liguodongiot/llm-action

LLM-Action is a Chinese-language knowledge repository covering the full engineering lifecycle of large language models (LLMs), the technology behind AI systems like ChatGPT. The primary audience is AI engineers and researchers who want practical, hands-on guidance for training, fine-tuning, deploying, and optimizing LLMs. The name reflects its focus on actionable knowledge rather than purely theoretical material. The content is organized into detailed sections. Training covers how to train LLMs from scratch and how to fine-tune (adapt) existing models using techniques like LoRA and QLoRA, methods that let you customize a model's behavior for a specific domain using far less computing power than full retraining. Inference covers frameworks and optimization techniques for running LLMs efficiently in production, including quantization (reducing model size to run on less hardware) and pruning (removing redundant parts of a model). The repository also covers model evaluation, prompt engineering (crafting effective instructions for LLMs), data engineering, distributed training across multiple GPUs, LLMOps (operations and deployment workflows for LLMs), and AI accelerator hardware. Each topic typically includes tutorial articles (hosted on Chinese platforms like Zhihu and CSDN) alongside practical code examples and notebooks. The readme and most content are written in Chinese. An AI engineer or ML researcher who wants battle-tested tutorials with working code for training and serving LLMs on Chinese or international models would use this as a reference. The code examples use Python.