Pruning models in Pytorch: How to rewind weights?

Question

Sorry in advance for my question, I'm still fairly new to Deep Learning and Pytorch.

I'm looking for strategies to increase inference speed, and I came upon Pruning. If I understood correctly, this is an iterative process:

1. Train model
2. Prune
3. Set non pruned weights of original model back to the initial weight values (rewinding)
4. Train again this model
5. Repeat 2 - 4

Can anyone give an example, or guide me on how to perform these steps, in particular rewinding, in Pytorch?

I also have a second question: How can only setting some weights to 0 increase inference speed without modifying the model architecture itself?

I've been searching around but I can't find examples on how to perform rewinding.

Note that currently I have no minimal code example to share, since I'm still trying to figure out how to do this.

Thank you!