Cool work! Considering using a biomedical embedding model?

[Andy-jqa]

Thank you for your wonderful work. It seems that the current system uses OpenAI's embedding models for dense retrieval, which might be sub-optimal (and costly). We have a biomedical embedding model trained by 255M real PubMed user logs. Would appreciate it if you could check out:

Paper: https://academic.oup.com/bioinformatics/article/39/11/btad651/7335842
Query encoder: https://huggingface.co/ncbi/MedCPT-Query-Encoder
Article encoder:https://huggingface.co/ncbi/MedCPT-Article-Encoder