Skip to content
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

Align optimizer swap buffer sizes #1036

Merged
merged 6 commits into from May 5, 2021
Merged

Align optimizer swap buffer sizes #1036

merged 6 commits into from May 5, 2021

Conversation

tjruwase
Copy link
Contributor

@tjruwase tjruwase commented May 3, 2021

Optimizer swap buffer sizes should be aligned, fix for #1033.
NVMe intra-request validation should be on entire file.

Optimizer swap buffer sizes should be aligned
@jeffra jeffra linked an issue May 3, 2021 that may be closed by this pull request
@tjruwase
Copy link
Contributor Author

tjruwase commented May 3, 2021

Added fix for #1037.

@jeffra jeffra linked an issue May 5, 2021 that may be closed by this pull request
@jeffra jeffra merged commit be5cd76 into master May 5, 2021
@stas00
Copy link
Contributor

stas00 commented May 5, 2021

This doesn't quite fix #1037 though. The problem is that the warning is not matching the traceback in its appearance in the log - sometimes it appears after the crash - these are 2 different streams. So it's very difficult to see and make a correlation with the JIT crash.

This needs to be an assert, IMHO - and not a warning.

For some reason I'm not allowed to re-open the Issue I created - odd.

@tjruwase
Copy link
Contributor Author

tjruwase commented May 5, 2021

@stas00, good point. I have re-opened the issue. I think we require a more general solution because assert is heavy-weight in some cases. But I agree we need to make the failure messages more meaningful.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

deepspeed.ops.op_builder.async_io.AsyncIOBuilder assert AssertionError: num_elems 513024> buffer 513000.
4 participants