Google has released TorchTPU, an open-source version of the PyTorch toolkit designed to run AI applications on its proprietary Tensor Processing Units. This move enables developers to migrate existing PyTorch models to Google Cloud by adjusting configurations without rewriting their code.
The initiative follows a similar announcement from Amazon, which introduced TorchNeuron to support its Trainium AI chips on AWS. Amazon is also contributing code to the main PyTorch project to decouple features from Nvidia hardware, aiming to increase compatibility with diverse chip architectures.
Google collaborated with Meta, Hugging Face, Inferact, and RadixArk to ensure TorchTPU integrates with established AI development tools. The company also launched a TPU agent suite to assist developers with migration and error resolution.
Additionally, Google updated its LiteRT tool to improve performance for PyTorch-based AI applications on mobile devices, laptops, and embedded systems. These efforts represent a strategic push by major cloud providers to reduce industry reliance on Nvidia GPUs.
