Google 与 Meta 开源 TorchTPU，为 TPU 带来原生 PyTorch 支持
原标题：Google and Meta Open-source TorchTPU to Bring Native PyTorch Support to Google TPUs
AI 摘要
Google 与 Meta 联合开源 TorchTPU，为 Google TPU 提供原生 PyTorch 后端，使现有 PyTorch 工作负载只需少量代码修改即可在 TPU 上运行，并支持 eager 与编译执行、分布式计算及硬件优化。该项目旨在替代此前基于 PyTorch/XLA 的方案，解决其需侵入式改代码、编译延迟高、缺乏原生 eager 模式等问题。Google 用约 5000 个 Hugging Face 模型验证兼容性，并在 Muse Glimmer 与 Qwen3 MoE 235B 上测试了推理与训练性能。
正文节选
TorchTPU is an open-source project developed by Google in collaboration with Meta that provides a native PyTorch backend for Google TPUs. It enables PyTorch workloads to execute on TPU hardware with minimal code modifications, supporting eager and compiled execution, distributed computing, and hardware-specific optimizations. Google began enabling PyTorch workloads to run on TPUs in 2020 through PyTorch/XLA, which leveraged JAX, XLA, and lazy tensors. According to Google, this approach had signi