Unify strides of o and do in attention backward#1
Merged
Conversation
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
When performing backward propagation,
oanddowill sometimes have different strides and fail the stride check:https://github.com/linxihui/dkernel/blob/main/dkernel/ops/sparse_attn_bwd.py#L675
This is true when training a model with GQA, where the key and value need to be repeated before passing to the kernel. When passing in
q,kandvwith size[B, S, H, D]and stride[S*H*D, H*D, D, 1]whereB == 1,the outputohas the same size and stride, butdohas stride[H*D, H*D, D, 1]. I have not tested training a model without GQA.According to this attention implementation of Megatron-LM, stride for a dimension that is 1 has no meaning, so the two strides mean the same thing. We adapt their solution to pass the stride check here.