How to move trainer from cpu to gpu after .fit

[xwinxu]

I am trying to do some evaluations that require my model be on cpu. I am trying to make trainer.test() run on CPU after being trained on GPU with trainer.fit. What's the best way to do this without needing to reload the checkpoint into the state dictionary? I am trying to avoid reloading the state dictionary because I notice I have to go through the dictionary and manually prune out the prefixes for the sub-modules of the model since wrapping them inside PTL adds the module names as an extra component. Thanks!