OpenEnv has a new home: github.com/huggingface/OpenEnv
Starting today, it's coordinated by a committee that includes Meta-PyTorch, Reflection, Unsloth, Modal, Prime Intellect, Nvidia, Mercor, Fleet AI, and Hugging Face
frontier labs train their models and their harnesses together. Claude knows Claude Code. GPT-5.5 knows Codex. that's not an accident, it's training. open-source models deserve the same magic, but pulling that off requires infrastructure that belongs to everyone, not one lab
OpenEnv is that layer. one api, any harness, any trainer, any environment
Rewards and training loops stay in TRL, Unsloth, wherever you already work. OpenEnv is the socket they all plug into
DeepSeek R1 dropped one year ago 🐳 and a lot has changed.
With @irenesolaiman , we’re launching a blog series about how that moment reshaped AI + open source in 2025, starting with strategic shifts and the explosion of new open models in China!
fine-tuning a 14B model with TRL + SFT on a free Colab (T4 GPU)? thanks to the latest TRL optimizations, you actually can! sharing a new notebook showing how to do it 😎
AI for Scientific Discovery Won't Work Without Fixing How We Collaborate.
My co-author @cgeorgiaw and I just published a paper challenging a core assumption: that the main barriers to AI in science are technical. They're not. They're social.
Key findings:
🚨 The "AI Scientist" myth delays progress: Waiting for AGI devalues human expertise and obscures science's real purpose: cultivating understanding, not just outputs. 📊 Wrong incentives: Datasets have 100x longer impact than models, yet data curation is undervalued. ⚠️ Broken collaboration: Domain scientists want understanding. ML researchers optimize performance. Without shared language, projects fail. 🔍 Fragmentation costs years: Harmonizing just 9 cancer files took 329 hours.
Why this matters: Upstream bottlenecks like efficient PDE solvers could accelerate discovery across multiple sciences. CASP mobilized a community around protein structure, enabling AlphaFold. We need this for dozens of challenges.
Thus, we're launching Hugging Science! A global community addressing these barriers through collaborative challenges, open toolkits, education, and community-owned infrastructure. Please find all the links below!
Smol course has a distinctive approach to teaching post-training, so I'm posting about how it’s different to other post-training courses, including the llm course that’s already available.
In short, the smol course is just more direct that any of the other course, and intended for semi-pro post trainers.
- It’s a minimal set of instructions on the core parts. - It’s intended to bootstrap real projects you're working on. - The material handsover to existing documentation for details - Likewise, it handsover to the LLM course for basics. - Assessment is based on a leaderboard, without reading all the material.
ModernBERT goes MULTILINGUAL! One of the most requested models I've seen, The Johns Hopkins University's CLSP has trained state-of-the-art massively multilingual encoders using the ModernBERT architecture: mmBERT.
Model details: - 2 model sizes: - jhu-clsp/mmBERT-small - jhu-clsp/mmBERT-base - Uses the ModernBERT architecture, but with the Gemma2 multilingual tokenizer (so: flash attention, alternating global/local attention, unpadding/sequence packing, etc.) - Maximum sequence length of 8192 tokens, on the high end for encoders - Trained on 1833 languages using DCLM, FineWeb2, and many more sources - 3 training phases: 2.3T tokens pretraining on 60 languages, 600B tokens mid-training on 110 languages, and 100B tokens decay training on all 1833 languages. - Both models are MIT Licensed, and the full datasets and intermediary checkpoints are also publicly released
Evaluation details: - Very competitive with ModernBERT at equivalent sizes on English (GLUE, MTEB v2 English after finetuning) - Consistently outperforms equivalently sized models on all Multilingual tasks (XTREME, classification, MTEB v2 Multilingual after finetuning) - In short: beats commonly used multilingual base models like mDistilBERT, XLM-R (multilingual RoBERTa), multilingual MiniLM, etc. - Additionally: the ModernBERT-based mmBERT is much faster than the alternatives due to its architectural benefits. Easily up to 2x throughput in common scenarios.
Based on these results, mmBERT should be the new go-to multilingual encoder base models at 300M and below. Do note that the mmBERT models are "base" models, i.e. they're currently only trained to perform Mask Filling. They'll need to be finetuned for downstream tasks like semantic search, classification, clustering, etc.
