[Question]: Constructing hybrid indices with Qdrant.

[gich2009]

Question Validation

• [X] I have searched both the documentation and discord for an answer.

Question

I get this error when I try to run a pipeline with qdrant as the vector store and the enable_hybrid=True. Am I missing a parameter or is this a bug?

File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/core/instrumentation/dispatcher.py", line 260, in wrapper result = func(*args, **kwargs) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/core/ingestion/pipeline.py", line 555, in run self.vector_store.add(nodes_with_embeddings) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 407, in add self._create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 643, in _create_collection raise exc # noqa: TRY201 File "/home/gich2009/Work/BAYESNET/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 616, in _create_collection self._client.create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/qdrant_client.py", line 2081, in create_collection return self._client.create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/qdrant_remote.py", line 2582, in create_collection sparse_vectors_config = RestToGrpc.convert_sparse_vector_config( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3474, in convert_sparse_vector_config map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3474, in map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3465, in convert_sparse_vector_params cls.convert_modifier(model.modifier) if model.modifier is not None else None File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3454, in convert_modifier raise ValueError(f"invalid Modifier model: {model}") ValueError: invalid Modifier model: none

[dosubot[bot]]

You are not missing any parameters when trying to run a pipeline with Qdrant as the vector store and enable_hybrid=True. The setup for enabling hybrid search with Qdrant is correct as shown in the example:

from llama_index.core import VectorStoreIndex, StorageContext
from llama_index.vector_stores.qdrant import QdrantVectorStore
from qdrant_client import QdrantClient, AsyncQdrantClient

# creates a persistent index to disk
client = QdrantClient(host="localhost", port=6333)
aclient = AsyncQdrantClient(host="localhost", port=6333)

# create our vector store with hybrid indexing enabled
vector_store = QdrantVectorStore(
    "llama2_paper",
    client=client,
    aclient=aclient,
    enable_hybrid=True,
    batch_size=20,
)

storage_context = StorageContext.from_defaults(vector_store=vector_store)
Settings.chunk_size = 512

index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context,
)

If you are encountering issues, it might be due to other factors not covered in the provided context. However, based on the given information, your setup appears to be correct ^[1].

Additionally, there are known issues or limitations documented for using Qdrant as a vector store with hybrid indexing in LlamaIndex. Specifically, if hybrid search is not enabled, attempting to perform a hybrid search will raise a ValueError:

if query.mode == VectorStoreQueryMode.HYBRID and not self.enable_hybrid:
    raise ValueError(
        "Hybrid search is not enabled. Please build the query with "
        "`enable_hybrid=True` in the constructor."
    )

This indicates that hybrid search must be explicitly enabled during the construction of the query; otherwise, it will not function correctly ^[2][3].

The error you are encountering seems to be related to the convert_modifier method in the qdrant_client library, which raises a ValueError for an invalid Modifier model. This might be an issue with the Qdrant client library itself, and you may need to check the specific configurations or versions you are using.

To continue talking to Dosu, mention @dosu.

[gich2009]

This is exactly what I am doing @dosubot

from llama_index.core import VectorStoreIndex, StorageContext from llama_index.vector_stores.qdrant import QdrantVectorStore from qdrant_client import QdrantClient, AsyncQdrantClient

creates a persistent index to disk

client = QdrantClient(host="localhost", port=6333) aclient = AsyncQdrantClient(host="localhost", port=6333)

create our vector store with hybrid indexing enabled

vector_store = QdrantVectorStore( collection_name=collection_name, client=client, aclient=aclient, enable_hybrid=True, batch_size=20, prefer_grpc=True, enable_hybrid=True, )

storage_context = StorageContext.from_defaults(vector_store=vector_store) Settings.chunk_size = 512

index = VectorStoreIndex.from_documents( documents, storage_context=storage_context, )

Has anyone else encountered this issue?

[logan-markewich]

I think its a bug with the handling added to bm42

For whatever reason, qdrants enum for the modifier has a NONE option, but apparently doesn't handle it? Very odd

        sparse_config = self._sparse_config or rest.SparseVectorParams(
            index=rest.SparseIndexParams(),
            modifier=(
                rest.Modifier.IDF
                if self.fastembed_sparse_model and "bm42" in self.fastembed_sparse_model
                else rest.Modifier.NONE
            ),
        )

[logan-markewich]

Actually I take it back, I wasn't able to replicate this in a fresh venv/google colab

Could be an issue with your qdrant client version? https://colab.research.google.com/drive/1TtsNPPLxLplvlvItrjBp7wE8oBrLLfgp?usp=sharing

[gich2009]

Let me try again @logan-markewich

[gich2009]

Hey @logan-markewich, it works for me too when I use an in memory client. Could you try using a managed Qdrant client?

[gich2009]

The error occurs consistently for me when I use a managed client.

File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/core/instrumentation/dispatcher.py", line 260, in wrapper result = func(*args, **kwargs) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/core/ingestion/pipeline.py", line 555, in run self.vector_store.add(nodes_with_embeddings) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 407, in add self._create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 643, in _create_collection raise exc # noqa: TRY201 File "/home/gich2009/Work/BAYESNET/.venv/lib/python3.10/site-packages/llama_index/vector_stores/qdrant/base.py", line 616, in _create_collection self._client.create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/qdrant_client.py", line 2081, in create_collection return self._client.create_collection( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/qdrant_remote.py", line 2582, in create_collection sparse_vectors_config = RestToGrpc.convert_sparse_vector_config( File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3474, in convert_sparse_vector_config map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3474, in map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3465, in convert_sparse_vector_params cls.convert_modifier(model.modifier) if model.modifier is not None else None File "/home/gich2009/Work/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3454, in convert_modifier raise ValueError(f"invalid Modifier model: {model}") ValueError: invalid Modifier model: none

[logan-markewich]

Do you need to bump the version of your deployed instance? Let me spin one up and try again

[gich2009]

I'm already on the latest version according to the console.

v1.11.0

[gich2009]

Hey @logan-markewich , I figured out what the issue is. The error occurs when prefer_grpc=True and enable_hybrid=True. enable_hybrid=True and prefer_grpc=False does not trigger the exception.

This is also true for a self hosted Qdrant container.

Not sure if it is the case that grpc cannot be used with hybrid indices.

[pydv9991]

hi @logan-markewich @gich2009 @joein I am also facing the issue when grpc = True , all is working fine with rest client but when i want to use prefer_grpc=True it gives error: using llama-index==0.10.67.post1 and qdrant client 1.11.0, sparse_vectors_config = RestToGrpc.convert_sparse_vector_config( File "/home/ubuntu/prashant/codex_genai/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3302, in convert_sparse_vector_config map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/ubuntu/prashant/codex_genai/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3302, in map=dict((key, cls.convert_sparse_vector_params(val)) for key, val in model.items()) File "/home/ubuntu/prashant/codex_genai/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3293, in convert_sparse_vector_params cls.convert_modifier(model.modifier) if model.modifier is not None else None File "/home/ubuntu/prashant/codex_genai/.venv/lib/python3.10/site-packages/qdrant_client/conversions/conversion.py", line 3282, in convert_modifier raise ValueError(f"invalid Modifier model: {model}") ValueError: invalid Modifier model: none

[joein]

I think its a bug with the handling added to bm42

For whatever reason, qdrants enum for the modifier has a NONE option, but apparently doesn't handle it? Very odd

        sparse_config = self._sparse_config or rest.SparseVectorParams(
            index=rest.SparseIndexParams(),
            modifier=(
                rest.Modifier.IDF
                if self.fastembed_sparse_model and "bm42" in self.fastembed_sparse_model
                else rest.Modifier.NONE
            ),
        )

It's indeed looks like a bug in conversion of modifier struct in qdrant-client, we'll fix it

However, the code you provided for the modifier also contains a bug, since bm42 is not the only model which requires an IDF modifier. If it is supposed that fastembed is installed, you can use qdrant_client.qdrant_fastembed.IDF_EMBEDDING_MODELS for a check, otherwise, you might need to check the models manually, and then you also need to add bm25.

It should also be possible just to replace rest.Modifier.NONE with a regular None, since modifier is an optional field

cc: @Anush008