AWS ships TorchNeuron beta for native PyTorch on Trainium

TorchNeuron removes one of Trainium's biggest adoption barriers: developers previously needed Neuron-specific tooling. With eager mode and torch.compile, porting is a device change.

The weekly roundup also includes Polly's Generative engine with 57 voices, ECS Gateway, SageMaker Training Sessions, and GA of AWS Continuum.

TorchNeuron targets NVIDIA's CUDA moat directly; Google's TPUs have similar PyTorch/XLA paths.

Watch performance parity and coverage of custom kernels.