Google and Meta release open-source TorchTPU
What happened
On October 21, early reports said Google and Meta had teamed up on an open-source TorchTPU, giving Google TPUs a native PyTorch backend with eager execution and torch.compile support. Later reports said TorchTPU lets PyTorch models and operators run directly on TPU devices, with custom ATen kernels and an XLA compilation cache, plus data-parallel and tensor-parallel distributed training and torch.compile() graph-mode execution. Those later reports credited the project to Google alone and did not mention Meta. Neither report gave a version number, performance figures or release plans.
Written by AI from the coverage · updated 58 minutes ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageGoogle open-sources TorchTPU, a native PyTorch back end for TPUs
Google 开源 TorchTPU，为 TPU 提供原生 PyTorch 后端，让 PyTorch 模型和算子可直接在 TPU 设备上运行。项目包含自定义 ATen kernel、XLA 计算编译缓存，支持数据并行与张量并行分布式训练，并集成 torch.compile() 图模式执行。
- AI HOT · ProductsGoogle 与 Meta 发布 TorchTPU：PyTorch 原生 TPU 后端
Google 与 Meta 合作推出开源 TorchTPU，为 Google TPU 提供原生 PyTorch 后端，支持即时执行与 torch.compile。
Heat over time
Not enough continuous observations to draw a trend yet.