AWS Neuron 2.32 Unlocks Advanced Training on Trainium Chips
AWS Neuron 2.32 brings fresh NKI programming features like on-device top-K instructions and variable-size collectives for Trn2/Trn3, plus 13 new kernels optimized for MoE training and sparse attention. You'll also get MXFP8 training support for blockwise MoE layers and a new Neuron Agentic Development skill that ports transformer models to vLLM—all available across AWS regions where Trainium instances run.
source: [aws/whats-new]