#shape-mismatch

Question 1: Single choice

easyAI-checkedPyTorch

net = nn.Sequential(nn.Linear(20, 256), nn.ReLU(), nn.Linear(128, 10))
X = torch.rand(2, 20)

The second linear layer expects 128 inputs, but the first one produces 256. What happens?

Building net already raises an error, because nn.Sequential checks neighbouring layers1
net(X) runs, because PyTorch resizes the second layer to 256 inputs as a lazy layer would2
net(X) runs, and PyTorch silently uses only the first 128 of the 256 hidden values3
Building net works; calling net(X) raises a shape-mismatch error at the second layer4