Retraining keras network (GRU model)

Hi fastai folks,
i have trained GRU neural net (keras) and saved weights ,when i want to train it with new upcoming data and do prediction for future , it starts to train from beginning ans shows (val_loss improved from inf to …) and saves model .i tried to save it to json and tried to reload it ,but still it shows same .is there any way that i can start training from the loss where i left ,instead of training it from start .(if the loss improved save it ,else use the older one)