Big events bring big moves. WAIC, one of China’s top AI events, starts tomorrow. Got a feeling we’ll see a wave of new models and fresh AI moves in the coming days👀

With the latest release, I want to make sure I get this message to the community: we are listening! Hugging Face we are very ambitious and we want `transformers` to accelerate the ecosystem and enable all hardwares / platforms! Let's build AGI together 🫣 Unbloat and Enable!

One of the early SmolLM3 checkpoints had super weird vibes despite scoring well on benchmarks. Turned out we had a bug in our data processing script that nuked all system messages 😵‍💫

New job title: Cowboy Post-trainer

works immediately out of the box. Hugging Face and the open llm ecosystem don't get enough credit for this.

Merry Christmas 🎁 GSPO is in TRL. Looking forward to see your reward curves 📈

College is about socialization — most of my social & musical life came from college. Wouldn’t trade that for any startup opportunity. Life != your career.

Introducing GLM-4.5 and GLM-4.5 Air: new flagship models designed to unify frontier reasoning, coding, and agentic capabilities. GLM-4.5: 355B total / 32B active parameters GLM-4.5-Air: 106B total / 12B active parameters API Pricing (per 1M tokens): GLM-4.5: $0.6 Input / $2.2

SmolLM3 now has an EU Summary of Public Content!!! 🇪🇺🤗 The Hugging Face SmolLM3 model is one of the strongest models its size while being fully open (incl. data!); it's now also (one of?) the first model to come with the AI Act-mandated summary of training content... 🧵👇1/4

Congrats to the E2B team 🔥!

The secret superpower in ML of just looking at the data a lot

The new Hugging Face jobs CLI is powered by uv 🤗 You can use `hf jobs uv run` to initiate a job from a standalone Python script.

GSPO is also now available in TRL: pip install git+github.com/huggingface/tr…

📢 TRL 0.20 drops: Fine-tune your VLM with GRPO! And it also includes GSPO. So basically, fine-tune your VLM with GSPO.

We just released the training and evaluation code to reproduce SmolLM3 🔥 🏋️‍♀️ Pretraining scripts (nanotron) 🤠 Post-training code for mid-training + SFT + APO (TRL/alignment-handbook) 👩‍⚖️ Evaluation scripts to reproduce all reported metrics github.com/huggingface/sm… We're

Today, we’re officially releasing the weights for AFM-4.5B and AFM-4.5B-Base on HuggingFace. This is a major milestone for Arcee.ai. AFM is designed to be flexible and high-performing across a wide range of deployment environments.

Life is so much better when your grad norms aren't NaNs

Like a LOT better