QdrantIndex loses client in composability

[Mikkolehtimaki]

When creating a ComposableGraph over a composite index of Qdrant indices, queries error in

File ...gpt_index/gpt_index/indices/query/vector_store/qdrant.py:61, in GPTQdrantIndexQuery.__init__(self, index_struct, embed_model, similarity_top_k, client, *
*kwargs)
     58     raise ValueError(import_err_msg)
     60 if client is None:
---> 61     raise ValueError("client cannot be None.")
     63 self._client = cast(qdrant_client.QdrantClient, client)

ValueError: client cannot be None.

Seems that the Qdrant client is not carried over to the composed index or graph. Created the query like this:

i1 = GPTQdrantIndex([d1], client=client, collection_name="i1")
i2 = GPTQdrantIndex([d2], client=client, collection_name="i2")
li = GPTListIndex([i1, i2])
graph  = ComposableGraph.build_from_index(li)

[Rohitjoshi9023]

Any update?

If anyone knows how we can use Vector Stores Indexes with ComposableGraph, please help https://stackoverflow.com/questions/75549286/gpt-index-issue-using-composablegraph-with-vector-stores

[Mikkolehtimaki]

I dug in a bit, the clients (I assume this affects weaviate as well) are not part of index_struct that is copied over in composability, causing the error. If I read it right :) May be a bigger change needed to fix it.

[jerryjliu]

@Mikkolehtimaki let me know if this is still an issue! apologies for the delay, 0.4.15 should have fixed this https://discord.gg/dGcwcsnxhU