Thewillman

Can you show all the prompt take in the whole BOLAA?

1

When I try to reproduct this work,I find that chatgpt can continue write the whole procedure of BOLAA without access any api but not answer the question directly in hotpotqa

Has anyone face with problems that DPO rewards accuracy stuck at 0.5 and the loss stuck at 0.6 to 0.8?

5

I just reconstruct the order of my dataset, I make the triplet whose prompt are similar to be put into a list (I want to add new loss function, but...