Custom sentence pairs sampling

[mariusmercier]

Hello,

Is there a way of inputting directly a custom pair sampling?

My model takes too much time when doing "oversampling", so I am using "undersampling", however I have specific negative pairs that I would like to include.
For each category I have a text that is the label of the category, so I would like that each labels are included in the negative pairs.

Thanks !