diff --git a/vllm/model_executor/models/baichuan.py b/vllm/model_executor/models/baichuan.py
index abaefa3cf7781..5cf5a199b7690 100644
--- a/vllm/model_executor/models/baichuan.py
+++ b/vllm/model_executor/models/baichuan.py
@@ -295,8 +295,6 @@ def forward(
 
 
 class BaiChuanBaseForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "W_pack": ["W_pack"],
         "gate_up_proj": [
diff --git a/vllm/model_executor/models/chatglm.py b/vllm/model_executor/models/chatglm.py
index bf64538ef54a3..5b5a69447e0b8 100644
--- a/vllm/model_executor/models/chatglm.py
+++ b/vllm/model_executor/models/chatglm.py
@@ -325,8 +325,6 @@ def forward(
 
 
 class ChatGLMForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "query_key_value": ["query_key_value"],
         "dense_h_to_4h": ["dense_h_to_4h"]
diff --git a/vllm/model_executor/models/gemma.py b/vllm/model_executor/models/gemma.py
index 9e071a155061b..ce97fc808c85e 100644
--- a/vllm/model_executor/models/gemma.py
+++ b/vllm/model_executor/models/gemma.py
@@ -291,8 +291,6 @@ def forward(
 
 
 class GemmaForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
diff --git a/vllm/model_executor/models/gpt_bigcode.py b/vllm/model_executor/models/gpt_bigcode.py
index 009d7b1498c22..17bbe4e312fc3 100644
--- a/vllm/model_executor/models/gpt_bigcode.py
+++ b/vllm/model_executor/models/gpt_bigcode.py
@@ -233,8 +233,6 @@ def forward(
 
 
 class GPTBigCodeForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {"c_attn": ["c_attn"]}
 
     supported_lora_modules = ["c_fc", "c_proj", "wte", "lm_head", "c_attn"]
diff --git a/vllm/model_executor/models/interfaces.py b/vllm/model_executor/models/interfaces.py
index a9eb397a5a97f..cb0fc154a74d8 100644
--- a/vllm/model_executor/models/interfaces.py
+++ b/vllm/model_executor/models/interfaces.py
@@ -13,7 +13,14 @@
 class SupportsVision(Protocol):
     """The interface required for all vision language models (VLMs)."""
 
-    supports_vision: ClassVar[Literal[True]]
+    supports_vision: ClassVar[Literal[True]] = True
+    """
+    A flag that indicates this model supports vision inputs.
+
+    Note:
+        There is no need to redefine this flag if this class is in the
+        MRO of your model class.
+    """
 
     def __init__(self, *, vlm_config: VisionLanguageConfig) -> None:
         ...
@@ -52,7 +59,14 @@ def supports_vision(
 class SupportsLoRA(Protocol):
     """The interface required for all models that support LoRA."""
 
-    supports_lora: ClassVar[Literal[True]]
+    supports_lora: ClassVar[Literal[True]] = True
+    """
+    A flag that indicates this model supports LoRA.
+
+    Note:
+        There is no need to redefine this flag if this class is in the
+        MRO of your model class.
+    """
 
     packed_modules_mapping: ClassVar[Dict[str, List[str]]]
     supported_lora_modules: ClassVar[List[str]]
diff --git a/vllm/model_executor/models/llama.py b/vllm/model_executor/models/llama.py
index f4918cbfef294..54d01701f04fb 100644
--- a/vllm/model_executor/models/llama.py
+++ b/vllm/model_executor/models/llama.py
@@ -299,8 +299,6 @@ def forward(
 
 
 class LlamaForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
diff --git a/vllm/model_executor/models/llava.py b/vllm/model_executor/models/llava.py
index 8e18b42b76734..125e3ddea7df3 100644
--- a/vllm/model_executor/models/llava.py
+++ b/vllm/model_executor/models/llava.py
@@ -88,8 +88,6 @@ class LlavaImageFeatureInputs(TypedDict):
 @MULTIMODAL_REGISTRY.register_dummy_data(get_dummy_image_data)
 class LlavaForConditionalGeneration(nn.Module, SupportsVision):
 
-    supports_vision = True
-
     def __init__(self,
                  config: LlavaConfig,
                  vlm_config: VisionLanguageConfig,
diff --git a/vllm/model_executor/models/llava_next.py b/vllm/model_executor/models/llava_next.py
index 5c03fb3705561..841818d8db6ff 100644
--- a/vllm/model_executor/models/llava_next.py
+++ b/vllm/model_executor/models/llava_next.py
@@ -108,8 +108,6 @@ def _image_pixel_processor(
 @MULTIMODAL_REGISTRY.register_dummy_data(_get_dummy_image_data)
 class LlavaNextForConditionalGeneration(nn.Module, SupportsVision):
 
-    supports_vision = True
-
     def __init__(self,
                  config: LlavaNextConfig,
                  vlm_config: VisionLanguageConfig,
diff --git a/vllm/model_executor/models/minicpm.py b/vllm/model_executor/models/minicpm.py
index ae17309bd5223..a76ed049828e7 100644
--- a/vllm/model_executor/models/minicpm.py
+++ b/vllm/model_executor/models/minicpm.py
@@ -392,8 +392,6 @@ def forward(
 
 
 class MiniCPMForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
diff --git a/vllm/model_executor/models/mixtral.py b/vllm/model_executor/models/mixtral.py
index 0bdcb21e514fd..a662db6d28d00 100644
--- a/vllm/model_executor/models/mixtral.py
+++ b/vllm/model_executor/models/mixtral.py
@@ -475,8 +475,6 @@ def forward(
 
 
 class MixtralForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     fall_back_to_pt_during_load = False
 
     packed_modules_mapping = {
diff --git a/vllm/model_executor/models/phi.py b/vllm/model_executor/models/phi.py
index d288bdd9d78f5..008fceb624f75 100644
--- a/vllm/model_executor/models/phi.py
+++ b/vllm/model_executor/models/phi.py
@@ -232,8 +232,6 @@ def forward(
 
 
 class PhiForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
diff --git a/vllm/model_executor/models/phi3v.py b/vllm/model_executor/models/phi3v.py
index 578e22beaa3d6..0bbe93241b139 100644
--- a/vllm/model_executor/models/phi3v.py
+++ b/vllm/model_executor/models/phi3v.py
@@ -32,12 +32,13 @@
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.models.clip import CLIPVisionModel
 from vllm.model_executor.models.llama import LlamaModel
-from vllm.model_executor.models.vlm_base import VisionLanguageModelBase
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.image import ImagePixelData, get_dummy_image_data
 from vllm.sequence import SamplerOutput
 
+from .interfaces import SupportsVision
+
 logger = init_logger(__name__)
 
 _KEYS_TO_MODIFY_MAPPING = {
@@ -317,18 +318,21 @@ def _image_processor(
 
 @MULTIMODAL_REGISTRY.register_image_pixel_input(_image_processor)
 @MULTIMODAL_REGISTRY.register_dummy_data(get_dummy_image_data)
-class Phi3VForCausalLM(VisionLanguageModelBase):
+class Phi3VForCausalLM(nn.Module, SupportsVision):
 
     def __init__(self,
                  config: PretrainedConfig,
-                 vision_language_config: VisionLanguageConfig,
+                 vlm_config: VisionLanguageConfig,
                  cache_config: Optional[CacheConfig] = None,
                  quant_config: Optional[QuantizationConfig] = None) -> None:
-        super().__init__(vision_language_config)
+        super().__init__()
+
         self.config = config
+        self.vlm_config = vlm_config
+
         self.model = LlamaModel(config, cache_config, quant_config)
         self.vision_embed_tokens = Phi3HDImageEmbedding(
-            vision_language_config, config, self.model.embed_tokens)
+            vlm_config, config, self.model.embed_tokens)
         self.lm_head = ParallelLMHead(config.vocab_size, config.hidden_size)
         self.logits_processor = LogitsProcessor(config.vocab_size)
         self.sampler = Sampler()
@@ -338,7 +342,7 @@ def _parse_and_validate_image_input(
         pixel_values = kwargs.pop("pixel_values", None)
         image_sizes = kwargs.pop("image_sizes", None)
 
-        expected_input_type = self.vision_language_config.image_input_type
+        expected_input_type = self.vlm_config.image_input_type
         ImageInputType = VisionLanguageConfig.ImageInputType
 
         if expected_input_type != ImageInputType.PIXEL_VALUES:
diff --git a/vllm/model_executor/models/qwen2.py b/vllm/model_executor/models/qwen2.py
index d351adcefc974..e2d725af63593 100644
--- a/vllm/model_executor/models/qwen2.py
+++ b/vllm/model_executor/models/qwen2.py
@@ -266,8 +266,6 @@ def forward(
 
 
 class Qwen2ForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
diff --git a/vllm/model_executor/models/xverse.py b/vllm/model_executor/models/xverse.py
index 639c3443bc369..b61721999ca9b 100644
--- a/vllm/model_executor/models/xverse.py
+++ b/vllm/model_executor/models/xverse.py
@@ -269,8 +269,6 @@ def forward(
 
 
 class XverseForCausalLM(nn.Module, SupportsLoRA):
-    supports_lora = True
-
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",