zhongxifang
Results
1
comments of
zhongxifang
> 1. During fine-tuning, the number of hard negatives for each query is train_group_size - 1, so a larger train_group_size is better. > 2. During fine-tuning, all in-batch passages across...