iiinova — where your digital self keeps evolving

Lewis Tunstall

@lewtun.bsky.social

🤗 LLM whisperer @huggingface 📖 Co-author of "NLP with Transformers" book 💥 Ex-particle physicist 🤘 Occasional guitarist 🇦🇺 in 🇨🇭

Posts Replies Media

Lewis Tunstall @lewtun.bsky.social · 21/05/2026

We hope the Carbon family of models will enable the broader community of computational biology have access to models that can literally run on your laptop. We are working on fine-tuning scripts that can run on small GPUs too and will share them soon!

000

Lewis Tunstall @lewtun.bsky.social · 21/05/2026

Carbon is built with a unique tokenizer: we split sequences in chunks of 6 bases, but during both training and inference we can work with single base resolution. The architecture combined with the tokenizer makes the model 275x faster than the previous SoTA (Evo2) at this size.

100

Lewis Tunstall @lewtun.bsky.social · 21/05/2026

We are releasing Carbon: a crazy fast DNA model Carbon is 275x faster than the next best model. So fast you can process the whole human genome on a single GPU in <2 days. We built a demo so you can explore how the model can generate DNA sequences and a lot more: huggingface.co/spaces/Huggi...

110

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

📊We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset. 🔎 Read our blog post for all the nitty gritty details: huggingface.co/blog/open-r1...

huggingface.co

Open R1: Update #2

A Blog post by Open R1 on Hugging Face

010

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

⏳ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that can’t be verified with a rules-based parser)

100

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

📀512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.

100

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

🐳 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.

100

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

What’s new compared to existing reasoning datasets? ♾ Based on NuminaMath 1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.

110

Lewis Tunstall @lewtun.bsky.social · 10/02/2025

Introducing OpenR1-Math-220k! huggingface.co/datasets/ope... The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch 💪 More details in 🧵

huggingface.co

open-r1/OpenR1-Math-220k · Datasets at Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

1163

Lewis Tunstall @lewtun.bsky.social · 25/01/2025

We are reproducing the full DeepSeek R1 data and training pipeline so everybody can use their recipe. Instead of doing it in secret we can do it together in the open! Follow along: github.com/huggingface/...

github.com

GitHub - huggingface/open-r1: Fully open reproduction of DeepSeek-R1

Fully open reproduction of DeepSeek-R1. Contribute to huggingface/open-r1 development by creating an account on GitHub.

620036

Lewis Tunstall @lewtun.bsky.social · 16/12/2024

Here's the links: - Blog post: huggingface.co/spaces/Huggi... - Code: github.com/huggingface/... Enjoy!

huggingface.co

Scaling test-time compute - a Hugging Face Space by HuggingFaceH4

Discover amazing ML apps made by the community

0160

Lewis Tunstall @lewtun.bsky.social · 16/12/2024

We outperform Llama 70B with Llama 3B on hard math by scaling test-time compute 🔥 How? By combining step-wise reward models with tree search algorithms :) We're open sourcing the full recipe and sharing a detailed blog post 👇

410921

Lewis Tunstall @lewtun.bsky.social · 12/12/2024

Hey ML peeps, we found a nice extension to beam search at Hugging Face that is far more scalable and produces more diverse candidates The basic idea is to split your N beams into N/M subtrees and then run greedy node selection in parallel Does anyone know what this algorithm is called?

060