llm-action
A Chinese-language collection of tutorials and hands-on notes on large language model engineering, covering training, inference, compression, evaluation and deployment techniques.
llm-action shares technical background and practical experience on large language models, covering their engineering and real-world application. Content is organized as a table of contents spanning training, inference, compression, evaluation, data engineering and deployment.
Training material covers full fine-tuning and parameter-efficient methods such as LoRA, QLoRA and P-Tuning v2, reinforcement learning from human feedback, distributed training techniques, and alignment techniques, with worked tutorials for models such as Alpaca, ChatGLM, Vicuna, and various LLaMA variants. Inference material covers serving frameworks and inference optimization; compression material covers quantization, pruning, knowledge distillation and low-rank decomposition; further sections cover evaluation, prompt engineering, LLM application development, AI compilers, and AI infrastructure such as accelerator cards and cluster networking.
The material is presented mainly as linked articles and tutorials, written in Chinese.