Super light weight architecture for ImageNet

Hello, I’m looking for a super light weight baseline architecture in PyTorch that achieves some level of performance on ImageNet (i.e. 30-40% top-1). I am aware of SqueezeNet, MobileNet etc. but I am looking for something even simpler with a lower requirement for accuracy. The idea is to simply provide a reference for how introductory textbook architectures (not too far from a simple CNN) can perform on this task.

Does something like this already exist (ideally with the architecture already written in PyTorch)?