InfraTech

Why we included this project

This notebook collection targets engineers who work with vLLM, SGLang, and similar serving stacks but have only a fuzzy idea of what the framework is doing under the hood. Each topic comes as a runnable Python notebook that works through KV cache management, RadixAttention, chunked prefill, speculative decoding, and quantization, paired with a Chinese-language article that goes deeper into the same material. The pairing makes it usable for solo study and for onboarding a team. The scope reaches beyond a single engine, with notebooks on parallel strategies, CUDA IPC tensor sharing, and a minimal training framework, so readers can build a mental model of the whole stack. The depth.