iiinova — where your digital self keeps evolving

Thomas Wolf @thomwolf.bsky.social · 27/09/2025

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👇

44713

Thomas Wolf @thomwolf.bsky.social · 17/09/2025

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.

1231

Thomas Wolf @thomwolf.bsky.social · 10/09/2025

first test in plugging the $100 Amazing Hands on a (cute) humanoid instead of grippers - pretty happy about it Hand: github.com/pollen-robot... Humanoid: www.pollen-robotics.com/reachy/

3338

Thomas Wolf @thomwolf.bsky.social · 09/09/2025

oh my god this is just crazy folks github.com/huggingface/...

1679

Thomas Wolf @thomwolf.bsky.social · 07/09/2025

wow, total BoM cost $660, folks open-source community >> closed source hyped robots Read more at Xlerobot.readthedocs.io

3396

Thomas Wolf @thomwolf.bsky.social · 29/05/2025

London's robot conference is going crazy over it! Today we're unveiling our 1st robot Hugging Face 🤝 Pollen Robotics A low-cost $250 open-source robot designed as an open-source platform for fun human computer interactions powered by HF Spaces, models and community. => discord.com/invite/jsvMR...

0427

Thomas Wolf @thomwolf.bsky.social · 19/05/2025

We've seen nothing yet! We hosted a 9-13 yo vibe-coding event with @robertkeus.bsky.social this weekend (h/t @antonosika.bsky.social and Lovable) takeaway? AI is unleashing a generation of wildly creative builders beyond anything I'd have imagined and they grow up knowing they can build anything!

2314

Thomas Wolf @thomwolf.bsky.social · 28/04/2025

[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).

1325

Thomas Wolf @thomwolf.bsky.social · 22/04/2025

being in a robotics lab in 2025 open-source hardware + good RL + impressive people => lot of fun join the OpenDuck community to build yours: discord.com/invite/UtJZs...

2412

Thomas Wolf @thomwolf.bsky.social · 04/04/2025

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. ---

2357

Thomas Wolf @thomwolf.bsky.social · 31/03/2025

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 --

77717

Thomas Wolf @thomwolf.bsky.social · 12/03/2025

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]

39918

Thomas Wolf @thomwolf.bsky.social · 10/03/2025

We're super excited to introduce the Academia Hub! A game-changer for school, university, courses, research, projects, collaborations, publications. Learn more: huggingface.co/docs/hub/aca... It's a per-school offer, not individual => your faculty/admin have to apply Go schools, go universities!

43610

Thomas Wolf @thomwolf.bsky.social · 06/03/2025

I shared a controversial take the other day at an event and I decided to write it down in a longer format: I’m afraid AI won't give us a "compressed 21st century" Here: thomwolf.io/blog/scienti... It's an extension of this interview discussion from the AI summit: youtu.be/AxBd3G0lFLs?...

1113234

Thomas Wolf @thomwolf.bsky.social · 28/02/2025

Oh! My TED talk is just out. I talk about how open-source is critical for AI safety and humanity's resilience. Yes, you don't hear that perspective often! :) It's called "What if AI just works" It's here: www.ted.com/talks/thomas...

1499

Thomas Wolf @thomwolf.bsky.social · 27/02/2025

A few words on DeepSeek new releases. Links are: - github.com/deepseek-ai/... - github.com/deepseek-ai/... - github.com/deepseek-ai/... and the Ultra-Scale Playbook at huggingface.co/spaces/nanot...

0515

Thomas Wolf @thomwolf.bsky.social · 25/02/2025

Hugging Face just entered the top 10 organizations on @github.com Close to 500,000 GitHub stars across our open-source libraries! Couldn't be more proud of what this 220-person team is accomplishing

1595

Thomas Wolf @thomwolf.bsky.social · 05/02/2025

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

1458

Thomas Wolf @thomwolf.bsky.social · 03/02/2025

« appending "Wait" multiple times to the model's generation » is our current most likely path to AGI :) See the fresh arxiv.org/abs/2501.19393 by Niklas Muennighoff et al.

06212

Thomas Wolf @thomwolf.bsky.social · 21/01/2025

The AI agent course keeps breaking records! Now breaking discord registration system You can still join at bit.ly/hf-learn-age... (let’s test discord infrastructure hahah)

0272

Thomas Wolf @thomwolf.bsky.social · 15/01/2025

Mind blown 👇 when people ask whether you need an agent framework at all! All evals should move to agentic evals in 2025 in my opinion. We’re just leaving so much capabilities of our models on the table. Benchmarked with smolagents: github.com/huggingface/...

2609

Thomas Wolf @thomwolf.bsky.social · 06/01/2025

New LLM evals entering and leaving the field saturated before the paper is even published in ML conferences 📸 matt_mccreary1

2609

Thomas Wolf @thomwolf.bsky.social · 03/01/2025

4 out of the 6 top trending repos on GitHub are from Chinese AI teams today 2025 is gonna look quite different from 2024

2404

Thomas Wolf @thomwolf.bsky.social · 01/01/2025

Our first release of 2025: 𝙨𝙢𝙤𝙡𝙖𝙜𝙚𝙣𝙩𝙨, 𝘁𝗵𝗲 𝘀𝗶𝗺𝗽𝗹𝗲𝘀𝘁 𝗹𝗶𝗯𝗿𝗮𝗿𝘆 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝘀𝘆𝘀𝘁𝗲𝗺𝘀! 💥 Main logic in ~1000 LoC 🧑‍💻 Agent writes its actions in code! LLMs are much better at writing code than current standard of writing JSON => higher perf 🌍 Any LLM support (h/t LiteLLM) 🛡️ Secure code exec (h/t E2B)

412318

Thomas Wolf @thomwolf.bsky.social · 23/12/2024

With the new OpenAI O3 moving performance from 5% up to 25% on FrontierMath it’s time to push open-source models upwards! We're super happy to release FineMath, the best open math dataset yet. A strong baseline to start training your own models Find it in the trending section of HuggingFace ;)

0599

Thomas Wolf @thomwolf.bsky.social · 14/12/2024

2025 will be the year of AI for science Leveraging all the things we've recently learned training AI models for 1000x impact in science and this will need data! More details: huggingface.co/blog/lemater... Thread: ...

26211

Thomas Wolf @thomwolf.bsky.social · 11/12/2024

The Open LLM Leaderboard got a new front page for Christmas Check it out at huggingface.co/spaces/open-...

26612

Thomas Wolf @thomwolf.bsky.social · 10/12/2024

When you’ve finished your day of emails

2360

Thomas Wolf @thomwolf.bsky.social · 09/12/2024

Four new visualisations of the rise of open-source AI models in 2024 added in our daily nuggets! - explore how various AI tasks have been growing - how community likes connect AI models together - the geography of models creators and followers Explore them here: huggingface.co/spaces/huggi...

1352

Thomas Wolf @thomwolf.bsky.social · 08/12/2024

The FineWeb team is happy to finally release "FineWeb2" 🥂🥳 FineWeb 2 extends the data driven approach to pre-training dataset design that was introduced in FineWeb 1 to now covers 1893 languages/scripts Details: huggingface.co/datasets/Hug... A detailed open-science tech report is coming soon

310613

Thomas Wolf @thomwolf.bsky.social · 06/12/2024

Exponentially growing number of open-source AI models over the course of the past 30 months – from a few thousands to over 1 million and more Interactive data viz: huggingface.co/spaces/huggi...

3449

Thomas Wolf @thomwolf.bsky.social · 04/12/2024

Top organizations releasing open-source AI models over the course of 2024 (cumulative likes) Interactive viz: huggingface-open-source-ai-year-in-review-2024.static.hf.space/index.html?d... Discussion: huggingface.co/spaces/huggi...

0313

Thomas Wolf @thomwolf.bsky.social · 04/12/2024

Most liked and most downloaded open-source AI models from 2022 to 2024 Interactive viz: aiworld.eu/embed/model/... Discussion: huggingface.co/spaces/huggi...

28520

Thomas Wolf @thomwolf.bsky.social · 26/11/2024

Releasing SmolVLM, a small 2 billion parameters Vision+Language Model (VLM) built for on-device/in-browser inference with images/videos. Outperforms all models at similar GPU RAM usage and tokens throughputs Blog post: huggingface.co/blog/smolvlm

423131

Thomas Wolf @thomwolf.bsky.social · 25/11/2024

A lot of gems and insights in this fresh unreleased slide-deck by Niels Rogge (well now it's released hahah - sorry Niels) => docs.google.com/presentation...

3667

Thomas Wolf @thomwolf.bsky.social · 24/11/2024

One other cool discovery was that the model could perfectly format numbers as R2L with just a few R2L examples, even when trained on L2R! Showing that these models aren’t just memorizing patterns from training data but are able to perform even better than the data it was originally trained on.

2250

Thomas Wolf @thomwolf.bsky.social · 24/11/2024

Exploring addition tasks with varying digit lengths, he found that R2L tokenization has better overall performance compared to the existing L2R tokenizer. Digits that are multiples of 3 have the same three-digit tokens so both L2R and R2L tokenization have the exact same results.

1131

Thomas Wolf @thomwolf.bsky.social · 24/11/2024

He wanted to test how models performed when subjected to a new tokenization scheme without re-training the model. Modifying Llama 3’s tokenizer to tokenize numbers from right to left (R2L) instead of left-to-right (L2R) with just a few lines of code 🧑‍💻. This affect the grouping of numbers by three:

1191

Thomas Wolf @thomwolf.bsky.social · 24/11/2024

It's Sunday morning so taking a minute for a nerdy thread (on math, tokenizers and LLMs) of the work of our intern Garreth By adding a few lines of code to the base Llama 3 tokenizer, he got a free boost in arithmetic performance 😮 [thread]

527134

Thomas Wolf @thomwolf.bsky.social · 22/11/2024

Excited to release SmolTalk: the secret recipe behind the best-in-class performance of SmolLM2 hf.co/datasets/Hug... The SmolLM team at Hugging Face curated new synthetic datasets to enhance instruction following and reasoning. SmolTalk also supports rewriting, summarization, and function calling.

61324

Thomas Wolf @thomwolf.bsky.social · 22/11/2024

Spoke at Slush this week. What a great conference for startup entrepreneurs! Had a blast

2230

Thomas Wolf @thomwolf.bsky.social · 26/10/2023

Software Heritage Foundation has just published a very interesting statement Context: you're a non-profit dedicated to preserving data, in this case data being all code ever written 🤯 Q: what should you do in today's LLM world? Close the doors to protect creators, open the doors to allow training?

131

Thomas Wolf @thomwolf.bsky.social · 25/10/2023

one of the most striking event of 2023: the rise of the GPU-poor models you can do a lot (and much cheaper) with well trained smaller models

150

Thomas Wolf @thomwolf.bsky.social · 18/09/2023

These new phi-1 & 1.5 models are fascinating ✅ performances crushing 10 times bigger models 😱 ✅ secret sauce coming from a ✨magic✨ textbook dataset 🕵️‍♂️ close to zero information on this dataset other than GPT3.5 generated ⚗️ Is it time for an "Open Textbook" project??

030

Thomas Wolf @thomwolf.bsky.social · 06/09/2023

TII just released Falcon 180B, a new LLM that beats OpenAI GPT-3.5 and is on par with Google's PaLM-2, making it one of the top 3 most capable LLMs publicly known. 👑 Falcon 180B is 2.5x larger than Llama 2. It was trained with 4x more computing power 😱 Read more here huggingface.co/blog/falcon-...

030