Tag
This thread recommends three key papers for mastering LLM inference: PagedAttention, Sarathi-Serve, and SGLang, which introduce efficient memory management, chunked prefills, and structured generation techniques used in modern inference engines like vLLM and TensorRT-LLM.