LLM-100-Lessons: A Systematic Learning Resource for Large Language Models and AI Agent Technologies

The LLM-100-Lessons repository on GitHub is a systematic learning resource for large language models (LLM) and AI Agent technologies. It aims to address the problem of knowledge fragmentation in LLM learning, covering over 100 core topics across the entire technical stack from pre-training to Agent architecture. It provides a structured learning path for learners and practitioners, helping them build a complete knowledge system.

LLM technology iterates rapidly, but the problem of knowledge fragmentation is prominent: related papers, blogs, and tutorials are scattered everywhere, lacking integration; moreover, LLM involves multiple subfields such as model architecture, pre-training, fine-tuning, alignment, inference optimization, and Agent systems, each with unique methodologies, making it easy for beginners to lose their way.

The LLM-100-Lessons repository was created to solve learning challenges. It collects in-depth analysis of over 100 core technical topics, organized into a complete curriculum system based on technical evolution and dependency relationships. It helps learners gradually build a comprehensive understanding from basic to advanced levels, serving as a one-stop structured knowledge base.

The repository covers the entire LLM technical stack:

• Model Pre-training: Data collection and cleaning, tokenizer design, pre-training objective functions, training infrastructure, etc.;
• Fine-tuning Techniques: Full-parameter fine-tuning, parameter-efficient fine-tuning (LoRA, Adapter), instruction fine-tuning, supervised fine-tuning (SFT), etc.;
• Alignment Strategies: RLHF, DPO, Constitutional AI, etc.;
• Agent System Architecture: Tool usage, planning and reasoning, memory management, multi-Agent collaboration, etc.

LLM-100-Lessons has unique advantages compared to other resources:

• More structured and easier to understand than official documents/papers;
• Free and faster to update than online courses (Coursera, etc.);
• More consistent in quality and systematic in organization than technical blogs;
• Provides in-depth content analysis instead of just listing links, unlike Awesome-LLM and similar repositories.

Recommendations for effectively using this resource:

1. Establish a roadmap: Choose a learning path based on your background and goals—beginners start with basics, while experienced users focus on cutting-edge topics;
2. Combine with practice: Implement or reproduce experiments when learning topics to deepen understanding;
3. Participate in the community: Ask questions and share insights in the discussion area to accelerate learning;
4. Track updates: Review the latest content regularly to keep your knowledge up-to-date.

Notes for use:

• Content accuracy: Maintained by the community—cross-verify key information with original papers/official documents;
• Timeliness: LLM technology evolves rapidly, so some content may be outdated;
• Balance between depth and breadth: Covers a wide range but has limited depth per topic—refer to original literature for deeper exploration;
• Language barrier: If primarily in Chinese, non-Chinese users may face limitations—pay attention to term translations.

Open-source knowledge bases like LLM-100-Lessons fill gaps in traditional educational resources, providing equal opportunities for global learners. The community-driven update mechanism allows it to keep up with cutting-edge technology. We look forward to more such projects to jointly build a rich, open, and high-quality LLM learning ecosystem, becoming a valuable knowledge asset for AI practitioners.