From b44b2f981c8029ae8c75010d70180fa915689162 Mon Sep 17 00:00:00 2001 From: psv666 <150513104+psv666@users.noreply.github.com> Date: Wed, 30 Sep 2026 18:07:55 +0800 Subject: [PATCH] [Bugfix] Fix Qwen3-Omni audio-input mean_e2el_ms regression (#8306) Signed-off-by: psv666 <2693925048@qq.com> Co-authored-by: amy-why-3459 --- .../model_executor/models/qwen3_omni/qwen3_omni_moe_thinker.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni_moe_thinker.py b/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni_moe_thinker.py index 5d5edf0df..63875f18d 100644 --- a/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni_moe_thinker.py +++ b/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni_moe_thinker.py @@ -678,6 +678,8 @@ class Qwen3OmniMoeThinkerProcessingInfo(Qwen2AudioProcessingInfo, Qwen2_5_VLProc spatial_merge_size=self.get_hf_config().vision_config.spatial_merge_size, target_sr=feature_extractor.sampling_rate, target_channels=1, + # Retain PyAV for long-audio latency after vLLM 0.30 changed the default. + audio_resample_method="pyav", expected_hidden_size=self._get_expected_hidden_size(), )