Google Unveils Torchtpu to Run Pytorch Apps on Tpus

WHO:

Techmeme - Google announced TorchTPU, described as its long-awaited version of PyTorch.

MarkTechPost - Google and Meta open-sourced TorchTPU.

WHAT:

Techmeme - TorchTPU is an open-source project intended to let AI applications run on TPUs and Google Cloud without rewriting code.

MarkTechPost - TorchTPU is a native PyTorch backend for Cloud TPUs.

MarkTechPost - TorchTPU runs in eager mode by default, compiles through `torch.compile` and XLA, and replaces PyTorch/XLA for new workloads.

HOW:

Techmeme - Running AI applications on TPUs and Google Cloud requires only simple configuration changes, rather than code rewrites.

MarkTechPost - TorchTPU supports eager execution by default and compilation through `torch.compile` and XLA.

(UPDATED October 21, 2026 00:36 UTC)

Google and Meta have open-sourced TorchTPU, a native PyTorch backend for Google Cloud TPUs. The project is designed to let AI applications run on TPUs without rewriting code, requiring only simple configuration changes. It supports eager execution by default and compilation through `torch.compile` and XLA, and is intended to replace PyTorch/XLA for new workloads.