Here we provide instructions for adding models to the repository. It is pretty simple.
First, create a file with your model's architecture. You are free to name it anything. Additionally, you can include other util files for the model here.
The second and last step involves making our config parser recognize your model file. To do this, first add an elif to the "prepare_models" function in the utils, that checks whether the model name provided in the experiment config matches the one for your new model. Second, add a function called "prepare_" that will handle all model preparation. You can see our currently implemented "prepare_" functions for examples, but all this function needs to contain is the import from your model file, as well the capability to load your pretrained base models, and initialize a "new" model which will act as the merged model. You may follow the template provided by the other prepare_ functions in the "prepare_models" utils.py function.