You must reduce the number of records in a vector index. What helps?
Why A is the answer
Larger chunks and less overlap
Larger chunks and less overlap produce fewer chunks.
In detail
- Each chunk becomes one record in the index.
- Larger chunks and less overlap produce fewer chunks.
- Measure retrieval after any change to chunking.
Use case
An illustrative scenario, not a real incident.
An index holds far more chunks than it needs, and storage costs grow. Larger chunks with less overlap cut the record count, and retrieval tests confirm that quality holds.
See it in different ways
flowchart LR A["Larger chunks"] --> C["Fewer records"] B["Less overlap"] --> C D["Smaller chunks"] --> F["More records"] E["More overlap"] --> F
stateDiagram-v2 [*] --> Raw Raw --> Text: parse or OCR Text --> Clean: remove extraneous content Clean --> Chunked: split Chunked --> Governed: Delta table in Unity Catalog Governed --> Indexed: Vector Search Indexed --> [*]
flowchart TD
Q{"What do you change?"}
Q -->|"Bigger chunks"| A0["Fewer records"]
Q -->|"Less overlap"| A1["Fewer records"]
Q -->|"Smaller chunks"| A2["More records"]Key point: Larger chunks and less overlap mean fewer records.
Why the other options are wrong
- B. Smaller chunks and more overlap
- This produces more records, not fewer.
- C. A larger embedding dimension
- Dimension changes the size of each vector, not the number of records.
- D. A longer system prompt
- A system prompt has no effect on the index.