Hello,

VValentin

Hello,
I'm implementing a MultiModalVectorStoreIndex based on an in-memory ChromaDB instance. I'm using an AzureOpenAI embedding model and I would like to understand why when I'm retrieving my nodes using as_retriever().retrieve(...), all the retrieved nodes have null embeddings value. I've checked that the related chroma db collection is containing embeddings value (using chroma_collection.query method )
Am I missing a configuration in the retrieve method or do you have any idea why I'm facing this issue ?

2 comments

LLogan M

the chroma vector store probably is not fetching/returning the embeddings when querying (likely to save memory and speed things up)

I don't think there is an option to enable that beyond overriding the query method in the vector store, or making a PR to add that option

VValentin

Thanks for your answer, I've found that I can retrieve those by doing the following steps:

Retrieve Nodes corresponding to the query

my_nodes_wth_score_lst = my_index.as_retriever().retrieve(my_query)

Extract the correspnding nodes object

nodes_lst = [n.node for n in my_nodes_wth_score_lst]

Get the corresponding embeddings for these nodes

nodes_lst = my_index._get_node_with_embedding(nodes_lst)

Add a reply

Find answers from the community

Hello,