-
Notifications
You must be signed in to change notification settings - Fork 40
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
Improve dropout performance by unrolling 4xint random masks. #85
Conversation
return x * scale | ||
|
||
|
||
UNROLL = 4 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Since UNROLL=4 is in each jit function, I suggest deleting this.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
lgtm
This PR applies the following changes to dropout.
Improved performance on A100 is show below.
dropout-performance:
M torch gems
0 5.000000e+02 0.199681 0.230627
1 4.096000e+03 1.646302 2.031746
2 1.638400e+04 8.094861 7.757576
3 6.553600e+04 28.151202 31.148290
4 1.000000e+05 40.716611 46.992483
5 1.000000e+09 952.747469 1326.879422