Make DMMHA kernel inside MHA optional for Whisper - #25166
Merged
kunal-vaishnavi merged 1 commit intoJun 25, 2025
Merged
Conversation
baijumeswani
approved these changes
Jun 25, 2025
baijumeswani
left a comment
baijumeswani
left a comment
Contributor
There was a problem hiding this comment.
Would this impact perf on whisper cuda models?
Contributor
Author
Yes, performance will be temporarily reduced for now. Once the DMMHA models are working with ONNX Runtime GenAI for word-level timestamps, the flag can be set to true by default. |
ankus-qti
pushed a commit
to CodeLinaro/onnxruntime
that referenced
this pull request
Nov 25, 2025
### Description This PR sets adding support for the `DecoderMaskedMultiHeadAttention` (DMMHA) kernel inside `MultiHeadAttention` (MHA) to false by default. ### Motivation and Context The models containing the extra inputs for DMMHA (i.e. `past_sequence_length` and `cache_indirection`) have some runtime issues. Additionally, not all execution providers implement the DMMHA kernel inside MHA and will therefore not support these extra inputs.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
This PR sets adding support for the
DecoderMaskedMultiHeadAttention(DMMHA) kernel insideMultiHeadAttention(MHA) to false by default.Motivation and Context
The models containing the extra inputs for DMMHA (i.e.
past_sequence_lengthandcache_indirection) have some runtime issues. Additionally, not all execution providers implement the DMMHA kernel inside MHA and will therefore not support these extra inputs.