RAG Pipeline Creation
#1
by
Vasanth
- opened
Is it possible to create a RAG pipeline with this model as an alternative to ColPali series of model on a corpus of documents?
@Vasanth
Thanks for your interest in our work!
VLM2Vec is a general multimodal representation model, with some of its training data being similar to the Text-Image retrieval task. Additionally, our evaluation dataset includes subsets adapted from existing document retrieval dataset, such as https://huggingface.co/datasets/Tevatron/wiki-ss-nq. Therefore, we believe VLM2Vec could be utilized in a RAG pipeline. We also plan to conduct further testing in this direction in the future (for example compare with Colpali on document retrieval task).
wenhu
changed discussion status to
closed