Edge51
探索编程、物理与投资 · ascend-vllm 开发者
Guangdong, China
Latest Posts
vLLM PagedAttention：Block 的完整生命周期
vLLM v0.20.2 · block_size = 16
vLLM Function Call 全链路深度解析
vLLM v0.20.2，聚焦 function call 的主链路分析。 版本与模型：Qwen3CoderToolParser（XML 格式工具调用）
LLM 推理基础：Prefill / Decode 与 KV Cache
Prefill vs Decode
vLLM 推理引擎：请求处理流程深度解析
整体流程
Explore
About
C++ backend developer, passionate about system programming, distributed systems, and clean architecture.
This site is where I share what I learn — code, physics, investing, and books. Most posts are notes to my future self, but if they help someone else along the way, that's even better.