AWS launches TorchNeuron to ease PyTorch applications onto Trainium chips

Amazon Web Services has launched TorchNeuron, a tool that lets developers run PyTorch AI applications on its Trainium chips with a single configuration change, according to the company. The move aims to make it easier for startups to use Amazon’s AI hardware and challenge Nvidia’s advantage, built partly on the close integration between its CUDA software and PyTorch.

Amazon says around 85% of AI applications use PyTorch, and says TorchNeuron requires no code rewrite. It argues that Trainium’s integration with AWS can help startups move from prototypes to production and reduce costs across model training and inference. AWS cites customer Deepgram, which it says processes 70% more audio per dollar on Trainium while maintaining production-grade accuracy; Amazon also plans to contribute changes to PyTorch to reduce its dependence on specific chips.