i have a Issue with WebsiteSearchTool: Embedding Dimension Mismatch

[Jasonk0825]

I'm encountering an error while trying to use the WebsiteSearchTool with Gemini Pro as the LLM. The error message is:

"Embedding dimension 768 does not match collection dimensionality 1536. This is commonly a side-effect when an embedding function, different from the one used to add the embeddings, is used to retrieve an embedding from the database."

To resolve this, I tried changing the embedding model to "models/text-embedding-004", but the same error persists. Here's my current configuration:

web_rag_tool = WebsiteSearchTool(
    config=dict(
        llm=dict(
            provider="google",
            config=dict(
                model="gemini-pro",
                temperature=0.7,
            ),
        ),
    )
)
        embedder=dict(
            provider="google",
            config=dict(
                model="models/text-embedding-004",
                task_type="retrieval_document",
            ),
        ),
    )
)

I'm unsure why the error mentions 768 dimensions when I'm using "models/text-embedding-004", which should produce 1024-dimensional embeddings.

Questions:

Is there a compatibility issue between Gemini Pro and the current embedding setup in WebsiteSearchTool? Are there any known issues or additional configurations needed when using Google's embedding models with this tool? How can I ensure that the correct embedding model is being used and properly recognized by the tool? Any guidance on resolving this dimension mismatch would be greatly appreciated. Thank you!

[Jatayu-u]

Hey @Jasonk0825, I am facing a similar issue. Did you get a solution for this?

[brukted]

+1