Loading blogs...
Posts tagged Performance.
Workstation WSL Proxy: compile OpenResty once into bwalia/wslproxy-openresty, warm Buildx/GHA layers, extract on bare metal, prefer DEPLOY_MODE=code — and why AI-assisted delivery made this cross-cutting work finishable.
Workstation 增强 LLMs:用于 KV 缓存的 PagedAttention 分页、vLLM 服务、用于代理的 Self-Debugging、PowerInfer 令牌率和 EG-MLA 内存削减 — 以及生产权衡。