Skip to content

Skip trtllm_alltoall tests on Thor #2448

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
yongwww merged 2 commits into from
Feb 1, 2026
Merged

Skip trtllm_alltoall tests on Thor #2448

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
84858f1
Skip trtllm_alltoall tests on SM110 (Thor) devices
dierksen Jan 30, 2026
615aa90
Clarify that trtllm_alltoall tests hang on Thor
dierksen Jan 30, 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
8 changes: 8 additions & 0 deletions tests/comm/test_trtllm_alltoall.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,14 @@
import torch

import flashinfer.comm.trtllm_alltoall as tllm_alltoall
from flashinfer.utils import get_compute_capability

# Skip all tests on SM110 (Thor) devices - these tests hang indefinitely on this architecture
pytestmark = pytest.mark.skipif(
torch.cuda.is_available()
and get_compute_capability(torch.device("cuda:0"))[0] == 11,
reason="Tests hang indefinitely on SM110 (Thor) devices",
)
Comment on lines +23 to +28
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist bot Jan 30, 2026

Choose a reason for hiding this comment

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

medium

The current implementation correctly skips tests on Thor devices. However, if CUDA is not available, the skipif condition evaluates to false, causing the tests to run and subsequently fail because they require CUDA. It is better practice for tests to be skipped if their environmental requirements are not met.

This suggestion refactors the logic to also skip all tests in this file if CUDA is not available. This provides a clearer result for developers running tests in a non-GPU environment and makes the skipping logic more robust and easier to read.

Suggested change
# Skip all tests on SM110 (Thor) devices - these tests hang indefinitely on this architecture
pytestmark = pytest.mark.skipif(
torch.cuda.is_available()
and get_compute_capability(torch.device("cuda:0"))[0] == 11,
reason="Tests hang indefinitely on SM110 (Thor) devices",
)
# Skip all tests on SM110 (Thor) devices and if CUDA is not available
_skip_reason = None
if not torch.cuda.is_available():
_skip_reason = "CUDA not available, skipping trtllm_alltoall tests"
elif get_compute_capability(torch.device("cuda:0"))[0] == 11:
_skip_reason = "Tests hang indefinitely on SM110 (Thor) devices"
pytestmark = pytest.mark.skipif(_skip_reason is not None, reason=_skip_reason)

Copy link
Copy Markdown
Collaborator

@bkryu bkryu Jan 30, 2026

Choose a reason for hiding this comment

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

CUDA will always be available when we test FlashInfer


has_setup_max_sm_count = False

Expand Down
Loading