Skip to content

Update attention fusion in speech component of Phi-4 mm #24513

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

Merged
merged 5 commits into from
Apr 26, 2025
Merged

Update attention fusion in speech component of Phi-4 mm #24513

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6fa631d
Update K path in conformer attention fusion
kunal-vaishnavi Apr 7, 2025
8023922
Keep original path when matching
kunal-vaishnavi Apr 23, 2025
c0e6edd
Merge branch 'main' into kvaishnavi/phi4-mm-speech
kunal-vaishnavi Apr 25, 2025
89aca0f
Merge branch 'main' into kvaishnavi/phi4-mm-speech
kunal-vaishnavi Apr 25, 2025
140ad53
Merge branch 'main' into kvaishnavi/phi4-mm-speech
kunal-vaishnavi Apr 26, 2025
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
11 changes: 8 additions & 3 deletions onnxruntime/python/tools/transformers/fusion_conformer_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,8 +126,14 @@ def fuse(self, normalize_node, input_name_to_nodes, output_name_to_node):
[1, 0, 0, 0, 0],
)
if k_nodes is None:
logger.debug("fuse_conformer_attention: failed to match k path")
return
k_nodes = self.model.match_parent_path(
matmul_qk,
["Transpose", "Reshape", "Add", "MatMul"],
[1, 0, 0, 0],
)
if k_nodes is None:
logger.debug("fuse_conformer_attention: failed to match k path")
return
else:
concat_k = k_nodes[1]
concat_parent = self.model.get_parent(concat_k, 0, None)
Expand Down Expand Up @@ -188,7 +194,6 @@ def fuse(self, normalize_node, input_name_to_nodes, output_name_to_node):
logger.debug("fuse_conformer_attention: MultiHeadAttention node creation failed")
return

self.increase_counter(new_node.op_type)
self.nodes_to_add.append(new_node)
self.node_name_to_graph_name[new_node.name] = self.this_graph_name

Expand Down
Loading