Install all requirements listed in requirements.txt.
Check arguments in data/save2json.py. Prepare data.
cd data
python save2json.py --subset train --encdec bart --pretrained_name facebook/bart-base --encrep first
python save2json.py --subset valid --encdec bart --pretrained_name facebook/bart-base --encrep first
cd ..
How to train?
Check all arguments needed in main.py, then run it
Check all arguments needed in evaluate.py, these aer generation parameters.
You can read on controlling sampling methods in How to Generate
It will generate responses from the save and save them comprehensively in generation_results/ in a json, where you will get corresponding scores - per sentence and averaged.