Loading and pushing multiple models to GPUs at the same

[rahulvigneswaran]

I have multiple pre-trained models at hand (one for each class). I have already loaded and frozen them.

Now I need to extract features from them and distill them into a new model.

For doing so, am adding all the models to a python dict so that I can use the class label to access the appropriate model and extract features from them. But the problem is that these pre-trained models inside the python dict are not moved automatically to GPUs. What am I doing wrong here?

[lecacosa]

Maybe we could add models to a nn.ModuleDict instead of a python dict.