AWS Neuron 2.33 adds native PyTorch support, Neuron Observer, and other new features

AWS Neuron 2.33 Release Notes

AWS Neuron 2.33 is now available, introducing several new features and improvements:

New Features

• TorchNeuron: The native PyTorch backend for AWS Trainium, in public beta. Run models written with the standard PyTorch APIs on Amazon EC2 Trn2 and Trn3 instances, in eager mode or with torch.compile.
• Neuron Observer: Detects Trainium health issues in under a second and identifies the affected node or device, in public beta.
• vLLM Neuron Plugin: Upgraded to vLLM 0.29.0, supporting native PyTorch compilation through TorchNeuron.
• vLLM Omni Neuron Plugin: For Wan2.2 video generation, now part of the Neuron release, supporting native PyTorch compilation through TorchNeuron.
• New Neuron AMI and Container: Include TorchNeuron, PyTorch 2.14, and TorchTitan.
• Neuron Explorer: Opens profiles captured with torch.profiler. Attributes allocations to the Python source lines that made them, for debugging out-of-memory errors in eager mode.
• Neuron Kernel Interface (NKI): Kernels can now use the full Python language for metaprogramming, including generators and class inheritance. 25 new experimental kernels added, including DeepSeek-V3.2, Qwen3.5, and Kimi Delta Attention.

What to do

• Review the AWS Neuron 2.33.0 Release Notes for detailed information.
• Explore the AWS TorchNeuron documentation for getting started with TorchNeuron.
• Check the AWS Neuron Observer documentation for using Neuron Observer.
• Read the AWS Neuron Kernel Interface documentation for metaprogramming with NKI.

Source: AWS release notes

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.