LLM-Wiki: A Structured Knowledge Base for Large Language Model Technologies

LLM-Wiki is a structured knowledge base project for large language model technologies maintained by VectorPeak, released on GitHub on June 17, 2026. This project covers core topics such as LLM fundamentals, agents, RAG, model training, evaluation methodologies, and AI engineering practices, aiming to provide systematic knowledge organization and archiving services for practitioners in different roles.

• Original Author/Maintainer: VectorPeak
• Source Platform: GitHub
• Original Link: https://github.com/VectorPeak/LLM-Wiki
• Release Date: 2026-06-17

LLM-Wiki aims to provide structured knowledge organization and archiving for the field of large language model (LLM) technologies, covering comprehensive content from basic model architectures to advanced application practices, including core topics such as agents, RAG, model training techniques, evaluation methodologies, and AI engineering practices.

1. LLM Fundamentals

Covers model architecture evolution, pre-training techniques, fine-tuning strategies (full-parameter fine-tuning, LoRA, QLoRA), and inference optimization (KV caching, quantization, speculative decoding)

2. Agent Systems

Includes agent architecture patterns (ReAct, Plan-and-Execute, Reflection), tool calling mechanisms, multi-agent collaboration, and memory management systems

3. RAG Technologies

Covers document processing pipelines, retrieval strategies (dense/sparse/hybrid retrieval), re-ranking techniques, and query optimization

4. Model Training and Fine-tuning

Includes data engineering, training strategies (pre-training, instruction fine-tuning, RLHF), distributed training, and training monitoring

5. Evaluation Methodologies

Covers benchmark tests (MMLU/GSM8K/HumanEval), automatic evaluation metrics, human evaluation, and domain-specific evaluation

6. AI Engineering Practices

Covers model deployment (vLLM/TensorRT-LLM/TGI), service architecture, cost control, and safety & alignment

1. Structured Organization: Adopts a Wiki-style hierarchical structure for easy navigation and retrieval
2. Continuous Updates: Follows the rapid development of the LLM field to maintain content timeliness
3. Practice-Oriented: Focuses on the operability of engineering practices
4. Community-Driven: Open collaboration model that gathers community wisdom

The value of LLM-Wiki lies in systematic knowledge integration:

• Researchers: Quickly understand the overall landscape of the field and grasp the context of technological evolution
• Engineers: Obtain directly implementable engineering practice guidelines
• Product Managers: Understand technical boundaries and make reasonable product plans
• Learners: Build a complete knowledge system and avoid fragmented learning

LLM-Wiki represents a new paradigm of knowledge management in the AI era and is a carrier for knowledge precipitation and inheritance in the LLM field. As new technologies such as multimodal models emerge, the knowledge base will continue to evolve. It is recommended that interested practitioners visit the GitHub link to learn more or participate in community collaboration.