Distance between entrypoint and query node

mklobucaric · December 20, 2024, 1:23pm

Description

I have created a sanbox in Weaviate Cloud to test functionality before putting the vectorstore in production. I am testing openai embeddings:
“text-embedding-3-small” and “text-embedding-3-large”.

I tried to change dimensions of “text-embedding-3-large” to 256, to see how sould retrieval go, but I got error below. “distance between entrypoint and query node: 256 vs 1536”

Can I change dimensions in Sandbox? Or do I have to setup something additional when I am doing hybrid or smilarity search?

This is my setup:

    client = weaviate.connect_to_weaviate_cloud(
        cluster_url=weaviate_url,
        auth_credentials=Auth.api_key(weaviate_api_key),
        headers={"X-OpenAI-Api-Key": openai_api_key},
        # headers={"X-Google-Studio-Api-Key": google_studio_api_key},
    )

        vectorizer_config = Configure.Vectorizer.text2vec_openai(
            model=model,
            model_version="3",
            dimensions=dimensions,
        )

        client.collections.create(
            name=collection_name,
            vectorizer_config=vectorizer_config,
            vector_index_config=Configure.VectorIndex.hnsw(
                distance_metric=VectorDistances.COSINE
            ),
            properties=properties,
        )

I have changed now everything to model: str = “text-embedding-3-small”, and dimensions: int = 1536, and now retrieval works fine…

Second somehow related question: Does, Google’s “text-embedding-004” embedding works in sandbox? I had earlier some trouble setting it up…

Third question, are this integrations necessary (since this integrations don’t support Google’s embeddings):

    # integrations = [
    #     Integrations.openai(
    #         api_key=openai_api_key,
    #         requests_per_minute_embeddings=3000,
    #         tokens_per_minute_embeddings=1000000,
    #     ),
    # ]
    # client.integrations.configure(integrations)

Server Setup Information

Weaviate Server Version:
Deployment Method: sanbox in Weaviate Cloud
Multi Node? Number of Running Nodes: no
Client Language and Version: Python v4
Multitenancy?: No

Any additional Information

"StatusCode.UNKNOWN\n\tdetails = “explorer: get class: vector search: object vector search at index …: vector search: knn search: distance between entrypoint and query node: 256 vs 1536: vector lengths don't match”\n\tdebug_error_string = "UNKNOWN:Error received from peer "

DudaNogueira · December 20, 2024, 1:59pm

hi @mklobucaric !!

Welcome to our community !!

This seems related to this bug we have recently discovered:

github.com/weaviate/weaviate

Hybrid search falling back to default vectorizer confs when dimensions is set

opened 08:15PM - 18 Dec 24 UTC

dudanogueira

bug

### How to reproduce this bug? Here is a reproducible code: ```python impor…t os import weaviate from weaviate import classes as wvc import weaviate.error_msgs client = weaviate.connect_to_local( headers={ "X-OpenAI-Api-Key": os.getenv("OPENAI_APIKEY", "CHANGE_ME"), } ) print(f"Client: {weaviate.__version__}, Server: {client.get_meta().get('version')}") client.collections.delete("Test") collection = client.collections.create( name="Test", vectorizer_config=wvc.config.Configure.Vectorizer.text2vec_openai( model="text-embedding-3-large", dimensions=1024, #type_="text", vectorize_collection_name=False ), properties=[ wvc.config.Property( name="text", data_type=wvc.config.DataType.TEXT, tokenization=wvc.config.Tokenization.WORD ) ] ) # Create a single object response = collection.data.insert( properties={ "text": "COVID-19 has many symptoms." } ) # objects indeed has 1024 dimensions response = collection.query.fetch_objects( limit=5, include_vector=True ) for obj in response.objects: print( f"fetch_objects: {obj.uuid} ({len(obj.vector['default'])}) | Properties: {obj.properties}") # you can perform a neartext response = collection.query.near_text( query="hybrid query with 1024 dimensions", #alpha=0.75, limit=5, include_vector=True ) for obj in response.objects: print( f"near text query: {obj.uuid} ({len(obj.vector['default'])}) | Properties: {obj.properties}") #but it fails to hybrid try: response = collection.query.hybrid( query="hybrid query with 1024 dimensions", alpha=0.75, limit=5, include_vector=True ) for obj in response.objects: print( f"hybrid query: {obj.uuid} ({len(obj.vector['default'])}) | Properties: {obj.properties}") except Exception as e: print("ERROR!!!", e) # if we close the client client.close() # and point it to a catch endpoint client = weaviate.connect_to_local( headers={ "X-OpenAI-Api-Key": os.getenv("OPENAI_APIKEY", "CHANGE_ME"), "X-OpenAI-BaseUrl": "https://webhook.site/beef60de-4d45-4c61-9928-b20fa619f91e", } ) collection = client.collections.get("Test") response = collection.query.hybrid( query="hybrid query with 1024 dimensions", #alpha=0.75, limit=5, include_vector=True ) for obj in response.objects: print( f"hybrid query: {obj.uuid} ({len(obj.vector['default'])}) | Properties: {obj.properties}") # we get this payload payload = { "input": [ "hybrid query with 1024 dimensions" ], "model": "text-embedding-3-small", "dimensions": 1536 } ``` ### What is the expected behavior? The hybrid search should work. It should generate the query vectorization payload as: ```json { "input": [ "hybrid query with 1024 dimensions" ], "model": "text-embedding-3-large", "dimensions": 1024 } ``` ### What is the actual behavior? The generated payload to vectorize a hybrid query is passing the wrong model and dimension as the payload: ```json { "input": [ "hybrid query with 1024 dimensions" ], "model": "text-embedding-3-small", "dimensions": 1536 } ``` ### Supporting information Client: 4.10.2, Server: 1.28.1 ### Server Version 1.28.1 ### Weaviate Setup Single Node ### Nodes count 1 ### Code of Conduct - [X] I have read and agree to the Weaviate's [Contributor Guide](https://weaviate.io/developers/contributor-guide) and [Code of Conduct](https://weaviate.io/service/code-of-conduct)

Can you confirm that near_text will work, but the failing query is a hybrid one?

Thanks!

mklobucaric · December 20, 2024, 3:16pm

Thank you!

Yes, you are right, nearText works but hybrid not…

Topic		Replies	Views
Hybrid Queries on new OpenAI Embedding Models failing server restart Support	15	369	January 8, 2025
Simple keyword search not working Support	4	981	September 14, 2023
Similarity search returns chunks that all have exactly the same distance value Support bug	3	636	November 29, 2023
Handling Varying Vector Sizes in Weaviate Indexing: Ensuring Consistency and Error Prevention Support	6	484	October 10, 2024
Indexing embeddings taking too long. What am I doing wrong? Support	4	1221	September 27, 2024

Distance between entrypoint and query node

Description

Server Setup Information

Any additional Information

Related topics