Architecture decision
GPU capacity determines embedding work; index size and query behavior determine database resources. Plan them separately.
GLOBALATLAS AI · Embedding infrastructure
Embedding infrastructure
Embedding pipelines encode input into vectors and usually combine model execution with a separate index or database.
Build a retrieval pipeline
- 01
Split source documents into inputs
- 02
Generate vectors with a fixed model version
- 03
Store vectors and metadata in a separate index
Configuration requirements
Specify model, input length, batch size and index growth. GPU inference and vector storage should be sized separately.