Fully connected input size

I have the following error, that seems related to my model initialization.

RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling `cublasSgemm( handle, opa, opb, m, n, k, &alpha, a, lda, b, ldb, &beta, c, ldc)
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 1, kernel_size=7, padding=3)
        self.conv2 = nn.Conv2d(1, 8, kernel_size=7, padding=3)
        self.conv3 = nn.Conv2d(8, 8, kernel_size=7, padding=3, stride=2)
        self.conv4 = nn.Conv2d(8, 8, kernel_size=7, padding=3, stride=2)
        self.conv5 = nn.Conv2d(8, 16, kernel_size=7, padding=3, stride=2)
        self.conv6 = nn.Conv2d(16, 32, kernel_size=7, padding=3, stride=2)
        self.conv7 = nn.Conv2d(32, 32, kernel_size=7, padding=3, stride=2)
        self.fc1 = nn.Linear(32*16*16,256) #Error : size ? 
        self.fc2 = nn.Linear(256,48)

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.conv6(x)
        x = self.conv7(x)
        x = self.fc1(x)
        x = self.fc2(x)
        return x

device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # PyTorch v0.4.0
model = Net().to(device)
summary(model, (1, 512, 512))

But it seems that my fully connected 1 is incorrect.
Anyone knows why?

Hello @Alexandra_Lorenzo

That error looks very uninformative, I think this is because of quite old version used (# PyTorch v0.4.0).

What I can see from you code that Linear takes output from Conv2d without any modifications.

x = self.conv7(x)
x = self.fc1(x)

This is definitely cause shape mismatch error. Linear takes 3d input, while Conv2d outputs 4d tensors. Please consider using poolings or flatening before dense layer