use the same training parameters as before including the model name and it should pick up from where it left off