Skip to content

Allow Memory Efficient Attention Kernel to run when local window size is set #21310

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
aciddelgado wants to merge 1 commit into from
Closed

Allow Memory Efficient Attention Kernel to run when local window size is set #21310

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion onnxruntime/contrib_ops/cuda/bert/group_query_attention.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,13 +160,15 @@ Status GroupQueryAttention<T>::ComputeInternal(OpKernelContext* context) const {
bool use_memory_efficient_attention =
!use_flash_attention &&
!disable_memory_efficient_attention_ &&
local_window_size_ == -1 &&
(sizeof(T) == 2 || parameters.sequence_length >= attention::kMinSeqLenForMemoryEfficientAttentionFp32) &&
has_memory_efficient_attention(sm, sizeof(T) == 2, parameters.head_size, parameters.head_size);
if (!use_flash_attention && !use_memory_efficient_attention && local_window_size_ != -1) {
return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
"Local attention UNSUPPORTED for sm < 80 on CUDA.");
}
if (use_memory_efficient_attention && local_window_size_ != -1) {
parameters.local_window_size = -1;
}
// allocate buffers
size_t kv_buffer_bytes = 0;
// need a buffer if we must ungroup kv
Expand Down