Google on Tuesday made TorchTPU, its long-awaited version of PyTorch, available as an open source project. The release puts Google alongside Amazon in an effort to weaken Nvidia's grip on the AI industry, where most AI apps are written to run on Nvidia's chips, known as GPUs, by default.

PyTorch is a widely used open source toolkit that developers and startups rely on to build, train and run AI models. Some accounts put its use at 85% of AI apps. Meta, then called Facebook, developed PyTorch in 2016, well before large language models became a household topic. Nvidia's GPUs were already the standard hardware for AI research at the time, so many of PyTorch's features were built to work with them. In 2022, Meta handed PyTorch over to the Linux Foundation, a nonprofit that oversees a number of major open source projects.

Google's TorchTPU lets AI apps run on Google's own AI chips, called Tensor Processing Units, or TPUs. It also makes it simple to move existing PyTorch apps onto TPUs and, by extension, onto Google Cloud. Developers can make the switch with a simple configuration change, without rewriting their apps.

Amazon made a similar move Monday when it released TorchNeuron, its own version of PyTorch. TorchNeuron allows AI apps to be ported to Amazon's custom AI chip, Trainium, and to Amazon Web Services without rewriting code. Amazon said it is going further. It is submitting code to the main PyTorch project that separates various features from Nvidia's chips, which would make the framework open to any hardware by default.

Google says it went a step beyond that. The company worked directly with Meta, Hugging Face, Inferact, which makes vLLM, RadixArk, which makes SGLang, and others to make sure TorchTPU's features and code libraries work well with these widely used AI development tools, just as the original PyTorch does.

Google also launched a dedicated tool, the TPU agent suite, to help developers move their apps to Google's chips and cloud and to fix errors along the way. The company updated LiteRT, its popular tool for running AI on phones, laptops and embedded devices, so it works better with PyTorch-based AI apps.

The changes will not cause developers to abandon Nvidia's hardware en masse. Moving between chips and clouds is now easier, though, and that is the point both Amazon and Google are pressing. Both companies are clearly working to erode Nvidia's advantage.

Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.