The definition of the split learning operations #6
Description
Hello,
Thank you for making your framework available.
I have a question regarding the Split Learning implementation.
There are SplitModel classes for several LM models. However, what I noticed is that when a model is created in the code, there is only one 'model' object. Moreover, during training, only a single forward() function call is used. Theoretically, when using SL, one would expect to see a list of model objects and a successive call of forward() (one for each model part). Additionally, one would expect the transmission of the activations/gradients. I understand that this code is a simulator, but it seems that the model remains as a whole.
I apologize if I have not understood the code properly, but where can I find the definition of these SL operations?