Datawhale and SGLang release open-source course building an inference engine from scratch

Datawhale and RadixArk (SGLang) released an open-source course to build a mini-sglang from scratch, then progress to reading real SGLang source code and contributing PRs. The course covers LLM reasoning concepts, KV cache, continuous batching, page attention, and speculative decoding. Part III, covering attention backends, quantization, and PD separation, is written by SGLang official team members, making it a rare authoritative system tutorial.