PyTorch Module类的__call__函数比forward函数的优先级高

使用以下的代码证明：

import torch.nn as nn

class A(nn.Module):
    def __init__(self):
        pass
    def forward(self, x):
        return x
    def __call__(self, x):
        print("Start running __call__ function")
        return self.forward(x)

a = A()
t1 = torch.rand((2, 5))
print(a(t1))

程序输出：

Start running __call__ function
tensor([[0.9715, 0.2507, 0.9550, 0.9101, 0.9695],
        [0.0976, 0.8725, 0.7675, 0.9817, 0.7045]])