Results ELECTRA RoBERTa + CNN klue/RoBERTa (base) epochs 15, batch_size 16 val_loss = 0.172, val_pearson = 0.921 klue/RoBERTa + CNN_layers epochs 15, batch_size 16 val_loss = 0.154, val_pearson = 0.926 sRoBERTa_Large klue/RoBERTa (base) epochs 200, batch_size 32 val_loss = 0.106, val_pearson = 0.9754