5 posts tagged with "LLM"

不是泛泛介绍 Ollama 是什么，而是专门测它的速度——在 Apple M1 Pro 上用 llama3.2:1b、qwen2.5:0.5b 跑出真实的 tokens/s 数据，同时梳理社区里"Ollama 比原生 llama.cpp 慢"的争议、和 vLLM/MLX 的对比，以及量化、上下文长度、GPU 开关这些影响速度的关键因素。

Blume is an open-source documentation framework built on Astro and Vite — maintain a folder of Markdown and it generates a static docs site with local search, AI-ready output, and a component library, without you having to build a React app yourself.

DSH Desktop is a community-built native desktop client for DeepSeek Harness, packaging the official Web UI, Host service, and plugin system into a download-and-run desktop app, unaffiliated with DeepSeek's official team.

Kimi CLI is Moonshot AI's terminal-based AI coding agent that can read and edit code, run shell commands, and hook into MCP and ACP editor protocols; the project has now been flagged by its own maintainers as being gradually wound down, with new users pointed to the successor project Kimi Code CLI.

wigolo is a local-first MCP server that turns web search, fetch, crawl, and structured extraction — six core tools total — into a local engine that needs no API key at all; this article covers actually setting it up and testing each of the six tools.