consult about the train_student.py

Ty you for your great work on reproducing the splitmean flow matching code. Can you give more training details of the train student.py. Because when I used the default setting, it can not generate a good result but a noisy one. 

Thank you
