How did you get hard/soft negatives?
#1
by
binarymax
- opened
Hi! Just a general question on the training script and the gooaq dataset...
(1) Is GooAQ only positive pairs?
(2) Does train_st_gooaq.py automatically choose hard/soft negatives?
This looks like a good starting point to adapt for training the ModernBERT to my own positive pairs dataset but I'd like to learn a bit more on the effort to get negatives beforehand.
Thanks!