Setting a default embedding model for vector stores

You can set the embedding model that different types of vector stores use to vectorize data into arrays of numerical values represented in a multidimensional space.

Before you begin

You must be a cluster administrator.

Procedure

Set the default embedding model for various types of vector stores such as in memory, Elasticsearch and watsonx.data™ Milvus by running the following command:

oc patch watsonxaiifm watsonxaiifm-cr \
--namespace=${PROJECT_CPD_INST_OPERANDS} \
--type=merge \
--patch='{"spec":{"default_models": {"embedding_model_memory": <model_id>, "embedding_model_elasticsearch": <model_id>, "embedding_model_watsonx_data": <model_id>}}}'

Use the following table to determine which embedding models can be configured for each type of vector data store.

✓ - The embedding model can be used as the default model by the vector store.

Embedding model ID	In memory vector store	Elasticsearch vector store	watsonx.data Milvus vector store
`sentence-transformers/all-minilm-l6-v2`	✓	✓	✓
`sentence-transformers/all-minilm-l12-v2`	✓	✓	✓
`.elser_model_1`	✓	✓
`.elser_model_2`	✓	✓
`.elser_model_2_linux-x86_64`	✓	✓
`ibm/granite-embedding-107m-multilingual`	✓	✓	✓
`ibm/granite-embedding-278m-multilingual`	✓	✓	✓
`intfloat/multilingual-e5-large`	✓	✓	✓
`ibm/slate-30m-english-rtrvr`	✓	✓	✓
`ibm/slate-125m-english-rtrvr`	✓	✓	✓

What to do next

To get started with using embedding models to improve semantic search, document comparison, and re-ranking content based on relevancy, see Supported encoder models.