Open-Source LLMs, Cooked for 'Chinese Babies'

What it does

This is a curated collection of Jupyter Notebook tutorials that walk Chinese beginners through the full lifecycle of running open-source LLMs on Linux: environment setup, local deployment, and fine-tuning via LoRA, p-tuning, or full-parameter distributed training. It covers over 50 domestic and international models—from Qwen and DeepSeek to LLaMA and Gemma—acting as a step-by-step bridge between wanting to run a model and actually running it.

The interesting bit

The project leans into its “for Chinese babies” branding, but the real substance is a relentless focus on lowering barriers for students and researchers without API budgets. It even maintains dedicated tracks for AMD GPUs and Huawei Ascend NPUs, hardware ecosystems that most mainstream Western tutorials ignore entirely.

Key highlights

• Covers 50+ models with per-model environment configs and deployment guides
• Includes end-to-end example projects: a Zhen Huan palace-drama roleplay bot, a social-etiquette assistant, and an advanced-math tutor
• Supports fine-tuning methods from LoRA to distributed full-parameter training
• Maintains separate AMD GPU and Ascend NPU sections with platform-specific optimizations
• Explicitly targets “the widest, most ordinary student groups” rather than ML engineers

Caveats

• The entire project is tutorials and documentation; there is no unified framework or library underneath
• Content is overwhelmingly Chinese-language and Linux-centric; Windows and macOS users are not the target audience

Verdict

Grab this if you are a Chinese-speaking student or hobbyist trying to escape API fees and actually run models locally. Skip it if you are looking for a reusable training framework or English-language quickstarts.