LLMBook zh.github.io

As LLM research has accelerated globally, high-quality Chinese-language documentation has lagged behind English resources, making this project particularly valuable for Chinese-speaking practitioners who prefer to study foundational material in their native language without losing technical precision through imperfect translation.

The content covers the progression from early transformer architectures through modern instruction-tuned and RLHF-aligned models, explaining concepts like attention mechanisms, tokenization strategies, pretraining data curation, fine-tuning approaches, and evaluation methodologies in depth.

Each chapter is written to be accessible to practitioners with a solid foundation in deep learning who want to understand LLMs specifically, rather than requiring readers to patch together understanding from scattered paper abstracts and blog posts.

Chinese AI researchers, engineers at major tech companies (Baidu, Alibaba DAMO, Tencent AI Lab, ByteDance), and graduate students working on NLP and foundation model research use this resource to build systematic understanding alongside their experimental work.

The GitHub Pages format keeps it freely accessible without paywalls, and the open-source structure welcomes contributions and corrections from the Chinese ML community.

For those outside China, the project also serves as a window into how the Chinese AI research community structures its technical education and knowledge transfer.