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.

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

Softmax Scheduling Cleanup #3046

Merged
merged 4 commits into from Mar 22, 2023
Merged

Softmax Scheduling Cleanup #3046

merged 4 commits into from Mar 22, 2023

Conversation

cmikeh2
Copy link
Contributor

@cmikeh2 cmikeh2 commented Mar 17, 2023

This PR cleans up softmax scheduling:

  1. Removes edge case whereby for long sequence lengths there would be a divide by 0 on the host (See [BUG] DeepSpeed Inference reports Signal code: Integer divide-by-zero when Seq length is 4096 for GPT2 #2985)
  2. Removes over allocation of threads for medium-sized sequences
  3. Adds clarifying comments for what the behavior should be.

@cmikeh2 cmikeh2 mentioned this pull request Mar 17, 2023
Closed
jeffra
jeffra approved these changes Mar 17, 2023
View reviewed changes
Copy link
Contributor

@jeffra jeffra left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As long as tests pass I am good with this change.

@jeffra jeffra added the merge-queue PRs ready to merge label Mar 17, 2023
@jeffra jeffra merged commit 1286e37 into master Mar 22, 2023
1 check failed
@mrwyattii mrwyattii deleted the cholmes/softmax-launch-fix branch July 7, 2023 02:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jeffra jeffra jeffra approved these changes

@RezaYazdaniAminabadi RezaYazdaniAminabadi Awaiting requested review from RezaYazdaniAminabadi

@awan-10 awan-10 Awaiting requested review from awan-10 awan-10 is a code owner

@arashb arashb Awaiting requested review from arashb arashb is a code owner

Assignees
No one assigned
Labels
merge-queue PRs ready to merge
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
2 participants
@cmikeh2 @jeffra