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

AWS Neuron 2.33 is now available, introducing TorchNeuron, the native PyTorch backend for AWS Trainium, in public beta.

• TorchNeuron enables standard PyTorch APIs on EC2 Trn2 and Trn3 instances in eager mode or with torch.compile
• Neuron Observer detects Trainium health issues in under a second and identifies affected nodes
• vLLM Neuron plugin upgraded to 0.29.0 with native PyTorch compilation support
• New Deep Learning AMI and Container include TorchNeuron, PyTorch 2.14, and TorchTitan
• Neuron Explorer profiles torch.profiler captures and attributes memory allocations to Python source lines
• NKI now supports full Python metaprogramming with 25 new experimental kernels for DeepSeek, Qwen, and Kimi models

These enhancements streamline PyTorch model training and deployment on AWS Trainium accelerators with improved debugging and monitoring capabilities.