diff --git a/vllm/lora/layers/column_parallel_linear.py b/vllm/lora/layers/column_parallel_linear.py
index 8273046bb6ecd..904025901fba7 100644
--- a/vllm/lora/layers/column_parallel_linear.py
+++ b/vllm/lora/layers/column_parallel_linear.py
@@ -340,12 +340,7 @@ class QKVParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
         packed_modules_list: list,
         model_config: PretrainedConfig | None = None,
     ) -> bool:
-        # Vision tower QKV has packed_modules_list=[] (already packed in checkpoint)
-        # Language models have packed_modules_list=[module_name]
-        # (single LoRA for qkv_proj)
-        return type(source_layer) is QKVParallelLinear and (
-            len(packed_modules_list) <= 1
-        )
+        return type(source_layer) is QKVParallelLinear and len(packed_modules_list) == 1
 
 
 class MergedQKVParallelLinearWithLoRA(MergedColumnParallelLinearWithLoRA):
diff --git a/vllm/model_executor/models/qwen2_5_vl.py b/vllm/model_executor/models/qwen2_5_vl.py
index 69fac625dde4c..76e2627b65ca6 100644
--- a/vllm/model_executor/models/qwen2_5_vl.py
+++ b/vllm/model_executor/models/qwen2_5_vl.py
@@ -1061,6 +1061,7 @@ class Qwen2_5_VLForConditionalGeneration(
     packed_modules_mapping = {
         "qkv_proj": ["q_proj", "k_proj", "v_proj"],
         "gate_up_proj": ["gate_proj", "up_proj"],
+        "qkv": ["qkv"],  # For vision tower's already-packed QKV
     }
 
     # To ensure correct weight loading and mapping.
diff --git a/vllm/model_executor/models/qwen3_vl.py b/vllm/model_executor/models/qwen3_vl.py
index eca21bad718f0..8a34e6e77faf6 100644
--- a/vllm/model_executor/models/qwen3_vl.py
+++ b/vllm/model_executor/models/qwen3_vl.py
@@ -1201,6 +1201,7 @@ class Qwen3VLForConditionalGeneration(
             "gate_proj",
             "up_proj",
         ],
+        "qkv": ["qkv"],  # For vision tower's already-packed QKV
     }
 
     supports_encoder_tp_data = True