AI News · Models ·
Google open-sources TorchTPU so PyTorch apps can run on its TPU chips
Google released TorchTPU as an open-source project, letting PyTorch AI apps run on its Tensor Processing Units with a configuration change instead of a rewrite. The launch came one day after Amazon released TorchNeuron for its Trainium chips. Google also introduced a TPU agent suite to help developers migrate apps.
Key points
- TorchTPU is open source and lets PyTorch apps run on Google TPUs with a configuration change.
- Google also launched a TPU agent suite to help migrate apps and fix errors.
- The release came a day after Amazon launched TorchNeuron for its Trainium chips.
What happened: Google announced on October 20 that TorchTPU, its version of PyTorch, is now available as an open-source project, TechCrunch reported. TorchTPU lets PyTorch AI apps run on Google's Tensor Processing Units (TPUs). According to the report, developers can move existing apps to TPUs and Google Cloud with a simple configuration change and without rewriting them.
The details: Google said it worked with Meta, Hugging Face, Inferact, the maker of vLLM, RadixArk, the maker of SGLang, and others so TorchTPU works with popular AI development tools. It also launched a TPU agent suite that helps developers move apps to its chips and cloud and fix errors along the way, and it updated LiteRT, its tool for running AI on phones, laptops and embedded devices, to work better with PyTorch apps. Pricing, performance comparisons and adoption figures were not reported.
Background: The release follows Amazon's launch of TorchNeuron on October 19, which lets PyTorch apps run on its Trainium chips. Amazon also said it is submitting code to the main PyTorch project to decouple features from Nvidia chips. PyTorch, first developed at Meta and handed to the Linux Foundation in 2022, was largely built around Nvidia GPUs, and TechCrunch said it is used by 85% of AI apps by some accounts.
What to watch: TechCrunch noted that developers are unlikely to abandon Nvidia hardware quickly. Nvidia's response to either release has not been disclosed.
Our take
Teams building on PyTorch may soon be able to compare Nvidia, Google and AWS hardware on price without rewriting their models.