semi-supervised-ImageNet1K-models | Semi-supervised ImageNet1K models | Dataset library

I'm trying to run a Resnet-like image classification model on a CPU, and want to know the breakdown of time it takes to run each layer of the model.

The issue I'm facing is the github link https://github.com/facebookresearch/semi-supervised-ImageNet1K-models has the model saved as a .pth file. It is very large (100s of MB), and I don't know exactly how it differs from pytorch except it's binary. I load the model from this file using the following script. But I don't see a way to modify the model or insert the t = time.time() variables/statements in between model layers to break down the time in each layer.

Questions:

1. Would running the model in the following script give a correct estimate of end-to-end time (t2-t1) it takes to run the model on the CPU, or would it also include pytorch compilation time?

2. How to insert time statements between consecutive layers to get a breakdown?

3. There is no inference/training script at the github link and only has the .pth file. So how exactly is one supposed to run inference or training? How to insert additional layers between consecutive layers of the .pth model and save them?