Closed XiaoqingNLP closed 6 years ago
Early stopping can be achieved in two ways in THUMT:
--validation <your validation source file> --references <your validation reference files>
is set in training, THUMT will evaluate on the validation set periodically. The checkpoints that achieves highest BLEU scores on the validation set will be saved in the train/eval folder. (You can halt or resume the training whenever you want.)keep_checkpoint_max
in parameters to reserve as many checkpoints as you want.
the early stopping to prevent overfitting is very useful to train a NMT model,but I didn't found it in thumt transformer.