Blueprint by nvidia

Serve LLMs with SGLang

High-throughput serving with RadixAttention, structured output, and an OpenAI-compatible API