Hi ! quick question: If we create an auto-embedding field with

model_config

but provide pre-computed embeddings during import, will Typesense: 1. Use our provided embeddings (not regenerate) 2. Still convert text queries to embeddings at search time? We have ~500k pre-computed CLIP embeddings we don’t want to regenerate, but need text search to work.

@Sahir Patel Unfortunately, it's not possible to mix auto-generated embeddings with pre-computed embeddings (sending them over). The embeddings model at search time and on document indexing must be the same.

got it , even if the pre-computed are from the same

clip-vit-b-p32

model that typesense uses internally at search time?

@Sahir Patel Actually we do ignore auto generation of the embedding field if the vector value for that field already exists. For text query, it will continue to use auto embedding configuration.