New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

1bit-SGD + Keras #1975

Closed
minimaxir opened this Issue Jun 6, 2017 · 4 comments

Comments

Projects
None yet
3 participants
@minimaxir

minimaxir commented Jun 6, 2017

How do use the 1bit-SGD optimizer in Keras w/ the CNTK backend? I am not seeing any documentation in that particular area.

Does just setting the normal SGD() optimizer do the trick?

@minimaxir

This comment has been minimized.

minimaxir commented Jun 6, 2017

Actually, just testing it, when setting the optimizer to 'sgd' instead of 'adam' for example in Keras, the optimization doesn't work at all.

@n17s

This comment has been minimized.

Member

n17s commented Jun 7, 2017

It depends what you want to do:

  • If you want to define the model with Keras, then "steal it" and train with CNTK, it should be possible to do it with the existing functionality.
  • if you want to train with Keras APIs, it would be more appropriate if you opened an issue on Keras supporting distributed training.
@minimaxir

This comment has been minimized.

minimaxir commented Jun 7, 2017

The latter.

The maintainer of Keras noted that Microsoft is responsible for the CNTK backend, which is why I commented here first.

@cha-zhang

This comment has been minimized.

Member

cha-zhang commented Jul 13, 2017

I don't think Keras has options to extend and use CNTK native features, and I agree with @n17s that this is not a CNTK issue.

I'm going to close this for now. Feel free to reopen it if you think this is important.

@cha-zhang cha-zhang closed this Jul 13, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment