add --fine-tune arg to train.py and test.py

[YiweiJiang2015]

add fine tune toggle to avoid saved config file being overridden by outer config during inference