How to use func::gen_upper_triangle

[XinDongol]

Why we want to use gen_upper_triangle like

SDNet/Utils/CoQAUtils.py

Line 80 in adef469

Get upper triangle matrix from start and end scores (batch)

instead of just calculating start_loss and end_loss separately like

start_loss = loss_fct(start_logits, start_positions)
end_loss = loss_fct(end_logits, end_positions)

https://github.com/huggingface/pytorch-transformers/blob/44dd941efb602433b7edc29612cbdd0a03bf14dc/pytorch_transformers/modeling_bert.py#L1250