diff --git a/tests/e2e/online_serving/test_qwen3_tts.py b/tests/e2e/online_serving/test_qwen3_tts.py index 9a1ff8b97..c430f7930 100644 --- a/tests/e2e/online_serving/test_qwen3_tts.py +++ b/tests/e2e/online_serving/test_qwen3_tts.py @@ -20,7 +20,7 @@ import pytest from tests.conftest import OmniServer from tests.utils import hardware_test -MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice" +MODEL = "/scratch/dyvm6xra/dyvm6xrauser49/rebase/models/hub/models--Qwen--Qwen3-TTS-12Hz-0.6B-CustomVoice/snapshots/85e237c12c027371202489a0ec509ded67b5e4b5" def get_stage_config(): @@ -46,6 +46,7 @@ def omni_server(): "120", "--trust-remote-code", "--enforce-eager", + "--disable-log-stats", ], ) as server: print("OmniServer started successfully") diff --git a/vllm_omni/entrypoints/omni_stage.py b/vllm_omni/entrypoints/omni_stage.py index e32adf737..3116e3294 100644 --- a/vllm_omni/entrypoints/omni_stage.py +++ b/vllm_omni/entrypoints/omni_stage.py @@ -1122,6 +1122,9 @@ async def _stage_worker_async( vllm_config=vllm_config, usage_context=usage_context, engine_args=omni_engine_args, + disable_log_stats=bool( + engine_args.get("disable_log_stats", True) or getattr(omni_engine_args, "disable_log_stats", True) + ), ) omni_stage.set_async_engine(stage_engine) if hasattr(omni_stage.async_engine, "log_stats") and omni_stage.async_engine.log_stats: