https://www.ruyaohui.com/tags/llm-inference/Recent content in LLM Inference on RiverYu's Homehttps://www.ruyaohui.com/images/cover.pnghttps://www.ruyaohui.com/images/cover.pngHugo -- 0.152.2zh-cn2026 RiverYu’s Home · Per aspera ad astraTue, 20 Oct 2026 00:00:00 +0000https://www.ruyaohui.com/posts/from-ai-request-to-computer-systems/Tue, 20 Oct 2026 00:00:00 +0000https://www.ruyaohui.com/posts/from-ai-request-to-computer-systems/追踪一次 AI 请求：从 UTF-8、线性代数、INT4、数据结构与计算图，到 Roofline、虚拟内存、KV 前缀缓存、SSE/TCP 和端到端性能。包含考研 408、可复核推导与跨层排障。