AWS unveils TorchNeuron for porting AI applications to Trainium — TechCrunch

Amazon Web Services has unveiled TorchNeuron, a tool designed to run AI applications on its proprietary Trainium chips. According to TechCrunch, developers will be able to run applications written using PyTorch on Trainium by changing a single configuration line, without rewriting program code.

TorchNeuron is adapted for PyTorch, an open-source suite of tools for creating, training and running artificial intelligence models. Amazon says that about 85% of AI applications are written using PyTorch. The framework was initially developed by Facebook and has been part of the Linux Foundation since 2022.

Competition with Nvidia

AWS said it seeks to reduce PyTorch developers' dependence on Nvidia hardware. Nvidia has for years ensured close compatibility between its CUDA software and PyTorch, making its hardware the standard choice for many AI applications.

More current news is available on the UA.News Telegram channel Telegram.

PyTorch users can already port applications to AMD graphics processors or Apple chips by changing the configuration. At the same time, AWS says Trainium offers not only compatibility with PyTorch but also access to Amazon's entire cloud infrastructure. The company contrasts this approach with specialized cloud providers that mostly use Nvidia hardware.

Training and inference

Colin Brace, vice president of technology software at AWS Annapurna Labs, said that Trainium can be used both for model training and for inference, meaning the use of a completed model to perform tasks. According to him, this could improve the economics of the full AI workflow cycle.

Brace also said that after switching to Trainium with native PyTorch support, the Deepgram service processes 70% more audio for every dollar spent while maintaining production-level accuracy. AWS said it is working on contributing changes to the core PyTorch project to reduce the framework's attachment to specific chips and make it more hardware-independent.