Pythorch Mish() improvement #293

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.

Sign up for GitHub

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

Jump to bottom

Open

AhmedMesih wants to merge 1 commit into WongKinYiu:yolov4-csp from AhmedMesih:yolov4-csp

AhmedMesih commented Jul 8, 2021

when I tested Tesla P100, using return F.mish(x) insted of return x * F.softplus(x).tanh() significantly reduced GPU memory usage.


          update

089d3f2

Owner

WongKinYiu commented Jul 9, 2021

thanks, mish-cuda even use less memory than nn.SiLU.
i think i have to wait ngc pytorch version 21.07 for testing if mish is supported or not.

AlexeyAB commented Jul 9, 2021

Yes, Mish was proposed 3 Sep 2019 pytorch/pytorch#25584 (comment) and 25 May 2021 it was added to Pytorch 1.9.0 pytorch/pytorch#58940
Also Mish is in the TensorFlow: https://www.tensorflow.org/addons/api_docs/python/tfa/activations/mish

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

Labels

None yet