SamKG

Yes, by 'refer' I mean as a pointer to the ctype object. What would prevent me from modifying the data directly?

@adamcrume I'd be happy to take a stab at fixing this - any pointers on where to start? Is it as simple as using the `add_inputs` method I mentioned earlier?

@kmantel I'd appreciate a second look at this - not sure if there's a better way to accomplish this, and my scheduler workaround may be a bit hacky

seems like a limitation with Pytorch; is there a specific usecase for deepcopying the pytorch model? ( @kmantel )

non-stateful pytorch representations may break things, since the initial parameters of the network are copied from the current context (but are never copied again afterwards). it should be fine to...

retest this please

retest this please

retest this please

retest this please

retest this please