Skip to content

Conversation

@CuriousPanCake
Copy link
Contributor

Details:

The previous fix 55eda00 for the model didn't account for GPU, which has a different model graph, so needs additional support as well.

Tickets:

Signed-off-by: Andrii Staikov [email protected]

@CuriousPanCake CuriousPanCake requested a review from a team as a code owner January 30, 2026 13:19
@github-actions github-actions bot added the category: transformations OpenVINO Runtime library - Transformations label Jan 30, 2026
@mryzhov mryzhov requested a review from Copilot February 2, 2026 09:41
@mryzhov mryzhov self-assigned this Feb 2, 2026
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR extends RoPE fusion support for the glm-4-9b-chat-hf model to handle GPU graph patterns, which differ from CPU patterns due to the use of Broadcast operations instead of Multiply operations in certain parts of the computation graph.

Changes:

  • Updated pattern matching to accept both Multiply and Broadcast operations in the ChatGLM interleave pattern builder
  • Added comprehensive GPU-specific test case for the glm-4-9b-chat-hf RoPE pattern
  • Renamed existing CPU test to clarify platform-specific testing

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File Description
fuse_rotary_positional_embeddings.cpp (transformations) Updated pattern matching to support Broadcast operation alongside Multiply for GPU compatibility
fuse_rotary_positional_embeddings.cpp (tests) Added GPU-specific test case and renamed CPU test for clarity

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copy link
Contributor

@mryzhov mryzhov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The tests became a little more complicated. Perhaps we should split the model generation into separate blocks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

category: transformations OpenVINO Runtime library - Transformations

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants