“
Hello everyone, this is Goodnote. The knowledge notes on large models RAG & Agent have been updated. The total word count is over 50,000. Due to space limitations, this article will only provide a summary. For detailed notes, please enter our public account and reply with ‘RAG’ and ‘Agent’ to obtain them.
RAG Notes
Table of Contents
-
Overall Architecture Design of RAG -
1. Overview -
1-Overview -
2-Indexing -
3-Retrieval -
4-Generation -
2. Optimizing Element Questions -
5-Multi Query Strategy -
6-RAG-Fusion Multi Query Result Fusion Strategy -
7-Decomposition Problem Decomposition Strategy -
8-Step Back Q&A Backtracking Strategy -
9-HyDE (Hypothetical Document Embeddings) -
Other Methods -
3. Routing Optimization and Problem Construction Strategy -
Routing -
Query Construction Optimization Strategy -
4. Index Generation Optimization -
12-Multi-representation Indexing -
13-RAPTOR -
14-ColBERT -
5. Retrieval and Generation Optimization Strategies -
Retrieval Optimization Strategy -
Retrieval -
15-Re-ranking -
16-Retrieval (CRAG) -
Generation Optimization Strategy -
17-Retrieval (Self-RAG) -
6. Others -
Reasons for RAG Emergence -
RAG Development Stages -
RAG Application Process -
RAG Optimization Methods -
Adaptive-RAG -
Differences between ReAct, CoT, and ToT -
Chunking Optimization -
Sentence Window Retrieval -
Embedding Development -
RAG Effectiveness Evaluation -
12 Issues Faced in Developing RAG Systems -
Context Compression -
Sensitive Information Processing -
Citation Sources -
7. Self-summary -
8. References
Summary
Agent Notes
Table of Contents
-
What is a Large Model Agent? -
Agent’s Thinking Mode -
Decomposing Sub-goals and Task Decomposition Methods -
Reflection and Improvement -
What Components Make Up a Large Model Agent? -
Planning -
Memory -
Tools -
Action -
Challenges of Agents -
References
Summary
Popular Articles Review
Welcome to subscribe to our collection of basic knowledge on artificial intelligence. Currently, the collections of Machine Learning Notes and Deep Learning Notes have been basically updated. Please click the image below to enter the collection directory.
