AI
Google open-sources TorchTPU, a PyTorch build for its chips
TechCrunch reports existing PyTorch apps can move to Google's TPUs with a configuration change.
TechCrunch's Julie Bort reports Google released TorchTPU, an open-source build of PyTorch (a popular toolkit for building AI models) that runs on Google's TPUs (its custom AI chips). Per TechCrunch, existing PyTorch apps can move to TPUs and Google Cloud "with a simple configuration change and no rewriting of their apps."
Google says it worked with Meta, Hugging Face, Inferact (vLLM) and RadixArk (SGLang), and also launched a TPU agent suite to help migrate apps and fix errors. Amazon released a similar project, TorchNeuron for its Trainium chips, on Monday Oct 19.
The caveat: TechCrunch cites "some accounts" for an unsourced 85% PyTorch share, says developers won't automatically leave Nvidia, and states no pricing.
On your desk: nothing to do today. For a solo builder the payoff is indirect, maybe cheaper training and inference options over time, so watch for pricing before you plan around it.
Read the original at TechCrunch: Google has joined Amazon in attacking Nvidia's stronghold with AI app developers →