Add ability to train from pre-trained model or resume training from checkpoint

[yahskapar]

See title, this should be as simple as adding a new config parameter or two and updating the init method of various trainer files.