Closed dhimanpd closed 6 years ago
@dhimanpd Can you post the traceback? You should be able to save weights from the multigpu parallelized model. Load weights during serial model defintion. Refer to def make_model_full
in the cifar10_cnn_mgpu.py
examples. Save the multigpu model's weights either by using a callback or save_weights after training.
Load the weights into the serial model. Also, for json save the serial_model's json. Then load it, and parallelize it with make_parallel. Saving weights from the parallelized multigpu model should work, if not please post the traceback and maybe a snippet of your code logic (parts where you load/save json and weights).
Hi, How can I save the model definition? Can you provide some info on how to save the model definition and reload model using json and load_weights. I am getting an exception when saving model definition.