Created by: golovenkov
We do not need to evaluate a model each batch when there are no train metrics to evaluate, i.e. when using empty list for "train_metrics" config: https://github.com/deepmipt/DeepPavlov/blob/99e7d890b6d366337ac97477bf3d31c048370561/deeppavlov/configs/ranking/ranking_ubuntu_v1_mt_word2vec_smn.json#L88 It can significantly speed up the training time.