trainer.resume with finetuning

[taghreed34]

Is trainer.resume() valid when finetuning is interrupted or it's just used with training? And if so, is using adam optimizer handled so momentum and scale vector aren't initialized from scratch?