Optimizer: scheduler

I am using MultiStepLR to change the learning rate which is great but there is a problem:

First let me sketch the context: I am using Colab to train, validate and test large networks (currently ResNet44 architectures). This obstructs me from doing the whole training process at once: I have to save the network from time to time. Since I am using MultiStepLR, it has to remember somehow in which epoch it was. My question is therefore, how does MultiStepLR knows which epoch it is and can I change that knowledge?