Qdrant query

SSeaCat

Hello! Thanks for the great library! I have a question about using Qdrant. I was able to save the index to qdrant collection but I can't figure out how to query it after. I found this code but stuck then

Plain Text

qdrant_vector_store = QdrantVectorStore(collection_name="test", client=qdrant_client)
query = VectorStoreQuery(query_str='Did Y Combinator have its name always?')

What should I do now? (The complement question is: should I save each separate index (based on a different set of documents) into a separate collection?) THANKS!

8 comments

LLogan M

Have you seen this example using qdrant?

https://github.com/jerryjliu/llama_index/blob/main/examples/vector_indices/QdrantIndexDemo.ipynb

SSeaCat

Yes, of course! But it doesn't show how I query the indices stored in the database, after. Let me describe the use case:
I'm working on an app allowing users to load their custom data and then query them. I'm building index (exactly as shown in this example), then save to the qdrant. As I should provide a unified code for every single user, at them moment they will query their own index, I have no idea which index they will query.
I assume I can create a separate collection for each index (say, based on the user's dataset's ID or so) and then I will have to find this collection (what is easy) and query it. So, how to do that?
I found this example for Qdrant:

Plain Text

    response = openai.Embedding.create(
        input="What is starship?",
        model="text-embedding-ada-002"
    )
    embeddings = response['data'][0]['embedding']

    search_result = qdrant_client.search(
        collection_name="mycollection",
        query_vector=embeddings, 
        limit=5
    )

I suppose LlamIndex should have something like that?

LLogan M

To later query an existing index that you saved on qdrant, im pretty sure you can do something like this

index = GPTQdrantIndex([], client=client, collection_name="collection_name")

index.query("my query")

SSeaCat

Thanks! I will try tomorrow and let you know how it works.

SSeaCat

Yep, it worked!! Back to my second question, should I create a separate collection for each index built from a specific set of documents? One consideration is the index dimensions - a collection may not have indices with different dimensions and from LlamaIndex docs and code it's not clear if it's always the same or not. Thanks!

LLogan M

Hmm, What do you mean by index dimensions?

If you have a system where different users are uploading documents, I think it makes sense to create an index/collection for each user

SSeaCat

I meant the dimensions of vector (I guess of embeddings), and from the qdrant documentation "A collection is a named set of points (vectors with a payload) among which you can search. Vectors within the same collection must have the same dimensionality and be compared by a single metric."

LLogan M

Right! As long as you use a consistent embedding model, the dimensions should always match (openAI embeddings are 1536 dimensions)

Add a reply

Find answers from the community

Qdrant query