Definition
The specific number of tokens or characters used to partition a document into discrete segments for vectorization. Smaller chunks increase retrieval precision but lose context, while larger chunks preserve context but risk introducing noise and exceeding the LLM's context window.
Refers to the length of individual segments during ingestion, not the total file size or the LLM's context limit.
"A loaf of bread sliced into individual pieces: too thin and they fall apart (loss of context); too thick and they won't fit in the toaster (context window overflow)."
- Chunk Overlap(Component)
- Embedding Model(Prerequisite)
- Vector Database(Component)
- Semantic Splitting(Alternative Method)
Conceptual Overview
The specific number of tokens or characters used to partition a document into discrete segments for vectorization. Smaller chunks increase retrieval precision but lose context, while larger chunks preserve context but risk introducing noise and exceeding the LLM's context window.
Disambiguation
Refers to the length of individual segments during ingestion, not the total file size or the LLM's context limit.
Visual Analog
A loaf of bread sliced into individual pieces: too thin and they fall apart (loss of context); too thick and they won't fit in the toaster (context window overflow).