Use FusedSDPA for MllamaVisionSdpaAttention #620
Merged
Conversation
This file contains 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
kdamaszk
requested review from
kzawora-intel,
madamczykhabana,
michalkuligowski and
mgawarkiewicz
as code owners
michalkuligowski
approved these changes
Jan 8, 2025
michalkuligowski
pushed a commit
that referenced
this pull request
Jan 8, 2025
This is a updated version from #650. Coupled with [Use FusedSDPA for MllamaVisionSdpaAttention #620], these two issues arising when running llama3.2 vision model can be resolved: GC fail when batchsize>1 on Gaudi3. Increased device memory consumption with Torch 2.5 compared to Torch 2.4. --------- Signed-off-by: yan ma <[email protected]> Co-authored-by: yisonzhu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Use
FusedSDPAinstead of regularF.scaled_dot_product_attentioninMllamaVisionSdpaAttentionmodule.The difference between these two ops is precision, since
F.scaled_dot_product_attentionconverts the input to float32 and performs all operations on this data type, whileFusedSDPAdoes not. However, it change accuracy only from 0.449 to 0.446 on accuracy test based on MMMU dataset and lm-evaluation-harness, while improves single prompt performance from ~550ms to ~100ms.