Skip to content

use combo kernel to fuse qk-norm and qk-rope #26682

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 2 commits into from
Oct 14, 2025
Merged

use combo kernel to fuse qk-norm and qk-rope #26682

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
506a52f
use combo kernel
BoyuanFeng Oct 13, 2025
00f4fca
respect user inductor config
BoyuanFeng Oct 13, 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
10 changes: 10 additions & 0 deletions vllm/config/compilation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -498,6 +498,16 @@ def __post_init__(self) -> None:
if isinstance(self.pass_config, dict):
self.pass_config = PassConfig(**self.pass_config)

if (
is_torch_equal_or_newer("2.9.0.dev")
and "combo_kernels" not in self.inductor_compile_config
and "benchmark_combo_kernel" not in self.inductor_compile_config
):
# use horizontal fusion, which is useful for fusing qk-norm and
# qk-rope when query and key have different shapes.
self.inductor_compile_config["combo_kernels"] = True
self.inductor_compile_config["benchmark_combo_kernel"] = True
Comment on lines +506 to +509
Copy link
Member

@mgoin mgoin Oct 14, 2025

Choose a reason for hiding this comment

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

I would appreciate a doc pointer to how this works so I can understand for future work. Currently this is very opaque

Copy link
Contributor Author

@BoyuanFeng BoyuanFeng Oct 14, 2025

Choose a reason for hiding this comment

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

we have some doc here. I will followup with more pytorch docs.


# migrate the deprecated flags
if not self.use_cudagraph:
logger.warning(
Expand Down