Oriel — AI & technology worth noticing, every day.

Google open-sources TorchTPU, a native PyTorch back end for TPUs

Repository · https://github.com/google-pytorch/torch_tpu

First seen by Oriel on .

AI summary

torch_tpu is a PyTorch backend for Google TPUs, combining custom ATen kernels and XLA compilation support to run models and operations on TPU devices.

  • It supports graph execution through torch.compile() and includes a compilation cache for XLA computations.
  • Distributed training examples cover Data Parallelism and Tensor Parallelism, alongside inference examples for models including Llama, Qwen, ResNet, and UNet.
  • The project provides debugging and benchmarking utilities, plus Marimo notebooks and a user guide for TPU VM users.
  • Python 3.11+ is supported, and the latest public release listed is torch_tpu==0.2.0 on PyPI.
  • Source builds use Bazel; wheel builds require --config=wheel_common, and installation instructions specify the CPU PyTorch package index.

#pytorch #tpu #xla #distributed training

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hacker News: 375, latest observation ; +350 over observed period since

Comments

Comment count reported by the source.

  • Hacker News: 88, latest observation ; +81 over observed period since

Aliases

SourceExternal IDURL
Hacker News50169053https://github.com/google-pytorch/torch_tpu Read discussion