diff --git a/src/mcore_bridge/config/parser.py b/src/mcore_bridge/config/parser.py
index b5351a5..7cf59c4 100644
--- a/src/mcore_bridge/config/parser.py
+++ b/src/mcore_bridge/config/parser.py
@@ -152,7 +152,8 @@ def hf_to_mcore_config(hf_config: PretrainedConfig) -> Dict[str, Any]:
     moe_n_hash_layers = res.pop('moe_n_hash_layers', None)
     rope_scaling = res.get('rope_scaling') or {}
     if llm_model_type in {'qwen3', 'qwen3_moe', 'qwen3_next'} or hf_model_type in {
-            'qwen3_omni_moe', 'qwen3_omni', 'qwen3_vl', 'qwen3_vl_moe', 'qwen3_5', 'qwen3_5_moe', 'llavaonevision1_5', 'minicpmv4_6'
+            'qwen3_omni_moe', 'qwen3_omni', 'qwen3_vl', 'qwen3_vl_moe', 'qwen3_5', 'qwen3_5_moe', 'llavaonevision1_5',
+            'minicpmv4_6'
     }:
         res['qk_layernorm'] = True
     if llm_model_type in {'qwen2_moe', 'qwen3_moe', 'qwen3_next'
diff --git a/src/mcore_bridge/model/mm_gpts/__init__.py b/src/mcore_bridge/model/mm_gpts/__init__.py
index a8950fa..896f5e1 100644
--- a/src/mcore_bridge/model/mm_gpts/__init__.py
+++ b/src/mcore_bridge/model/mm_gpts/__init__.py
@@ -1,2 +1,3 @@
 # Copyright (c) ModelScope Contributors. All rights reserved.
-from . import gemma4, glm, internvl, kimi_vl, llama4, llava, minicpmv4_6, qwen, qwen3_5, qwen3_5_gdn, qwen3_asr, qwen3_omni, qwen3_vl
+from . import (gemma4, glm, internvl, kimi_vl, llama4, llava, minicpmv4_6, qwen, qwen3_5, qwen3_5_gdn, qwen3_asr,
+               qwen3_omni, qwen3_vl)
diff --git a/src/mcore_bridge/model/mm_gpts/minicpmv4_6.py b/src/mcore_bridge/model/mm_gpts/minicpmv4_6.py
index 16b5b73..a11c094 100644
--- a/src/mcore_bridge/model/mm_gpts/minicpmv4_6.py
+++ b/src/mcore_bridge/model/mm_gpts/minicpmv4_6.py
@@ -17,11 +17,8 @@ class MiniCPMV46Vit(HuggingFaceVit):
     _aligner = ['merger']
 
     def prepare_model(self, hf_config: PretrainedConfig):
-        from transformers.models.minicpmv4_6.modeling_minicpmv4_6 import (
-            MiniCPMV4_6VisionModel,
-            MiniCPMV4_6Merger,
-            MiniCPMV4_6Model
-        )
+        from transformers.models.minicpmv4_6.modeling_minicpmv4_6 import (MiniCPMV4_6Merger, MiniCPMV4_6Model,
+                                                                          MiniCPMV4_6VisionModel)
         self.vision_tower = MiniCPMV4_6VisionModel._from_config(hf_config.vision_config)
         self.merger = MiniCPMV4_6Merger(hf_config).to(dtype=self.vision_tower.dtype)
         self.model_cls = MiniCPMV4_6Model
@@ -37,10 +34,8 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
         if pixel_values is None and pixel_values_videos is None:
             patch_size = hf_config.vision_config.patch_size
             dummy_pv = torch.zeros(
-                1, 3, 4 * patch_size, 4 * patch_size,
-                device=inputs_embeds.device, dtype=self.vision_tower.dtype)
-            dummy_ts = torch.tensor(
-                [[4, 4]], device=inputs_embeds.device, dtype=torch.int32)
+                1, 3, 4 * patch_size, 4 * patch_size, device=inputs_embeds.device, dtype=self.vision_tower.dtype)
+            dummy_ts = torch.tensor([[4, 4]], device=inputs_embeds.device, dtype=torch.int32)
             with self.patch_hf_config():
                 vision_output = self.model_cls.get_image_features(self, dummy_pv, dummy_ts)
             image_embeds = torch.cat(vision_output.pooler_output, dim=0)
@@ -52,11 +47,11 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
                     vision_output = self.model_cls.get_image_features(
                         self, pixel_values[:1].to(dtype=self.vision_tower.dtype), target_sizes)
                     image_features = (
-                        torch.cat(vision_output.pooler_output, dim=0)
-                        .to(device=inputs_embeds.device, dtype=inputs_embeds.dtype)
-                        .repeat(num_beams, 1))
-                    mask = self.model_cls.get_placeholder_mask(
-                        self, input_ids, inputs_embeds, image_features, hf_config.image_token_id)
+                        torch.cat(vision_output.pooler_output,
+                                  dim=0).to(device=inputs_embeds.device,
+                                            dtype=inputs_embeds.dtype).repeat(num_beams, 1))
+                    mask = self.model_cls.get_placeholder_mask(self, input_ids, inputs_embeds, image_features,
+                                                               hf_config.image_token_id)
                 inputs_embeds = inputs_embeds.masked_scatter(mask, image_features)
 
             if pixel_values_videos is not None:
@@ -65,11 +60,11 @@ def get_inputs_embeds(self, inputs_embeds, **kwargs):
                     vision_output = self.model_cls.get_video_features(
                         self, pixel_values_videos[:1].to(dtype=self.vision_tower.dtype), target_sizes_videos)
                     video_features = (
-                        torch.cat(vision_output.pooler_output, dim=0)
-                        .to(device=inputs_embeds.device, dtype=inputs_embeds.dtype)
-                        .repeat(num_beams, 1))
-                    mask = self.model_cls.get_placeholder_mask(
-                        self, input_ids, inputs_embeds, video_features, hf_config.video_token_id)
+                        torch.cat(vision_output.pooler_output,
+                                  dim=0).to(device=inputs_embeds.device,
+                                            dtype=inputs_embeds.dtype).repeat(num_beams, 1))
+                    mask = self.model_cls.get_placeholder_mask(self, input_ids, inputs_embeds, video_features,
+                                                               hf_config.video_token_id)
                 inputs_embeds = inputs_embeds.masked_scatter(mask, video_features)
 
         return inputs_embeds