-
Notifications
You must be signed in to change notification settings - Fork 270
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
[minor] OSS: bring DDP in the benchmark #130
Conversation
(the values in the benchmark regression test will need to be updated, that's expected. The current commit will fail because of that) |
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!
benchmarks/oss.py
Outdated
@@ -91,6 +94,7 @@ def train( | |||
optimizer = ddp.optimizer | |||
model = ddp | |||
else: | |||
model = DDP(model, device_ids=[rank]) |
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.
should we toggle find_unused_parameters to True/False and see its impacts?
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.
ahh, good point ! it was not yet up for review actually, looks like there's something wrong with SDP, looking into that :) I'll test this option
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.
oops, sorry, didn't notice that! :-)
looks like something is wrong in SDP depending on the world size, and this benchmark change exposed that |
…is correct wrt DDP, but SDP needs to be fixed
Before submitting
What does this PR do?
A suggestion from @msbaines on another PR made me think of that, the benchmark job that we had was not using DDP so not super realistic. The fake pictures were also the same on each rank, which prevents catching some sync issues, fixing that
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
Did you have fun?
Make sure you had fun coding 🙃