shorten the information in tqdm during training ML algorithms based on the log level

[Ghostvv]

Description of Problem:
currently DIETClassifier output a lot of info in tqdm:

Epochs:  11%|▍   | 9/80 [03:58<30:47, 26.02s/it, t_loss=6.703, i_loss=0.451, entity_loss=0.044, role_loss=0.053, i_acc=0.961, entity_f1=0.883, role_f1=0.717]

which makes it unusable for small sized terminals.
we should consider removing some information like loss if log level is info
also consider shortening entity_loss=0.044, role_loss=0.053 to e_loss=0.044, r_loss=0.053

Overview of the Solution:

Examples (if relevant):

Blockers (if relevant):

Definition of Done: