#nn-lazylinear

Question 1: Single choice

easyAI-checkedPyTorch

net = nn.Sequential(nn.Linear(20, 256), nn.ReLU(), nn.LazyLinear(10))

Before any data has passed through net, which parameters have been initialized?

All of them, because LazyLinear uses a default input size of 1 until real data arrives1
Only the first layer's; the LazyLinear weights stay uninitialized until the first forward pass2
All of them, because once the first layer's input size is known PyTorch infers the rest at once3
None of them, because a single lazy layer makes the whole network wait for the first forward pass4