https://socake.github.io/categories/ai-%E5%B7%A5%E7%A8%8B/Recent content in AI 工程 on 黄文卓 | DevOps EngineerHugo -- gohugo.iozh-CN17691281867@163.com (Wenzhuo Huang)17691281867@163.com (Wenzhuo Huang)© 2026 Wenzhuo HuangMon, 06 Apr 2026 11:30:00 +0800https://socake.github.io/posts/autogen-multi-agent-practice/Mon, 06 Apr 2026 11:30:00 +080017691281867@163.com (Wenzhuo Huang)https://socake.github.io/posts/autogen-multi-agent-practice/AutoGen 把多 Agent 协作从玩具推向生产。本文讲清它的核心抽象 (Conversable Agent / Group Chat / 工具调用)，以及从 demo 到生产要处理的那些事。https://socake.github.io/posts/litellm-gateway-proxy/Thu, 02 Apr 2026 14:00:00 +080017691281867@163.com (Wenzhuo Huang)https://socake.github.io/posts/litellm-gateway-proxy/LiteLLM 是 LLM 多模型接入的事实标准。本文讲清它的 Proxy 模式部署、Model Config、Virtual Key、Router Fallback、成本追踪和踩坑实录。https://socake.github.io/posts/unsloth-efficient-finetuning/Sun, 22 Mar 2026 09:15:00 +080017691281867@163.com (Wenzhuo Huang)https://socake.github.io/posts/unsloth-efficient-finetuning/Unsloth 用手写 Triton kernel 把单卡 LoRA 微调速度和显存压到极致。本文讲清 Unsloth 的原理、和 LLaMA Factory/TRL 的组合用法，以及真实使用的坑。https://socake.github.io/posts/llamafactory-finetuning/Wed, 18 Mar 2026 11:20:00 +080017691281867@163.com (Wenzhuo Huang)https://socake.github.io/posts/llamafactory-finetuning/LLaMA Factory 把大模型微调的很多 trick 工程化了。本文按一个完整项目的节奏讲：数据、SFT、LoRA、DPO、合并、评估和常见坑。