Model serving Inference Optimization GPU Infrastructure

Miles

An asynchronous reinforcement-learning framework for large language models

A platform for LLM post-training and deployment-oriented workflows, including on-policy distillation and technical serving-stack work with SGLang.

Recent stories

1 linked story

newsSECONDARY2026-04-25

SGLang supports DeepSeek V4 with 199 tok/s on B200 and 240 tok/s at 900K context

SGLang and Miles published a technical breakdown of their DeepSeek V4 day-zero stack, including ShadowRadix caching, Flash Compressor, FP4 expert-weight handling, and measured B200/H200 throughput. That gives deployers concrete serving and training-path numbers for V4 beyond generic launch-day compatibility claims.