diff --git a/vllm/model_executor/models/qwen3_vl_moe.py b/vllm/model_executor/models/qwen3_vl_moe.py index 17164d1735..4413e1213b 100644 --- a/vllm/model_executor/models/qwen3_vl_moe.py +++ b/vllm/model_executor/models/qwen3_vl_moe.py @@ -216,6 +216,7 @@ class Qwen3VLMoeForConditionalGeneration( multimodal_config = vllm_config.model_config.multimodal_config self.config = config + self.model_config = vllm_config.model_config self._tokenizer = cached_tokenizer_from_config(vllm_config.model_config) self.multimodal_config = multimodal_config self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"