AWS and RadixArk release SGLang-Neuron for Trainium inference

Amazon says SGLang-Neuron is now publicly available through an AWS Deep Learning Container; the plugin was released by RadixArk in collaboration with Annapurna Labs. It lets teams serve models and generate reinforcement learning rollouts on Trainium, AWS's family of AI chips, using the same SGLang server interface and generation APIs they already use. Installing the plugin registers the Neuron platform, and setting SGLANG_PLATFORM=neuron selects it at server startup. Trainium-specific models, attention, mixture-of-experts and graph-runner code sit in the plugin, while SGLang core keeps the API server, request lifecycle, RadixAttention and prefix-cache bookkeeping; execution runs on native PyTorch compiled with torch.compile. Prefill/decode disaggregation is supported, moving the prompt KV cache to a decode server over Elastic Fabric Adapter. AWS points to a getting-started guide and repository for supported models and validated configurations.