EfficientNet | Paper Walkthrough & PyTorch Implementation

[@sharmakartikeya]

I wanted to thank you for the implementation. You really helped me implementing this myself. However, there are a few things I noted:
1. When I print the parameters for all models from version B1 to B7, they come out to be far less than that claimed by the paper. Any reasons for that or am I missing something?
2. As in ResNets, the strides are done in the first occurrence of every block. Other implementations of EfficientNet have also incorporated the stride=2 convblocks in the first occurrence as opposed to your implementation where you have taken it as the last occurrence. Any reasons for this?
For example, here is an excerpt from MobileNet V2 paper -
All layers in the same sequence have the same
number c of output channels. The first layer of each
sequence has a stride s and all others use stride 1