Skip to content

[Bugfix] Add regression test for allreduce RMS fusion with PP #35960

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
robellliu-dev wants to merge 6 commits into
base: main
Choose a base branch
from
Open

[Bugfix] Add regression test for allreduce RMS fusion with PP #35960

Changes from 3 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5ceaaee
[Bugfix] Add regression test for allreduce RMS fusion with PP
robellliu-dev Mar 4, 2026
1966511
[Bugfix] Format config regression test for pre-commit
robellliu-dev Mar 4, 2026
2b69071
Merge pull request #3 from robellliu-dev/codex/fix-issue-32730-and-cr…
robellliu-dev Mar 4, 2026
a27a45f
Merge branch 'main' into main
robellliu-dev Mar 9, 2026
20c927b
Improve allreduce RMS fusion gating regression coverage
robellliu-dev Mar 9, 2026
ca8b242
Merge branch 'vllm-project:main' into main
robellliu-dev Mar 11, 2026
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
21 changes: 21 additions & 0 deletions tests/test_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@
from vllm.config.vllm import (
OPTIMIZATION_LEVEL_TO_CONFIG,
OptimizationLevel,
enable_allreduce_rms_fusion,
)
from vllm.platforms import current_platform

Expand Down Expand Up @@ -58,6 +59,26 @@ def test_async_scheduling_with_pipeline_parallelism_is_allowed():
assert cfg.scheduler_config.async_scheduling is True


def test_enable_allreduce_rms_fusion_disabled_for_pp():
cfg = VllmConfig(
parallel_config=ParallelConfig(
tensor_parallel_size=2,
pipeline_parallel_size=1,
data_parallel_size=1,
)
)

with (
patch("vllm.utils.flashinfer.has_flashinfer", return_value=True),
patch.object(current_platform, "is_cuda", return_value=True),
patch.object(current_platform, "is_device_capability", return_value=True),
):
assert enable_allreduce_rms_fusion(cfg)

cfg.parallel_config.pipeline_parallel_size = 2
assert not enable_allreduce_rms_fusion(cfg)

@gemini-code-assist gemini-code-assist Bot Mar 4, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

high

The current test correctly covers the pipeline parallelism case as intended. However, to make it more robust and prevent future regressions, it would be beneficial to cover all gating conditions within enable_allreduce_rms_fusion, including tensor_parallel_size and data_parallel_size.

A parameterized test would be a clean way to test all ParallelConfig combinations and improve the test's clarity and maintainability.

@pytest.mark.parametrize(
    ("parallel_config", "should_be_enabled"),
    [
        # Should be enabled with only TP > 1
        (ParallelConfig(tensor_parallel_size=2, pipeline_parallel_size=1, data_parallel_size=1), True),
        # Should be disabled with TP <= 1
        (ParallelConfig(tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1), False),
        # Should be disabled with PP > 1
        (ParallelConfig(tensor_parallel_size=2, pipeline_parallel_size=2, data_parallel_size=1), False),
        # Should be disabled with DP > 1
        (ParallelConfig(tensor_parallel_size=2, pipeline_parallel_size=1, data_parallel_size=2), False),
    ],
    ids=["TP-only", "No-TP", "With-PP", "With-DP"]
)
def test_enable_allreduce_rms_fusion_gating(parallel_config, should_be_enabled):
    cfg = VllmConfig(parallel_config=parallel_config)

    with (
        patch("vllm.utils.flashinfer.has_flashinfer", return_value=True),
        patch.object(current_platform, "is_cuda", return_value=True),
        patch.object(current_platform, "is_device_capability", return_value=True),
    ):
        assert enable_allreduce_rms_fusion(cfg) is should_be_enabled



@dataclass
class _TestConfigFields:
a: int
Expand Down