Google and Meta Open-Source TorchTPU for PyTorch on TPUs
Google and Meta’s PyTorch team released TorchTPU as an open-source native backend for running PyTorch workloads on Google Cloud TPUs. It supports standard PyTorch APIs, eager-mode debugging and torch.compile optimization, with integrations spanning Hugging Face Transformers and Diffusers, vLLM, SGLang, llm-d, TorchTitan and LiteRT. Google says its training setup was validated on up to 8,000 Ironwood TPUs and achieved as much as 52% model FLOPS utilization with Qwen3-8B; these figures are vendor claims.
TorchTPU gives developers a native route for migrating PyTorch workloads to Google’s TPU infrastructure while continuing to use familiar training, inference and deployment tools. It also extends Google’s AI software ecosystem beyond the TensorFlow and JAX frameworks historically associated with TPUs.
If you ever tried to run PyTorch on a TPU, you probably ported your code to torch_xla, or you ended up having to learn JAX. With TorchTPU, a native PyTorch backend, TPU is now just another device. It opens up native support of TPU in the Hugging Face ecosystem. Transformers, Diffusers and Kernels run natively on TPU. Pick a model from the Hub, send it to "tpu", and you're done.
Wait, what's a TPU? 🤔
TPU stands for Tensor Processing Unit. TPUs are Google's own AI accelerators. They are chips built for one job: the matrix math that neural networks spend most of their time on. Google has been designing them for about ten years, and uses them to train and serve its own models, including Gemini.
TPUs are available on Google Cloud, from a single chip up to pods of thousands of chips. They shine when you need a lot of compute, for training or for serving at scale.
#Google#Meta#PyTorch#TorchTPU#Google TPU#Google Cloud#Hugging Face#Transformers#Diffusers#vLLM#SGLang#LiteRT