{"_id":"672926c86c4d0ec128d7496c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5e48005437cb5b49818287a5/oges-i7Dd9Fs0FDrh3bEs.png","fullname":"Hugging Face Science","name":"science","type":"org","isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":287,"isUserFollowing":false},"items":[{"_id":"685debc24991716944278525","position":0,"type":"paper","id":"2506.20920","title":"FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data\n Processing to Every Language","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2506.20920.png","upvotes":77,"publishedAt":"2025-06-26T01:01:47.000Z","isUpvotedByUser":false},{"_id":"68236e093faba8afbbfa6b23","position":1,"type":"paper","id":"2504.05299","title":"SmolVLM: Redefining small and efficient multimodal models","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2504.05299.png","upvotes":205,"publishedAt":"2025-04-07T17:58:57.000Z","isUpvotedByUser":false},{"_id":"68236e8fda8c7fd4ce83aea9","position":2,"type":"paper","id":"2504.01833","title":"YourBench: Easy Custom Evaluation Sets for Everyone","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2504.01833.png","upvotes":22,"publishedAt":"2025-04-02T15:40:24.000Z","isUpvotedByUser":false},{"_id":"68236e02f678c559394e5b28","position":3,"type":"paper","id":"2502.02737","title":"SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2502.02737.png","upvotes":254,"publishedAt":"2025-02-04T21:43:16.000Z","isUpvotedByUser":false}],"position":1,"theme":"indigo","private":false,"shareUrl":"https://hf.co/collections/science/papers","upvotes":4,"isUpvotedByUser":false}],"datasets":[],"models":[],"paperPreviews":[],"spaces":[],"buckets":[],"numBuckets":0,"numDatasets":0,"numModels":0,"numSpaces":1,"lastOrgActivities":[{"time":"2026-02-13T21:24:27.469Z","user":"lewtun","userAvatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1594651707950-noauth.jpeg","type":"paper-daily","paper":{"id":"2602.12176","title":"Single-minus gluon tree amplitudes are nonzero","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2602.12176.png","upvotes":7,"publishedAt":"2026-02-12T17:09:06.000Z","isUpvotedByUser":true}},{"time":"2026-02-12T15:51:44.252Z","user":"lewtun","userAvatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1594651707950-noauth.jpeg","type":"paper-daily","paper":{"id":"2602.03773","title":"Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RL","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2602.03773.png","upvotes":9,"publishedAt":"2026-02-03T17:34:04.000Z","isUpvotedByUser":true}},{"time":"2026-01-30T16:49:59.657Z","user":"julien-c","userAvatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5dd96eb166059660ed1ee413/NQtzmrDdbG0H8qkZvRyGk.jpeg","type":"paper-daily","paper":{"id":"2601.21571","title":"Shaping capabilities with token-level data filtering","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2601.21571.png","upvotes":27,"publishedAt":"2026-01-29T11:34:01.000Z","isUpvotedByUser":true}}],"acceptLanguages":["*"],"canReadRepos":false,"canReadSpaces":false,"blogPosts":[],"currentRepoPage":0,"filters":{},"paperView":false}">AI & ML interests
None defined yet.
Recent Activity
View all activity
Organization Card
Hugging Face Research
The science team at Hugging Face is dedicated to advancing machine learning research in ways that maximize value for the whole community.
🛠️ Tooling & Infrastructure
The foundation of ML research is tooling and infrastructure and we are working on a range of tools such as datatrove, nanotron, TRL, LeRobot, and lighteval.
📑 Datasets
High quality datasets are the powerhouse of LLMs and require special care and skills to build. We focus on building high-quality datasets such as no-robots, FineWeb, The Stack, and FineVideo.
🤖 Open Models
The datatsets and training recipes of most state-of-the-art models are not released. We build cutting-edge models and release the full training pipeline as well fostering more innovation and reproducibility, such as Zephyr, StarCoder2, or SmolLM3.
🌸 Collaborations
Research and collaboration go hand in hand. That's why we like to organize and participate in large open collaborations such as BigScience and BigCode, as well as lots of smaller partnerships such as Leaderboards on the Hub.
⚙️ Infrastructre
The research team is organized in small teams with typically <4 people and the science cluster consists of 96 x 8xH100 nodes as well as an auto-scalable CPU cluster for dataset processing. In this setup, even a small research team can build and push out impactful artifacts.
📖 Educational material
Besides writing tech reports of research projects we also like to write more educational content to help newcomers get started to the field or practitioners. We built for example the alignment handbook, the evaluation guidebook, the pretraining tutorial, or the FineWeb blog.
Release Timeline
This is the release timeline so far and follow the links by clicking on the elements:
JAN 2025
JAN 2024
🔥 Warming up
🤗 Join us!
We are actively hiring for both full-time and internships. Check out hf.co/jobs
Hugging Face Science team papers
-
FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language
Paper • 2506.20920 • Published • 77 -
SmolVLM: Redefining small and efficient multimodal models
Paper • 2504.05299 • Published • 205 -
YourBench: Easy Custom Evaluation Sets for Everyone
Paper • 2504.01833 • Published • 22 -
SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
Paper • 2502.02737 • Published • 254
Hugging Face Science team releases
Hugging Face Science team papers
-
FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language
Paper • 2506.20920 • Published • 77 -
SmolVLM: Redefining small and efficient multimodal models
Paper • 2504.05299 • Published • 205 -
YourBench: Easy Custom Evaluation Sets for Everyone
Paper • 2504.01833 • Published • 22 -
SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
Paper • 2502.02737 • Published • 254
models
0
None public yet
datasets
0
None public yet