diff --git a/pyproject.toml b/pyproject.toml
index 97a4ae1..74532e7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -35,7 +35,7 @@ dependencies = [
 
 [project.optional-dependencies]
 vllm = [
-  "vllm==0.14.0",
+  "vllm>=0.15.2rc1",
 ]
 
 [project.urls]
diff --git a/qwen_asr/core/vllm_backend/qwen3_asr.py b/qwen_asr/core/vllm_backend/qwen3_asr.py
index db36219..631dfe0 100644
--- a/qwen_asr/core/vllm_backend/qwen3_asr.py
+++ b/qwen_asr/core/vllm_backend/qwen3_asr.py
@@ -198,7 +198,6 @@ def __init__(
             num_heads=self.num_local_heads,
             head_size=self.head_dim,
             scale=self.scaling,
-            multimodal_config=multimodal_config,
         )
 
     def forward(
@@ -359,15 +358,9 @@ def __init__(
         self.proj2 = nn.Linear(config.d_model, config.output_dim)
 
         # Get attention backend
-        attn_backend_override = (
-            multimodal_config.mm_encoder_attn_backend
-            if multimodal_config is not None
-            else None
-        )
         self.attn_backend = get_vit_attn_backend(
             head_size=config.d_model // config.encoder_attention_heads,
             dtype=torch.get_default_dtype(),
-            attn_backend_override=attn_backend_override,
         )
 
     def compute_attn_mask_seqlen(self, cu_seqlens: torch.Tensor) -> torch.Tensor | None:
@@ -553,6 +546,13 @@ def get_feature_extractor(self, **kwargs: object) -> WhisperFeatureExtractor:
     def get_supported_mm_limits(self) -> Mapping[str, int | None]:
         return {"audio": None}
 
+    def build_data_parser(self) -> MultiModalDataParser:
+        """Build data parser for vllm >= 0.15.2"""
+        feature_extractor = self.get_feature_extractor()
+        return Qwen3ASRMultiModalDataParser(
+            target_sr=feature_extractor.sampling_rate,
+        )
+
 
 class Qwen3ASRDummyInputsBuilder(BaseDummyInputsBuilder[Qwen3ASRProcessingInfo]):
     def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:
@@ -622,12 +622,6 @@ def _parse_audio_data(
 class Qwen3ASRMultiModalProcessor(
     Qwen3OmniMoeThinkerMultiModalProcessor,
 ):
-    def _get_data_parser(self) -> MultiModalDataParser:
-        feature_extractor = self.info.get_feature_extractor()
-        return Qwen3ASRMultiModalDataParser(
-            target_sr=feature_extractor.sampling_rate,
-        )
-
     def _get_mm_fields_config(
         self,
         hf_inputs: BatchFeature,
@@ -994,4 +988,4 @@ def get_generation_prompt(
             "prompt_token_ids": prompt_token_ids,
             "multi_modal_data": {"audio": audio},
         }
-        return cast(PromptType, prompt_dict)
\ No newline at end of file
+        return cast(PromptType, prompt_dict)
diff --git a/qwen_asr/inference/qwen3_asr.py b/qwen_asr/inference/qwen3_asr.py
index d99b915..c6639d3 100644
--- a/qwen_asr/inference/qwen3_asr.py
+++ b/qwen_asr/inference/qwen3_asr.py
@@ -47,10 +47,18 @@
 )
 
 try:
-    from qwen_asr.core.vllm_backend import Qwen3ASRForConditionalGeneration
     from vllm import ModelRegistry
-    ModelRegistry.register_model("Qwen3ASRForConditionalGeneration", Qwen3ASRForConditionalGeneration)
-except:
+
+    try:
+        from vllm.transformers_utils.config import _CONFIG_REGISTRY
+        _CONFIG_REGISTRY["qwen3_asr"] = Qwen3ASRConfig
+    except (ImportError, AttributeError):
+        pass
+
+    model_class_path = "qwen_asr.core.vllm_backend.qwen3_asr:Qwen3ASRForConditionalGeneration"
+    ModelRegistry.register_model("qwen3_asr", model_class_path)
+    ModelRegistry.register_model("Qwen3ASRForConditionalGeneration", model_class_path)
+except Exception as e:
     pass
 
 
@@ -827,4 +835,4 @@ def finish_streaming_transcribe(self, state: ASRStreamingState) -> ASRStreamingS
         state.text = txt
 
         state.chunk_id += 1
-        return state
\ No newline at end of file
+        return state