dalao.ai

AI news

News about Hugging Face

See everything

Hugging Face ships LeRobot 0.7.0 with live text control of robots and streaming training on huge datasets

Hugging Face's open-source robotics library LeRobot released version 0.7.0 on November 5. Operators can change instructions, ask what the robot sees, or hand it a goal while it keeps moving, without stopping the run. Training can stream datasets from the Hub, object storage, or LanceDB instead of downloading them first—the Hub now holds nearly 300 TB of open Physical AI data in LeRobot format. The release also adds remote inference so large policies can run on a GPU server while the robot stays light, plus new world-action and vision-language-action models and a training stack aimed at multi-billion-parameter policies.

Why it matters It turns live language control and disk-less training into default open tooling, so robotics teams can reuse the stack instead of building orchestration from scratch.

Hugging Face
Open source

Hugging Face publishes Ultimate MoE Training Handbook after about 500,000 GPU-hours of work

On November 5 Hugging Face’s Nouamane Tazi, Leandro von Werra and Thomas Wolf published The Ultimate MoE Training Handbook. They say about 500,000 GPU-hours went into optimizing mixture-of-experts training, distilled into roughly 30,000 words and more than 50 interactive figures—from a first single-node run through routers, load balancing and cluster communication. Readers can follow recipes to train a small MoE with about 6.86 billion total parameters and roughly 1.22 billion active per token. The handbook lives on a Hugging Face Space.

Why it matters Few open resources walk through MoE training settings end to end; teams moving from dense models get a concrete playbook they can run.

Hugging Face
Open source

Hugging Face proposes Model Integration Contracts for Transformers, using shrunk models to catch regressions on CPU

Hugging Face proposed transformers-mic, packaging model tests as Model Integration Contracts (MICs): a shrunk model, a runnable usage example and human-reviewed baseline outputs, run on CPU. Replaying five months of commits, a 66 MB UMT5 fixture (source checkpoint 1,227 MB) flagged a bug that sat in main for almost two months: with SDPA enabled, UMT5's decoder could attend to future tokens.

Why it matters Transformers supports over 750 model types and can't run its full GPU tests on every PR; this lets regressions be caught on CPU at the PR stage.

Hugging Face
Product updatesoriginally

US Justice Department makes inquiries into Nvidia's $12.93 billion deal to buy Hugging Face

Reuters reported on Oct 28, citing two people familiar with the matter, that the US Justice Department has asked how Nvidia's $12.93 billion deal for open-source AI platform Hugging Face would affect competition, including whether it would give Nvidia a key distribution point for AI. CNBC later confirmed it. Nvidia declined to comment; the deal is expected to close in the first half of 2027.

Why it matters Hugging Face is the main hub for open models and datasets; the review could affect whether the deal closes and on what terms developers and rival chip and cloud providers keep using the platform.

NVIDIAHugging Face
Funding & companiesoriginally