Open danyaljj opened 3 years ago
It looks good! I propose to merge it to a different branch (e.g. bart-hf-v4
) because re-runing finetuning with a newer version of HF is not tested. People can use HF v4 if they do not want to re-run finetuning but only want to run inference.
PR corresponding to #17