How does this compare to ms-marco cross encoders?
Thanks for sharing this model with the community.
I couldn't find a direct comparison of re-ranking performance with cross-encoder/ms-marco-MiniLM-L-12-v2 (and other models in the same family). Has anybody looked into it?
Thanks for your interest in our work!
The aim of bge-reranker is to train a multilingual model (mainly for English and Chinese currently), and you can see the training dataset at https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker. cross-encoder/ms-marco-MiniLM-L-12-v2 is only for english.
The current evaluation is based on Chinese data and cross-language data, lacking evaluation for English. We plan to extend the evaluation dataset and add more models in the future.
Is there any update about the comparation between cross-encoder/ms-marco-MiniLM-L-12-v2 and BAAI/bge-reranker-base? Thank you.
Is there any update about the comparation between cross-encoder/ms-marco-MiniLM-L-12-v2 and BAAI/bge-reranker-base? Thank you.
Apologies, but we have not conducted this comparison.
If you have any experimental results, we would greatly appreciate your feedback.