Update augment branch to use kernels for zero-fill before layernorm and rmsnorm#1
Merged
cbcase merged 5 commits intov0.13-augmentfrom Oct 17, 2024
Merged
Conversation
MarkusRabe
approved these changes
Oct 17, 2024
MarkusRabe
reviewed
Oct 17, 2024
MarkusRabe
left a comment
MarkusRabe
left a comment
There was a problem hiding this comment.
My only comment might be that it might be helpful to keep the existing code runnable, just hidden by a flag.
MarkusRabe
reviewed
Oct 17, 2024
| params.barrier = reinterpret_cast<int*>(barrier->data.dptr); | ||
| } | ||
|
|
||
| const char *envval = std::getenv("NVTE_FORCE_MEMSET"); |
There was a problem hiding this comment.
The only nit is that it would be good to document the flag. This is a fall back to the original behavior of library.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR updates the TE code to use a fill kernel (not memset) to zero-out buffers before calling the fast fp8 layernorm and rmsnorm kernels. We are making this change because the memset ops introduce gaps in cuda graph execution.