[ 仅 API 调用方式不一致 ]torch.nn.Module

torch.nn.Module(*args, **kwargs)

paddle.nn.Layer(name_scope=None, dtype='float32')

# PyTorch 写法
class MLP(torch.nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(MLP, self).__init__()
        self.fc1 = torch.nn.Linear(input_size, hidden_size)
        with torch.no_grad():
            self.fc1.weight.fill_(1.0)
            self.fc1.bias.fill_(0.1)

    def forward(self, x):
        x = self.fc1(x)
        return x

# Paddle 写法
class MLP(paddle.nn.Layer):
    def __init__(self, input_size, hidden_size, output_size):
        super(MLP, self).__init__()
        self.fc1 = paddle.nn.Linear(in_features=input_size, out_features=hidden_size)
        with paddle.no_grad():
            self.fc1.weight.fill_(1.0)
            self.fc1.bias.fill_(0.1)

    def forward(self, x):
        x = self.fc1(x)
        return x