I want to reproduce the accuracy of paper, but it is difficult to reproduce the accuracy of paper because the performance of competent model is about 70%, which is significantly lower than paper.
Is it possible to get the checkpoints of COMPETENT MODEL and SCRATCH MODEL or the random seed values used?
Thanks for the good work.
I want to reproduce the accuracy of paper, but it is difficult to reproduce the accuracy of paper because the performance of competent model is about 70%, which is significantly lower than paper.
Is it possible to get the checkpoints of COMPETENT MODEL and SCRATCH MODEL or the random seed values used?