This category/tag is for discussions and questions about running PyTorch workloads on Google Cloud TPUs using the native torch_tpu backend and the broader PyTorch ecosystem including TorchTitan, vLLM, SGLang, TorchRec, and Hugging Face. Designed to provide a native and familiar experience for the PyTorch community without complex code rewrites, PyTorch on TPU empowers developers to efficiently train, fine-tune, and serve machine learning models at scale while staying entirely within the PyTorch ecosystem they already know.