Amazon released TorchNeuron. It lets PyTorch code run on AWS Trainium chips with no rewrite. A single configuration change is the only step required. The same AI training script produced identical results on both chip types. TorchNeuron plugs into PyTorch’s standard hardware registration. It will be open-sourced on October 20 at PyTorch Conference North America. Startups and enterprises in robotics, video, and drug discovery already use it. Amazon is contributing changes upstream to PyTorch to make it more hardware-agnostic. TorchNeuron supports Trainium2 and Trainium3 chips. It integrates with TorchTitan for distributed training.
What changed
PyTorch code required rewrites to run on AWS Trainium chips.
What it unlocks
Running existing PyTorch code on AWS Trainium chips with one configuration change.
- 85% of AI developers use PyTorch
- Apache 2.0 license
- PyTorch Conference North America on October 20
What you need to act on it
- PyTorch codebase
- AWS Trainium hardware
Sources