Definition
The systematic approach to ingesting and retrieving information from data sources exceeding an LLM's context window, balancing the trade-off between granular retrieval (small chunks) and semantic coherence (global context) via strategies like hierarchical indexing or recursive summarization.
Distinguished from simple file parsing by its focus on token-limit management and preserving long-range dependencies.
"A 100-foot scroll being cut into a numbered book with a table of contents to fit into a standard briefcase."
- Chunking(Component)
- Context Window(Prerequisite)
- Hierarchical Indexing(Component)
- Lost in the Middle(Relevant Phenomenon)
Conceptual Overview
The systematic approach to ingesting and retrieving information from data sources exceeding an LLM's context window, balancing the trade-off between granular retrieval (small chunks) and semantic coherence (global context) via strategies like hierarchical indexing or recursive summarization.
Disambiguation
Distinguished from simple file parsing by its focus on token-limit management and preserving long-range dependencies.
Visual Analog
A 100-foot scroll being cut into a numbered book with a table of contents to fit into a standard briefcase.