Troubleshooting Chroma Vector Store and Docstore Setup

Question

Hey, I'm running a chroma vector store and a very basic storagecontext and docstore setup. For some reason whenever I try to peek into my chromadb after indexing a handful of sample documents, it returns an empty dict as follows: {'ids': [], 'embeddings': array([], dtype=float64), 'documents': [], 'uris': None, 'data': None, 'metadatas': [], 'included': [, , ]}. When I print my docstore from index.docstore.docs, it states that I do have documents. I've been debugging it for a bit, playing around with persist paths and other configs but I can't seem to find where the problem resides

Logan M · Answer

The docstore shouldn't be populated unless you explicitly populate it?

Logan M · Answer

Sounds like you aren't passing in the chroma vector store properly and its defaulting the the simple vector store

OverclockedClock · Answer

When I'm persisting my index I see a deafult__vector_store.json, I guess that implies that it's using the simplevectorstore, right?

OverclockedClock · Answer

Hmmm, I am not adding any docs to my docstore anywhere as far as I'm concerned

Logan M · Answer

https://colab.research.google.com/drive/1CyhBujHkkodIM7m-JXd7vZp2uDGMo3wQ?usp=sharing

Logan M · Answer

Works fine for me

Logan M · Answer

You dont need to call persist or load

OverclockedClock · Answer

Ohh really?

Logan M · Answer

Yea, its also shown in the docs

Logan M · Answer

https://docs.llamaindex.ai/en/stable/examples/vector_stores/ChromaIndexDemo/#basic-example-including-saving-to-disk

Logan M · Answer

For most vector stores, everything gets tossed into the vector db itself

OverclockedClock · Answer

🤦‍♂️ apologies for not reading the manual properly

Logan M · Answer

no worries! There's so many docs pages

OverclockedClock · Answer

If I'm loading an index from storage though, don't I need to persist and load it from somewhere?

Logan M · Answer

thats really only for the defaults

Logan M · Answer

chroma is managing its own storage

OverclockedClock · Answer

So I can just re-initialize a chroma client, initialize chromavectorstore, and then create a vectorstoreindex and I'm set?

Logan M · Answer

Yes, using the from_vector_store() method

OverclockedClock · Answer

Oh my days

Logan M · Answer

One gotcha is make sure you keep the embedding model the same. Either setting Settings.embed_model or passing it in

.from_documents(..., embed_model=embed_model)
.from_vector_store(..., embed_model=embed_model)

OverclockedClock · Answer

Thank you so much, I'll take a look at it

Find answers from the community

Troubleshooting Chroma Vector Store and Docstore Setup