pytorch 学习：torch.mean()

U = U_w + U_j  # b*c*es*1
print("concat:", U.shape)
p = torch.mean(U, dim=-1, out=None)  # b*c*es
print(Ugp.shape)