Open
Description
For each new implementation, I need to copy and paste the visualizer.py and train.py. This seems kind of redundant, especially since the training loop is nearly the exact same for most of the models implemented so far.
I should at least create a parent class with a template train method, which I can pass in methods I write for forward passes and loss computations. This way, I can simply subclass this parent class for each model, to avoid having duplicate train loops.
Metadata
Metadata
Assignees
Labels
No labels