[ 仅 API 调用方式不一致 ]torch.autograd.function.FunctionCtx.saved_tensors

torch.autograd.function.FunctionCtx.saved_tensors

paddle.autograd.PyLayerContext.saved_tensor()

# PyTorch 写法
class cus_tanh(torch.autograd.Function):
    @staticmethod
    def forward(ctx, x):
        y = torch.nn.functional.tanh(x)
        ctx.save_for_backward(x, y)
        return y

    @staticmethod
    def backward(ctx, dy):
        x, y = ctx.saved_tensors
        grad = y + dy + 1
        return grad

# Paddle 写法
class cus_tanh(paddle.autograd.PyLayer):
    @staticmethod
    def forward(ctx, x):
        y = paddle.nn.functional.tanh(x=x)
        ctx.save_for_backward(x, y)
        return y

    @staticmethod
    def backward(ctx, dy):
        x, y = ctx.saved_tensor()
        grad = y + dy + 1
        return grad