finetuned model correctly

fine-tuned for 30 epochs, reaches f1=0.77 on validation-set (original reached 0.5

corrected format