Meta AI Updates: October 21, 2026

1. Google and Meta’s PyTorch Team Open-Source TorchTPU, a Native PyTorch Backend for TPUs

Meta. TorchTPU, a project started at Google in mid-2025 in collaboration with the PyTorch team at Meta, was introduced on the PyTorch blog at PyTorch Conference North America as a native PyTorch backend for Google TPUs. Unlike PyTorch/XLA, it has no JAX runtime dependency, prioritizes eager execution (including a new DeferAndFuse eager mode), and builds torch.compile on TorchDynamo, so existing workloads can move with changes such as .to("cuda") to .to("tpu"). The team reports support for about 450 ATen ops, custom kernels via JAX/Pallas and Helion, validated out-of-box execution on the 5,000 most downloaded Hugging Face models, and integrations with vLLM, SGLang, TorchTitan, Ray, and TorchRec. Four repositories (TorchTPU, vLLM-TorchTPU, SGLang-TorchTPU, and TorchTitan) are being open-sourced at the conference. Source