Skip to content

[ONNX FE] don't skip total_sequence_length for GroupQueryAttention Node #32981

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

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
+6 −7
Open

[ONNX FE] don't skip total_sequence_length for GroupQueryAttention Node #32981

Show file tree
Hide file tree
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
5 changes: 3 additions & 2 deletions ...ransformations/src/transformations/op_conversions/group_query_attention_decomposition.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,8 +70,9 @@ ov::OutputVector ov::pass::GroupQueryAttentionDecomposition::decompose(
auto past_key = node->input_value(3);
auto past_value = node->input_value(4);
auto seqlens_k = node->input_value(5);
auto cos_cache = node->input_value(6);
auto sin_cache = node->input_value(7);
auto total_sequence_length = node->input_value(6);
auto cos_cache = node->input_value(7);
auto sin_cache = node->input_value(8);

// The length of all tokens (past + current) is `seqlens_k` + 1.
// current = Q.shape[2], past = `seqlens_k` + 1 - current
Expand Down
3 changes: 0 additions & 3 deletions src/frontends/onnx/frontend/src/op/com.microsoft/group_query_attention.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,9 +90,6 @@ ov::OutputVector group_query_attention(const ov::frontend::onnx::Node& node) {
}

for (int i = 3; i < 9; ++i) {
// skip total_sequence_length
if (i == 6)
continue;
ov_op_inputs.push_back(onnx_op_inputs[i]);
}
return std::make_shared<internal::GroupQueryAttention>(ov_op_inputs,
Expand Down
5 changes: 3 additions & 2 deletions src/plugins/intel_npu/src/plugin/npuw/llm_compiled_model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -115,8 +115,9 @@ class GroupQueryAttentionDecomposition : public ov::pass::MatcherPass {
auto past_key = node->input_value(3);
auto past_value = node->input_value(4);
auto seqlens_k = node->input_value(5);
auto cos_cache = node->input_value(6);
auto sin_cache = node->input_value(7);
auto total_sequence_length = node->input_value(6);
auto cos_cache = node->input_value(7);
auto sin_cache = node->input_value(8);

// The length of all tokens (past + current) is `seqlens_k` + 1.
// current = Q.shape[2], past = `seqlens_k` + 1 - current
Expand Down
Loading