Why is keras training slow?

[chenzi]

I use aws ec2 p2.xlarge server 12G gpu 60G Memory， But training is still slow.

[chenzi]

I get an error RuntimeWarning: Can save best model only with val_acc available, skipping.
'skipping.' % (self.monitor), RuntimeWarning)

[GantMan]

You're having around the same speed I have.

I think training is slower in this Keras model for 2 reasons. There is real-time image manipulation to augment the training set, and the amount of data per epoch is huge. Given the batch size, I'm using.
To get this model to 93% I trained on my machine for days.

As for your error message:
Did you change anything in the code? If so consult this link keras-team/keras#836

[GantMan]

Another thing, since your computer you're using has more power than mine, you might want to try removing my workers=0 line. You might get a training boost.

[chenzi]

I tryed set wokrder=1000 then The training speed is faster.