TorchTPU

Latest news

1. 20 Oct
   Google launches native vLLM and SGLang PyTorch engines on TPU

   Google announced new native PyTorch inference backends for vLLM and SGLang running on TPU, built on a new TorchTPU runtime. The engines include production-ready serving features and custom Pallas kernels to make serving on TPU faster and easier to debug for PyTorch developers.

   Developers · 1 source

   

2. 20 Oct
   Google adds TorchTPU support to TorchTitan for Cloud TPUs

   Google announced integration of a native TorchTPU backend into PyTorch and a TorchTitan fork adapted for training on Google Cloud TPUs. The fork and recipes aim to let developers run large PyTorch models like Llama 3 and Qwen 3 on TPUs with minimal code changes and high-performance distributed training.

   Developers · 1 source

   

Articles

No articles about TorchTPU yet.