AWS announced Neuron 2.33 is now available. It includes TorchNeuron, a native PyTorch backend for Trainium chips, in public beta. Developers can run standard PyTorch code on EC2 Trn2 and Trn3 instances, either in eager mode or with torch.compile. The release also adds Neuron Observer, which spots Trainium health problems in under a second and points to the faulty node or device.
Why it matters
PyTorch users can now train and run models on AWS Trainium without custom code changes, speeding up AI development on cheaper hardware.
