Save checkpoint with minimal overhead

[standard-aaron]

I want to save a checkpoint of my model with absolutely no training whatsoever (i.e. just with randomly initialized weights) AFAICT there's no way to do this without invoking a Trainer method that requires arguments/setup beyond the model itself:

e.g.
-.predict that requires setting up a DataLoader, Dataset, etc.
-.fit also requires a DataLoader (either as an argument or as a method to the mdoel)

Is there a simpler way to save a checkpoint that won't require mocking up any data?