glossary.md•5.58 kB
# Glossary
| **Term** | Definition |
|------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| **Bi-encoder** | A model that compresses the meaning of a document or query into a single vector. Used in the first stage of retrieval. |
| **Context Stuffing** | Overloading the context window with too much information, which can degrade LLM performance. |
| **Context Window** | The maximum amount of text that an LLM can process at once. |
| **LLM Recall** | The ability of an LLM to find specific information within its context window. |
| **Recall** | A metric that measures how many relevant documents are retrieved in a search. |
| **Reranker (Cross-encoder)** | A model that takes a query and a document as input and outputs a similarity score. This score is used to reorder documents by relevance. |
| **Retrieval Augmented Generation (RAG)** | A technique that combines the power of Large Language Models (LLMs) with external knowledge sources to generate more accurate and comprehensive responses. |
| **Semantic Search** | Searching for information based on the meaning of words and phrases, rather than just matching keywords. |
| **Two-Stage Retrieval** | A system that first retrieves a large set of potentially relevant documents using a fast retriever (like a vector search) and then reranks them using a more accurate similarity score generated by a slower reranker before presenting them to an LLM. This approach combines the speed of the first-stage retrieval with the accuracy of the second-stage reranking, resulting in a more efficient and effective RAG pipeline. |
| **Vector Search** | A technique used to perform semantic search by converting text into numerical vectors and comparing their proximity in a vector space. |