Skip to content

[CI][Qwen3-Omni] Keep the deploy-YAML literal in the structured-config sampling check #7803

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
NickCao merged 1 commit into from
Sep 18, 2026
Merged

[CI][Qwen3-Omni] Keep the deploy-YAML literal in the structured-config sampling check #7803

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 6 additions & 4 deletions tests/e2e/offline_inference/test_qwen3_omni.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,16 +112,18 @@ def test_structured_multistage_config_reaches_runtime(omni_runner, offline_clien
assert code2wav_vllm.scheduler_config.async_scheduling is False

# Sampling defaults are consumed from StageClient metadata, rather than
# from either the resolver output or VllmConfig.
# from either the resolver output or VllmConfig. These literals mirror
# vllm_omni/deploy/qwen3_omni_moe.yaml exactly.
expected_sampling = (
{"temperature": 0.0, "max_tokens": 2048},
{"temperature": 0.9, "top_k": 50, "max_tokens": 4096, "repetition_penalty": 1.05},
{
"temperature": 0.0,
"top_p": 1.0,
# SamplingParams normalizes the disabled greedy-sampling sentinel
# from the deploy YAML (top_k=-1) to its runtime value (top_k=0).
"top_k": 0,
# Deploy YAML literal. The resolver check below compares it verbatim;
# the runtime check normalizes it through SamplingParams, which turns
# the greedy-sampling sentinel -1 into 0.
"top_k": -1,
"max_tokens": 65536,
"repetition_penalty": 1.1,
},
Expand Down
Loading