diff --git a/vllm/transformers_utils/configs/deepseek_vl2.py b/vllm/transformers_utils/configs/deepseek_vl2.py
index 03f24319e2..3d3e20fea8 100644
--- a/vllm/transformers_utils/configs/deepseek_vl2.py
+++ b/vllm/transformers_utils/configs/deepseek_vl2.py
@@ -101,7 +101,6 @@ else:
 
 class DeepseekVLV2Config(PretrainedConfig):
     model_type = "deepseek_vl_v2"
-    architectures: list[str] | None = None
 
     tile_tag: str = "2D"
     global_view_pos: str = "head"
@@ -114,17 +113,11 @@ class DeepseekVLV2Config(PretrainedConfig):
         candidate_resolutions: tuple[tuple[int, int]] = ((384, 384),),
         **kwargs,
     ):
-        if "architectures" not in kwargs:
-            kwargs["architectures"] = ["DeepseekVLV2ForCausalLM"]
+        architectures = kwargs.setdefault("architectures", ["DeepseekVLV2ForCausalLM"])
 
-        vision_config = kwargs.pop("vision_config", {})
-        self.vision_config = VisionEncoderConfig(**vision_config)
-
-        projector_config = kwargs.pop("projector_config", {})
-        self.projector_config = MlpProjectorConfig(**projector_config)
-
-        language_config = kwargs.pop("language_config", {})
-        self.text_config = DeepseekVLV2TextConfig(**language_config)
+        self.vision_config = VisionEncoderConfig(**kwargs.pop("vision_config", {}))
+        self.projector_config = MlpProjectorConfig(**kwargs.pop("projector_config", {}))
+        self.text_config = DeepseekVLV2TextConfig(**kwargs.pop("language_config", {}))
 
         self.tile_tag = tile_tag
         self.global_view_pos = global_view_pos
@@ -132,8 +125,8 @@ class DeepseekVLV2Config(PretrainedConfig):
         self.vocab_size = self.text_config.vocab_size
 
         # update model_type for OCR models
-        if "DeepseekOCRForCausalLM" in kwargs["architectures"]:
-            self.model_type = "deepseek_ocr"
-        elif "DeepseekOCR2ForCausalLM" in kwargs["architectures"]:
-            self.model_type = "deepseek_ocr2"
+        if "DeepseekOCRForCausalLM" in architectures:
+            kwargs["model_type"] = "deepseek_ocr"
+        elif "DeepseekOCR2ForCausalLM" in architectures:
+            kwargs["model_type"] = "deepseek_ocr2"
         super().__init__(**kwargs)