resolve

Author: mickqian

python/sglang/srt/models/qwen3_omni_moe.py

@@ -423,9 +423,6 @@ def __init__(
         quant_config: Optional[QuantizationConfig] = None,
         prefix: str = "",
     ):
-        # for k, v in config.text_config.__dict__.items():
-        #     setattr(config, k, v)
-        # config.vocab_size = config.text_config.vocab_size
         super().__init__(
             config, quant_config, prefix, language_model_cls=Qwen3MoeLLMModel
         )

python/sglang/srt/multimodal/processors/qwen_vl.py

@@ -312,7 +312,7 @@ async def process_mm_data_async(
             use_audio_in_video=False,
             audio_seqlens=audio_feature_lengths,
             audio_token_id=getattr(self.hf_config, "audio_token_id", None),
-            audio_start_token_id=getattr(self.hf_config, "audio_start_token_id", None),
+            audio_start_token_id=self.audio_start_token_id,
             position_id_per_seconds=getattr(
                 self.hf_config, "position_id_per_seconds", None
             ),