bertmodel=HooshvareLab/bert-base-parsbert-uncased dataset=SeyedAli/Persian-Text-Emotion epoch=6 Training Loss =0.000300