Skeleton for Attention(23) on CUDA #25684
Draft
+1,296
−237
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
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Comment on lines
+30
to
+31
| T* output_qk // Q*K output | ||
| ) { |
There was a problem hiding this comment.
Suggested change
| T* output_qk // Q*K output | |
| ) { | |
| T* output_qk // Q*K output | |
| ) { |
| template class NaiveAttention<T>; \ | ||
| template void ComputeAttentionProbs<T>(cudaStream_t stream, T * attention_probs, const T* Q, const T* K, const Tensor* mask_index, \ | ||
| const AttentionParameters& parameters, const T* past_key, T* present_key, \ | ||
| T* output_qk); \ |
There was a problem hiding this comment.
Suggested change
| T* output_qk); \ | |
| T* output_qk); \ |
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| @@ -0,0 +1,420 @@ | |||
| /* | |||
Check warning
Code scanning / lintrunner
CLANGFORMAT/format Warning
|
MultiHeadAttention cuda implementation is close to onnx Attention definition. |
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Outdated suggestions cannot be applied.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
Draft for Attention(23) on CUDA.
Two directions for the implementations.
contribops
As it is, it does not seem to support SoftCap options and all the modes for the qk output. What about bfloat16?
cublasLtMatMul
Follows the same implementation the one made on CPU, relies on cublasLtMatMul, it should handle all types (bfloat16, float8). It misses cuda code for the cache copy and softcap.