model: qwen3-omni (thinker-only) (#10911)

Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>

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

@@ -155,7 +155,6 @@ class BaseMultimodalProcessor(ABC):
     ):
         self.hf_config = hf_config
         self._processor = _processor
-        self.arch = hf_config.architectures[0]
         self.server_args = server_args
         self.transport_mode = transport_mode
 
@@ -191,6 +190,7 @@ class BaseMultimodalProcessor(ABC):
             "input_features": Modality.AUDIO,
             "input_features_mask": Modality.AUDIO,
             "audio_attention_mask": Modality.AUDIO,
+            "feature_attention_mask": Modality.AUDIO,
             # Video-related attributes
             "pixel_values_videos": Modality.VIDEO,
             "second_per_grid_ts": Modality.VIDEO,
@@ -222,6 +222,7 @@ class BaseMultimodalProcessor(ABC):
             if self._processor.__class__.__name__ in {
                 "Gemma3nProcessor",
                 "Qwen2AudioProcessor",
+                "Qwen3OmniMoeProcessor",
             }:
                 # Note(Xinyuan): for gemma3n, ref: https://github.com/huggingface/transformers/blob/ccf2ca162e33f381e454cdb74bf4b41a51ab976d/src/transformers/models/gemma3n/processing_gemma3n.py#L107
                 kwargs["audio"] = audios