Thomas Wolf

Thomas Wolf

@thomwolf.bsky.social

Co-founder @huggingface

12.8K followers1.4K following122 posts

EU–INC is the single best thing Europe could do to catch-up in the AI race A simple unified pan-European startup structure, with modern employee ownership and simple access to capital, able to tap into Europe’s full talent pool. ‼️ but it’s at high risk of not seeing the light of day. You can help👇

View full thread

EU–INC is the single best thing Europe could do to catch-up in the AI race A simple unified pan-European startup structure, with modern employee ownership and simple access to capital, able to tap into Europe’s full talent pool. ‼️ but it’s at high risk of not seeing the light of day. You can help👇

🦄This is exactly what ambitious founders need to start and scale and catch up with the rest of the world: - A single new pan-European legal entity - A single central EU-level registry - Standardized investment documents - Standardized EU-wide stock options - Local taxes & employment

Reachy-Mini update🚨 Good: quality up++, sound/elec better than alpha. First tiny batches ship late Sep. Less good: scaling to thousands units (vs. 100s planned) taking longer than expected Still on track to ship first 3000 pre-orders before Xmas🎄(Lite & Wireless), most shipping early Dec.

Thomas Wolf reposted

This is a really good blog post about how modern speech AI models work, highly recommended ⬇️ #MLSky

Thomas Wolf@thomwolf.bsky.social

[Past weekends side project] A quick introduction to contextual speech models I made some interactive visualisations to explain a bit better how recent contextual speech models and audio tokenization work (e.g. Sesame's demo).

[Past weekends side project] A quick introduction to contextual speech models I made some interactive visualisations to explain a bit better how recent contextual speech models and audio tokenization work (e.g. Sesame's demo).

At Hugging Face—in robotics and across all AI—we believe in a future where AI and robots are open-source and affordable; community-built and safe; hackable and fun. We've had so much understanding and passion working with Pollen Robotics over the past year that we decided to join forces! ---

You can already find our open-source humanoid robot platform Reachy 2 on the Pollen website and we're excited to build and share more open-source robots with the world in the coming months. I spoke with @jeremyakahn.bsky.social about it in Fortune: fortune.com/2025/04/14/a...

fortune.comHugging Face gets into the humanoid robot game with Pollen Robotics purchaseThe acquisition will see Hugging Face, best known for open-source AI models, selling hardware for the first time

Fascinating example of applied research! No. 1 question we get with +1.5M models on the hub: how to find the best one for your use-case? Here comes YourBench 🪄 a research work led by Sumuk Shashidhar, Alina Lozovskaya and team. ---

From a single or a few document examples, it automatically generates a custom eval with questions and test a series of models on it to find the best fit for your field! Cherry on the cake: it's all open-source and open-science with full publication, code, and more. ---

Wow - open-source LLMs for code are so back Just one week after the latest DeepSeek release, here is the All-Hands team dropping a 32B model with matching performance on software engineering tasks benchmarks (SWE-Bench). 20 times smaller than the DeepSeek model --

Results obtained by post-training w. reinforcement learning in a code environment: remember the 2025 trend of reasoning models in math/code/etc trained with RL - yep, it strikes again Feels like the open-source world starts to crack the secret code generation sauce of closed models like Claude --

We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1 And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder [1/3]

It's beating Claude 3.7 on (competitive) programming –a domain Anthropic has been historically really strong at– and it's getting close to o1-mini/R1 on olympiad level coding with just 7B parameters! [2/3]

Thomas Wolf reposted

Replying to @valeoai.bsky.social

We’re just scratching the surface of the "big models + raw video + closed-loop sim" stack. More to come! 🚀 Retweet if you’re excited, and follow @valeoai.bsky.social for updates! 💚 🙏 Thanks for reading! [10/10]

Thomas Wolf reposted

LLM Reasoning labs will be eating good today🍔 We commandeered the HF cluster for a few days and generated 1.2M reasoning-filled solutions to 500k NuminaMath problems with DeepSeek-R1 🐳 Have fun!

Thomas Wolf reposted

OK! My Google colleague Thang Luong shared some exciting updates about AlphaGeometry2! AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% compared to 54% previously! Paper: arxiv.org/abs/2502.03544 See full list of authors on link

Gus@gusthema.bsky.social

Hey @jeffdean.bsky.social , help bring posts like the AlphaGeometry2 one to this network please! Using the other place as the only place is bad for the field!

From an open-research point of view, maybe the greatest thing about DeepSeek–R1 is how its RL training technique appears so straightforward/simple in comparison to the cumbersome approaches we were starting to think necessary for reasoning like Process Reward Models or Monte Carlo Tree Search. [1/2]

Once again it's party-time for the AI community which just discovered a whole new land of possibilities for building smarter AI models, and distillation in smaller models is one of the most exciting of these new promises. [2/2]

We've just released the new Spaces search and it's quite mind blowing Explore over 400k AI Apps in the most intuitive way background removal, image-to-3D, comic factory, sound transcription, image editing, clothes virtual try-on, etc All made by AI builders for AI builders huggingface.co/spaces

I wrote some reflections on DeepSeek, open-source, AI, US and China, starting from Dario's recent essay calling for stronger export controls. I mostly disagree with his essay and think it missed the point You can read it here: thomwolf.io/blog/deepsee...

thomwolf.io🐳 Some notes on "DeepSeek and export control"Finally took time to go over Dario's essay on DeepSeek and export control and wrote some notes. I mostly disagree and I think it missed the point.

Every time you declare open-source AI dead another team rise up like a Whac-A-Mole. Intelligence will be commoditized! The remaining moat will be the quality of the product and how it integrate with co-factors. That’s were billions dollars AI companies will be built.