I am not able to reproduce the rouge scores reported in the BART paper with the Xsum dataset. I am writing this to re-confirm the hyperparameters used to reproduce the results in the paper with the Xsum dataset. It will be of great help if anyone can point out the exact hyperparameters used for training to replicate the results in the paper.
Here is my version of the code. BART-training/no_augCheck_Xsum.py at main · Sai-Ashish/BART-training · GitHub