diff --git a/vllm/model_executor/models/llava.py b/vllm/model_executor/models/llava.py
index de3777cad0589..19effcbfc5512 100644
--- a/vllm/model_executor/models/llava.py
+++ b/vllm/model_executor/models/llava.py
@@ -75,19 +75,20 @@ def __init__(self,
                  vision_hidden_size: int,
                  text_hidden_size: int,
                  projector_hidden_act: str,
+                 multimodal_projector_bias: bool,
                  quant_config: Optional[QuantizationConfig] = None,
                  prefix: str = ""):
         super().__init__()
 
         self.linear_1 = ColumnParallelLinear(vision_hidden_size,
                                              text_hidden_size,
-                                             bias=True,
+                                             bias=multimodal_projector_bias,
                                              quant_config=quant_config,
                                              prefix=f"{prefix}.linear_1")
         self.act = get_act_fn(projector_hidden_act)
         self.linear_2 = RowParallelLinear(text_hidden_size,
                                           text_hidden_size,
-                                          bias=True,
+                                          bias=multimodal_projector_bias,
                                           quant_config=quant_config,
                                           prefix=f"{prefix}.linear_2")
 
@@ -503,6 +504,7 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
             vision_hidden_size=config.vision_config.hidden_size,
             text_hidden_size=config.text_config.hidden_size,
             projector_hidden_act=config.projector_hidden_act,
+            multimodal_projector_bias=config.multimodal_projector_bias,
             quant_config=quant_config,
             prefix=maybe_prefix(prefix, "multi_modal_projector"))
 
diff --git a/vllm/model_executor/models/llava_next.py b/vllm/model_executor/models/llava_next.py
index 185edcb8de11f..defdeb54afb6a 100644
--- a/vllm/model_executor/models/llava_next.py
+++ b/vllm/model_executor/models/llava_next.py
@@ -231,7 +231,8 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
         self.multi_modal_projector = LlavaMultiModalProjector(
             vision_hidden_size=vision_hidden_size,
             text_hidden_size=config.text_config.hidden_size,
-            projector_hidden_act=config.projector_hidden_act)
+            projector_hidden_act=config.projector_hidden_act,
+            multimodal_projector_bias=config.multimodal_projector_bias)
 
         self.language_model = init_vllm_registered_model(
             vllm_config=vllm_config,
diff --git a/vllm/model_executor/models/llava_next_video.py b/vllm/model_executor/models/llava_next_video.py
index a5002513554db..d70ae2f148ff9 100644
--- a/vllm/model_executor/models/llava_next_video.py
+++ b/vllm/model_executor/models/llava_next_video.py
@@ -253,16 +253,16 @@ def forward(self, image_features: torch.Tensor):
 class LlavaNextMultiModalProjector(nn.Module):
 
     def __init__(self, vision_hidden_size: int, text_hidden_size: int,
-                 projector_hidden_act: str):
+                 projector_hidden_act: str, multimodal_projector_bias: bool):
         super().__init__()
 
         self.linear_1 = nn.Linear(vision_hidden_size,
                                   text_hidden_size,
-                                  bias=True)
+                                  bias=multimodal_projector_bias)
         self.act = get_act_fn(projector_hidden_act)
         self.linear_2 = nn.Linear(text_hidden_size,
                                   text_hidden_size,
-                                  bias=True)
+                                  bias=multimodal_projector_bias)
 
     def forward(self, image_features: torch.Tensor) -> torch.Tensor:
         hidden_states = self.linear_1(image_features)
@@ -298,7 +298,8 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
         self.multi_modal_projector = LlavaNextMultiModalProjector(
             vision_hidden_size=config.vision_config.hidden_size,
             text_hidden_size=config.text_config.hidden_size,
-            projector_hidden_act=config.projector_hidden_act)
+            projector_hidden_act=config.projector_hidden_act,
+            multimodal_projector_bias=config.multimodal_projector_bias)
         self.language_model = init_vllm_registered_model(
             vllm_config=vllm_config,
             hf_config=config.text_config,
diff --git a/vllm/model_executor/models/llava_onevision.py b/vllm/model_executor/models/llava_onevision.py
index ac502000c3ee7..f1c06cd85967c 100644
--- a/vllm/model_executor/models/llava_onevision.py
+++ b/vllm/model_executor/models/llava_onevision.py
@@ -372,11 +372,11 @@ def __init__(self, config: LlavaOnevisionConfig):
 
         self.linear_1 = nn.Linear(config.vision_config.hidden_size,
                                   config.text_config.hidden_size,
-                                  bias=True)
+                                  bias=config.multimodal_projector_bias)
         self.act = get_act_fn(config.projector_hidden_act)
         self.linear_2 = nn.Linear(config.text_config.hidden_size,
                                   config.text_config.hidden_size,
-                                  bias=True)
+                                  bias=config.multimodal_projector_bias)
 
     def forward(self, image_features: torch.Tensor) -> torch.Tensor:
         hidden_states = self.linear_1(image_features)