Skip to content

[CI] Accuracy issue of qwen3-next-w8a8 nightly test fix. #5746

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
wangxiyuan merged 2 commits into from
Jan 9, 2026
Merged

[CI] Accuracy issue of qwen3-next-w8a8 nightly test fix. #5746

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fbef514
[CI] check for runtime errors in qwen3-next-w8a8 under eager mode.
InSec Jan 8, 2026
2d7e838
[CI] Accuracy issue of qwen3-next-w8a8 nightly test fix
InSec Jan 9, 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
2 changes: 1 addition & 1 deletion tests/e2e/nightly/single_node/models/test_qwen3_next_w8a8.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ async def test_models(model: str) -> None:
"--gpu-memory-utilization",
"0.65",
"--compilation-config",
'{"cudagraph_capture_sizes": [32], "cudagraph_mode":"FULL_DECODE_ONLY"}',
'{"cudagraph_capture_sizes": [32]}',
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot Jan 9, 2026

Choose a reason for hiding this comment

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

high

While disabling FULL_DECODE_ONLY mode is a valid temporary fix for the accuracy issue, the reason for this change is not apparent from the code alone. To improve maintainability and prevent future regressions, it's crucial to add a comment explaining that this is a temporary workaround. This will provide context for other developers who might work on this test in the future and prevent them from re-enabling it without understanding the implications.

Suggested change
'{"cudagraph_capture_sizes": [32]}',
'{"cudagraph_capture_sizes": [32]}', # TODO: Re-enable 'cudagraph_mode: "FULL_DECODE_ONLY"' once accuracy issue is resolved.

]
request_keyword_args: dict[str, Any] = {
**api_keyword_args,
Expand Down
Loading