A Chinese LLM bootcamp for the GPU-less and API-anxious

What it does

This repo is a Chinese-language onboarding track for developers entering the AI/LLM space. It starts with calling APIs through the OpenAI SDK—using domestic providers like DeepSeek, Aliyun, and Zhipu to bypass the “how do I get a key” paralysis—and gradually escalates to local model deployment, LoRA fine-tuning, and Stable Diffusion. Every major step has a Kaggle or Colab notebook attached, so you can run code even if your machine lacks a GPU.

The interesting bit

The author recently switched the entire project to uv for environment management despite admitting it is “not friendly,” reasoning that long-term pain beats short-term comfort. The repo also doubles as a complete Chinese-mirror homework set for HUNG-YI LEE’s 2024 Generative AI course, complete with a small CodePlayground for one-line AI experiments.

Key highlights

• Zero-GPU onboarding: API-focused tutorials run on Kaggle/Colab; GPU-heavy sections are tagged LLM or SD so you know what you’re signing up for.
• DeepSeek-centric API primer: detailed guides on obtaining and using DeepSeek through multiple Chinese cloud providers, all via the standard OpenAI SDK.
• Curated path from prompt engineering to LoRA fine-tuning and image generation, with Docker base images provided for local setup.
• Includes PaperNotes for reading foundational LLM papers and a CodePlayground for quick script experiments.
• Recently added MCP (Model Context Protocol) deep-dives and Claude Code usage guides.

Caveats

• Colab links are currently broken because the original account lacked a recovery email; the author is migrating them to a new account.
• Kaggle blocks Gradio, so some interactive demos are restricted to local or Colab execution.
• The entire curriculum is in Chinese.

Verdict

Ideal for Chinese-speaking developers who have watched the theory lectures but keep stalling at the “get an API key and run something” stage. If you already deploy and fine-tune local models routinely, or you don’t read Chinese, this is not your stop.