Models & Labs
Hugging Face natively supports PyTorch on Google TPUs
Hugging Face Bloghigh confidence
Why it matters
- →Eliminates the need for torch_xla or JAX when running PyTorch on Google TPUs.
- →Enables direct model loading in Hugging Face Transformers and Diffusers via simple device mapping.
- →Provides significant inference speedups through native Pallas kernel integration.
Hugging Face and Google have announced native support for TPUs within the PyTorch ecosystem via TorchTPU, eliminating the need for the previous torch_xla workaround. This integration allows Hugging Face libraries like Transformers and Diffusers to run directly on TPU hardware using standard device mapping commands. The update includes optimized Pallas kernels that deliver substantial performance improvements, with tests showing up to 21.5x faster decoding speeds on Mixture-of-Experts models. While the implementation is still maturing with a 96.5% test pass rate, it marks a significant shift toward unified hardware support for PyTorch developers.
Read originalThe story around this
TopicHugging Face PyTorch And Swift
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More from Hugging Face Blog
More in Models & Labs
Models & Labsmodels
llama.cpp b12029 adds CUDA 13 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally supporting NVIDIA's latest CUDA 13 stack alongside AMD's ROCm 10.0. The inclusion of native builds for Linux arm64 with Snapdragon chips, covering CPU, Adreno GPU, and Hexagon NPU, signals a serious push into mobile edge deployment. While Apple Silicon KleidiAI is disabled in this specific build, the broader hardware coverage means developers no longer need to compile from source for most modern accelerators. It’s less about new features and more about removing friction for anyone running local models on bleeding-edge or diverse hardware.
llama.cpp Releases
© The Verge AIModels & Labsother
