[Question] How to use cache model (inception-2015-12-05.pkl) for metric

Due to network limitation, the training will be stopped since it cannot download the inception-2015-12-05.pkl for metrics. I download the .pkl off-line, how can I direct the train.py to use the cached inception-2015-12-05.pkl rather than trying to download?
thanks!