#here is my code
#i use subprocess to train, then the problem is i can not add --eval-bleu
#everytime i add bleu score line, it throws an unknown error,
#and i did not find any accuracy metrics in fairseq. where can i add
#custom accuracy or how can i run training code with bleu score? I think
#ppl score is not efficient like bleu or accuracy.
import subprocess
# Train each model in the ensemble
command = [
'python3',
'fairseq/train.py',
'--optimizer', 'adam',
'--lr', '5e-5',
'--min-lr', '5e-10',
'--lr-shrink', '0.5',
'--max-tokens', '1000',
'--arch', 'transformer',
'--clip-norm', '0.5',
'--max-epoch', '30',
'--save-dir', 'my_test_checkpoint/cyrillic2bichig2',
'cyrillic2bichig-bin'
]
subprocess.run(command)