Dust Community Icon

Evaluating the Embedding Model for Multimodal Support in Dust

·
·

TOPIC: Embedding Model in Dust. With all the multimodal models (text, audio, video, images) coming online and available on Dust, I am wondering whether we need to change the embedding model as well ? Currently the embedding model provider is preselected to be OpenAI. But I am not sure whether this preselected model is embedding pure text representations or more than that. I am not even sure whether is can result in an issue. Could someone give some guidance on how to think about it? Thank you,

  • Avatar of Remi
    Remi
    ·
    ·

    Thanks for you question Alexander AP Preis ! TLDR: I suggest staying with the default model. The main limitation is that Dust will only use text content or PDF turned to text to generate answers. Ex: it won’t work if you have a a screenshot of a org chart in a Notion page. If that case, I suggest having the content as text in a toggle below the image.