Hi,

First, thanks for all the recent developments on the library!

I try to mine hard examples using sentence_transformers.util.mine_hard_negatives. I am using a reranker which slows down the process.
As I have a 4 GPUs machine, I would like to use them all to speed up. But even by putting the argument use_multi_process to True, during the rescoring with the cross-encoder, only 1 GPU is used.

Has anyone else experienced the same issue?

Thanks!