f范数criterion = torch.nn.MSELoss(reduction='mean')1,2范数coef_q = torch.norm(Q,p =1,dim =1)loss_q = torch.norm(coef_q,p =2)