Choosing a vector store for a RAG experiment
Before you run a retrieval-augmented generation experiment in AutoAI, you must choose a vector database to store and retrieve data.
For AutoAI RAG experiments, you can choose from three vector store options:
- In-memory Chroma database
- Milvus database
- Elasticsearch
If you are using a vector store in a coded solution, use this format for collection names: autoai_rag_a0b1c2d3_ymdHMS> where y-year, m-month, d-day, H-hours, M-minute, S-second.
Chroma vector database
An in-memory Chroma vector store stores data on your computer's RAM. The index of vectorized documents is fast for querying and retrieving data. Due to memory constraints, the in-memory database is built for small to medium-sized datasets.
If you don't specify a connection to a vector store, the vectorized content is saved to the default, in-memory Chroma database. The content does not persist beyond the experiment, so the Chroma option is not a viable production method for deploying a RAG pattern.
Milvus vector database
A Milvus vector store database has a scalable and distributed architecture. It offers high performance in retrieving data when it has optimized indexing and supports many indexing algorithms and distance metrics. It's built for large datasets and is more durable than a Chroma database.
The vectorized content persists for future patterns. For details, see Working with Milvus.
A Milvus database uses this schema:
| Field | Type |
|---|---|
| document_id | VarChar |
| start_index | Int64 |
| sequence_number | Int64 |
| text | VarChar |
| pk | Int64 |
| vector | FloatVector |
Elasticsearch vector database
Elasticsearch is a search and analytics engine that has vector store search capabilities. Like Milvus, it is built for large datasets and has a durable and scalable architecture. Using the Elasticsearch vector database is beneficial for complex queries because it has strong search and filtering capabilities due to its search engine roots.
Next steps
Follow the fast path to automate the search for a RAG pattern .
Parent topic: Building RAG experiments with AutoAI