GQA Rotary and Packed QKV with Flash #18906
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
tianleiwu
reviewed
Jan 2, 2024
tianleiwu
reviewed
Jan 4, 2024
onnxruntime/contrib_ops/cuda/bert/group_query_attention_impl.cu
Outdated
Show resolved
Hide resolved
tianleiwu
reviewed
Jan 4, 2024
yufenglee
reviewed
Jan 13, 2024
tianleiwu
reviewed
Jan 22, 2024
tianleiwu
reviewed
Jan 22, 2024
tianleiwu
previously approved these changes
Jan 22, 2024
tianleiwu
previously approved these changes
Jan 22, 2024
tianleiwu
approved these changes
Jan 23, 2024
YUNQIUGUO
pushed a commit
that referenced
this pull request
Jan 30, 2024
### Description These changes add rotary embedding and packed qkv input to gqa. As of now, the changes are only supported with Flash-Attention (SM >= 80) but should soon be supported with Memory Efficient Attention as well. ### Motivation and Context With the fusion of rotary embedding into this Attention op, we hope to observe some perf gain. The packed QKV should also provide some perf gain in the context of certain models, like Llama2, that would benefit from running ops on the fused QKV matrix, rather than the separate Q, K, and V. --------- Co-authored-by: Yufeng Li <[email protected]>
kunal-vaishnavi
added a commit
that referenced
this pull request
Mar 13, 2024
### Description This PR updates the replacement of MultiHeadAttention (MHA) with GroupQueryAttention (GQA). It is related to the changes in [this PR](#18906). ### Motivation and Context The updated replacement of MHA with GQA includes the following fusion changes. - Apply sliding window within GQA - Fuse the rotary embeddings within GQA - Fuse the 3 MatMuls into 1 packed MatMul if possible - Fuse the 3 Adds into 1 packed Add if possible
YUNQIUGUO
pushed a commit
that referenced
this pull request
Mar 21, 2024
### Description This PR updates the replacement of MultiHeadAttention (MHA) with GroupQueryAttention (GQA). It is related to the changes in [this PR](#18906). ### Motivation and Context The updated replacement of MHA with GQA includes the following fusion changes. - Apply sliding window within GQA - Fuse the rotary embeddings within GQA - Fuse the 3 MatMuls into 1 packed MatMul if possible - Fuse the 3 Adds into 1 packed Add if possible
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.
Description
These changes add rotary embedding and packed qkv input to gqa. As of now, the changes are only supported with Flash-Attention (SM >= 80) but should soon be supported with Memory Efficient Attention as well.
Motivation and Context
With the fusion of rotary embedding into this Attention op, we hope to observe some perf gain. The packed QKV should also provide some perf gain in the context of certain models, like Llama2, that would benefit from running ops on the fused QKV matrix, rather than the separate Q, K, and V.