Hugging Face ❤️ TorchTPU: TPUs, natively
Hugging Face published a post about TorchTPU, presented as a way to use TPUs natively. The headline signals support for the backend in the Hugging Face ecosystem.
Related stories
AnyLanguageModel 1.0: one year of local models on Apple platforms
Hugging Face marks version 1.0 of AnyLanguageModel, looking back on a year of running local models on Apple platforms.
MPS in PyTorch 2.15
A Hugging Face post covers the state of MPS, the Apple-silicon GPU backend, in PyTorch 2.15.
Mistral Large 4 trails Muse Spark Open by one point, but leads it on agent tests
On Artificial Analysis's index, Meta's Muse Spark Open scores 39 against 38 for the Mistral Large 4 preview, which still wins five of ten tests, including Terminal-Bench and AutomationBench. Large 4 generates far more tokens to run the index and costs more at launch pricing; Muse's licence restricts large-scale resale, while Large 4's licence is unpublished and its weights are promised by October 31.
Amazon's TorchNeuron lets PyTorch code run on AWS Trainium chips
Amazon has launched TorchNeuron, an open-source backend that lets PyTorch code run on its Trainium AI chips with a one-line change. It enters public beta on October 20 under the Apache 2.0 license, supports Trainium2 and Trainium3, and does not yet cover every PyTorch feature.
Nvidia launches new open models to run agents in the enterprise and on PCs
Nvidia has released new open models aimed at companies that want to run their own models in their own data centers. The launch also pushes private AI agents that run on the device.