finetune and use memonger get no memory saving

[BrettLL]

when I use your code to finetune resnet152 and add code:
if args.finetune:
symbol, arg_params, aux_params = mx.model.load_checkpoint(args.pretrained_model, args.model_load_epoch)
(symbol, arg_params) = get_fine_tune_model(symbol, arg_params, args.num_classes, 'flatten0')
finetune result is ok.
But when I use memonger option, it cost the same gpu memory as I don't use this one.Does it mean that when I finetune a net and I can not use option to save memory?? thanks~

[tornadomeet]

memonger should also support fine-tune if set correctly.