How to replace the embedded model for a knowledge graph that has completed all the building steps？

[YumeAyai]

I have constructed a knowledge graph using a general-purpose embedding model. However, the model performs poorly in my specific domain or use case. I now need to replace it with a new, more suitable embedding model. How can I update the embedding vectors in my existing vector database to reflect the new model while maintaining the integrity of the knowledge graph?

[ntsarb]

I have constructed a knowledge graph using a general-purpose embedding model. However, the model performs poorly in my specific domain or use case. I now need to replace it with a new, more suitable embedding model. How can I update the embedding vectors in my existing vector database to reflect the new model while maintaining the integrity of the knowledge graph?

To the best of my knowledge, it is not possible, in which case you have to re-index the source material with the new model.

[oscarhoffmann3487]

You can rerun the text embedding step of your pipeline using a new embedding model. To do this, simply add a generate_text_embeddings entry under workflows in your settings.yaml file. This triggers the pipeline to regenerate embeddings directly, without needing to reprocess all other steps.