The main requirements are:
Get pre-trained BERT model
Download BERT-BASE-CASED and put it under ./pretrained
.
Train and select the model
python run.py --dataset=WebNLG --train=train --rounds=4
python run.py --dataset=WebNLG_star --train=train --rounds=2
python run.py --dataset=NYT24 --train=train --rounds=3
python run.py --dataset=NYT24_star --train=train --rounds=2
python run.py --dataset=NYT29 --train=train --rounds=3
Evaluate on the test set
python run.py --dataset=WebNLG --train=test --rounds=4
python run.py --dataset=WebNLG_star --train=test --rounds=2
python run.py --dataset=NYT24 --train=test --rounds=3
python run.py --dataset=NYT24_star --train=test --rounds=2
python run.py --dataset=NYT29 --train=test --rounds=3
Parts of our codes come from bert4keras.