Google 开源 TorchTPU，让 PyTorch 代码脱离 Nvidia GPU 跑在 TPU 上
Google 把 TorchTPU 开源了，PyTorch 代码不用改就能跑 TPU，调试还能单步执行，想摆脱 Nvidia 锁定的可以试试。
Google 开源了 TorchTPU，用它替代 PyTorch/XLA 这一层。此前 TPU 依赖 PyTorch/XLA 把操作排队交给编译器，调试方式与 GPU 不同。TorchTPU 让每个操作按原样逐步运行，方便调试，生产环境再切到 torch.compile 提速。实测训练 Qwen 3 8B 时最高用到 TPU 峰值算力的 52%，Amazon 也提供类似的 PyTorch 原生后端。
Google and Amazon now both offer native PyTorch backends that let code leave Nvidia GPUs.
Google just open-sourced TorchTPU that replaces PyTorch/XLA, letting developers debug TPU models step by step.
TorchTPU used up to 52% of peak TPU compute training Qwen 3 8B.
TPUs previously reached PyTorch through PyTorch/XLA, which queued operations for a compiler, so debugging differed from GPU work.
TorchTPU replaces that layer, running each operation as written for debugging, then switching to torch.compile for production speed.
