SGLang explains serving and RL rollouts on AWS Trainium

SGLang says model serving and reinforcement learning rollouts on AWS Trainium work through its existing interface.

The session, presented by AWS's Pinak Panigrahi and RadixArk's Alex Nails, covers the TorchNeuron plugin beta, NKI kernels, parallelism, speculative decoding, and an introduction to Miles. The report is a single post from SGLang's X account, which links the talk and does not include performance figures.