This repository provides a comprehensive, step-by-step guide and practical demos for learning AI and Large Language Models (LLMs), targeting beginners and students. It bridges the gap from API usage to local model deployment and fine-tuning, enabling learning even without a dedicated GPU, and includes Chinese mirrored assignments for Li Hongyi's 2024 Generative AI course.

How It Works

The project adopts a progressive learning approach, starting with simple API calls (e.g., OpenAI SDK compatible) and gradually moving towards local LLM deployment, fine-tuning (LoRA, DPO), and advanced concepts like RAG and quantization. It leverages online platforms like Kaggle and Colab for accessible execution and provides detailed explanations of underlying mechanisms, including model parameters, memory usage, and sampling strategies.

Quick Start & Requirements

• Installation: Clone the repository using git clone https://github.com/Hoper-J/AI-Guide-and-Demos-zh_CN.git.
• Environment: Recommended to use Conda for virtual environments (conda create -n aigc python=3.9, conda activate aigc).
• Core Dependencies: PyTorch with CUDA support (e.g., pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118).
• Optional: Docker is available for a pre-configured environment.
• Resources: Online notebooks (Kaggle/Colab) are provided for GPU-accelerated learning. Local execution may require significant VRAM for LLM tasks.
• Docs: Li Hongyi's Generative AI Course

Highlighted Details

• Comprehensive coverage from API usage to local LLM fine-tuning and deployment.
• Includes Chinese mirrored assignments and code for Li Hongyi's Generative AI course.
• Offers a "Code Playground" for experimenting with AI scripts.
• Detailed explanations of concepts like LoRA, Beam Search, Top-K/Top-P sampling, and quantization.
• Supports local LLM execution using formats like GGUF with Llama-cpp-python.

Maintenance & Community

The project is actively maintained by Hoper-J. Community interaction points are not explicitly listed in the README.

Licensing & Compatibility

The repository's code and content are generally presented for educational purposes. Specific licensing for individual components or datasets is not detailed in the README.

Limitations & Caveats

Some advanced LLM and Stable Diffusion tasks may require significant GPU memory (VRAM). While many examples are designed to run without a GPU via APIs or specific quantization, local fine-tuning and deployment will have hardware requirements. The project does not provide instructions for circumventing internet access restrictions ("科学上网").