Silicon neutrality in action! 🤗
With TorchTPU, @Google TPUs are now native @PyTorch devices, supported in 🤗 Transformers, Diffusers & Kernels.
Same familiar workflows. More freedom to choose your hardware.
Blog ⤵️
Agent Manager is finally out!
Orchestrate dozens of agents across harnesses (Claude Code, Codex, OpenCode etc.) effortlessly.
Agents communicate and self-organize to help you achieve complex tasks and always keep running on HF spaces.
Try it: huggingface.co/spaces/lvwerr…
Silicon neutrality means freedom to choose your hardware.
With @awscloud , we're bringing native AWS Trainium support to 🤗 Transformers, Diffusers, TRL & Kernels through TorchNeuron.
Same PyTorch code. Familiar workflows. More choice.
huggingface.co/blog/hf-torch…
Before: custom modeling classes, a dedicated trainer, lazy tensors and static shapes.
Now: eager mode, torch.compile, FSDP and TP, straight through our libraries.
89% of the top 100 text-generation architectures run out of the box, up from 46%.
Your training script isn’t tied to Trainium. Your checkpoint is a standard Transformers checkpoint you can push to the Hub and load on any hardware.
Neuron kernels from AWS are on the Hub at aws-neuron/kernels, with up to 2.4x speedups on a ViT encoder.
That’s what silicon neutrality means for builders: choice without lock-in.
Huge thanks to the AWS Neuron team!
Hugging Face’s kickoff last night was absolute FIRE 🔥
A giant mega duck. DJs. Food, drinks, and music. Great conversations everywhere—catching up with familiar faces and making new connections. What a night. 🦆
Huge props to @jeffboudier and the @huggingface team for putting together such a stellar event! we missed you @ClementDelangue.
Today, the ExecuTorch Hackathon is in full swing at Founders, Inc., with teams building across three tracks:
💻 Compute — on-device AI on Snapdragon X Elite-powered Copilot+ PCs
📱 Mobile + XR — AI experiences on Samsung Galaxy S25 Ultra and Galaxy XR
🛠️ IoT — edge AI, sensing, and control on Arduino VENTUNO Q boards
From the dance floor to building on real hardware. Open Source AI Week is starting with a real bang—and we’re only getting started. 🚀
Hackathon details:
luma.com/executorch-hackatho…
Thanks to hackathon sponsors:
@AIatMeta@Qualcomm@github@open_intel_fdn@fdotinc@hackersquadio@PyTorch#OpenSourceAIWeek#OpenIntelligence
What a night at Hugging Face's Open Together party in SF 🤗 1,000+ people celebrating open models!
Fun memory: back when Hugging Face was a tiny startup, I sat a few desks from them at Station F. Now they're the home of open-source AI. As someone who checks the Hub every day, being in a room full of fellow open-source fans was so exciting.
Friday night at @huggingface’s Open Together in SF. 🤗
Here to catch up on the latest in open-source AI and meet some of the people building it.
If you’re here, come say hi!
@_akhaliq where you at ?? 👀
I had an awesome time at the @huggingface party! I got to meet the 🤗 MCP maintainer @evalstate, dance with the 🤗 mascot, and meet 2 adorable microducks.
Tonight is the night. Today is the day AI Builders unite.
I hear MEGADUCK is coming. Will you?
Open Together, 6pm on 10/16 at The Midway, SF. Free.
luma.com/OpenTogether
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.
Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.
Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token context window
→ Natively multimodal across text, image, video, audio
Muse Spark Open is out, and TRL supports it from day 0 🤗
Fine-tune the 816B MoE (42B active) with SFTTrainer: transformers shards the model and splits the experts across GPUs at load time, and our example trains a LoRA adapter on 8 nodes of 8 H100s.