Keras takes a very long time to make the first prediction after model.load ()

Question

I load the model into keras with model.load () and find that the first prediction takes more than 10 times longer than calculated than to follow the predictions, any ideas why this might happen, or suggestions, it would be highly recognized that speeding up the prediction cycle entry level will be greatly appreciated.

I am using a Tensorflow server with processor processing.

Thanks for the help, Denym

Answer 1

Ok, so I found an answer that works for me:

, keras model.load, json/yaml .h5 keras.

model.load 5 , , .

json .h5 10 100 , , , , , .