Amazon Opens Its AI Chips to All PyTorch Developers
Your AI experiments could soon run on cheaper, faster Amazon chips—no code changes needed.
AWS is announcing TorchNeuron, an open source native PyTorch backend for Trainium, so the model code, libraries, and tools you already use run on Trainium. It builds on PyTorch’s accelerator-agnostic extension points, including the PrivateUse1 device backend and torch.accelerator APIs.
TorchNeuron supports eager execution for research and debugging, including experimental adaptive eager mode, torch.compile for production performance, and the Neuron Kernel Interface for custom kernels on Trainium2 and Trainium3. Getting started uses .to('neuron'), and modern code written against torch.accelerator runs unmodified.
TorchNeuron will be released as open source under Apache 2.0 at github.com/aws-neuron/torch-neuronx, going live as public Beta on Oct. 20, ahead of its open source launch at PyTorch Conference North America on Oct. 20.
- TorchNeuron lets AI models built with PyTorch run on Amazon's Trainium chips with minimal code changes.
- This could lower costs for AI services and speed up development of new AI features.
- It's in beta and only works on the newest Trainium chips, so not all AI models will work immediately.
Why It Matters
Cheaper AI development could lead to lower prices and smarter AI features in apps you use daily.