Skip to content
Trending storyDeveloping

AWS releases TorchNeuron, a native PyTorch backend for Trainium chips

2 articles2 sourcessince Oct 19Last article 2h ago ·

Overview

AISummary of 2 articles

Amazon Science announces TorchNeuron, a native PyTorch backend that lets existing PyTorch code run on AWS Trainium chips with a single configuration change and no rewrite.

The PyTorch Blog says it supports eager execution, torch.compile, and the Neuron Kernel Interface on Trainium2 and Trainium3. AWS says it tested 100k Hugging Face models and plans to close gaps in operator coverage in later releases.

The source code will be released under Apache 2.0 at github.com/aws-neuron/torch-neuronx, going live as a public beta on October 20 at PyTorchCon. AWS says TorchAO and FP8 are currently in private beta and will be generally available in the next release.

Written by AI from the articles below · updated Oct 19, 9:31 AM ET

Check the sources:

Article timeline

The articles in this story. Times are ET.

Oct 19
  1. Amazon ScienceOfficial
    Amazon Science announces TorchNeuron, running PyTorch natively on AWS Trainium

    AIAmazon Science announces TorchNeuron, which lets PyTorch code run natively on AWS Trainium chips with a single configuration change and no rewrite. The project will be open sourced at PyTorchCon on October 20.

    Video from @AmazonScience's post
  2. PyTorch BlogOfficialPick
    AWS announces TorchNeuron, an open source native PyTorch backend for Trainium

    AIAWS announces TorchNeuron, an open source native PyTorch backend that lets existing PyTorch code run on AWS Trainium through the standard device API. It supports eager execution, torch.compile, and the Neuron Kernel Interface on Trainium2 and Trainium3, with the source code released under Apache 2.0 and a public beta starting Oct. 20. The post reports that 100k Hugging Face models were tested, with gaps in operator coverage that AWS plans to close in later releases.

Heat Trend

Not enough continuous observations to show a trend yet.