Hybrid search with custom vector embeddings for query

Support
1

Description

I’m using a custom vectorizer that generates multi-vector embeddings (i.e., a 2D array of vectors) and storing them in a named vector field. When using the nearVector operator, everything works fine with the multi-vector format.

However, when I attempt to use the hybrid search with the same multi-vector format, it doesn’t seem to accept the 2D embedding array. My goal is to perform a hybrid search using both a text query and my custom multi-vector embedding.

Is hybrid search currently compatible with multi-vector embeddings in Weaviate? Or am I missing a specific configuration or usage requirement?

Server Setup

  • Weaviate version: 1.31.0

  • Vector configuration: Custom vectorizer with multi-vector embeddings enabled on a named vector field

Query Example

graphql

{
  Get {
    Article(
      limit: 5
      hybrid: {
        query: "Text query"
        vector: [[], [], []]  # multi-vector embedding
        targetVectors: ["embedding_text_vector"]
      }
    ) {
      title
      _additional {
        score
      }
    }
  }
}

Any guidance on whether this is supported—or how to make it work—would be greatly appreciated.

2

Hi @pawwi,

Unfortunately, this is not supported yet. However this can be a great feature request.

Could you please submit this as feature request?

Best regards,
Mohamed Shahin
Weaviate Support Engineer
(Ireland, UTC±00:00/+01:00)