diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py
index 6078d8aca..af7a05f95 100644
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -265,7 +265,6 @@ class ServerArgs:
     context_length: Optional[int] = None
     is_embedding: bool = False
     enable_multimodal: Optional[bool] = None
-    limit_mm_data_per_request: Optional[Union[str, Dict[str, int]]] = None
     revision: Optional[str] = None
     model_impl: str = "auto"
 
@@ -637,6 +636,7 @@ class ServerArgs:
     enable_prefix_mm_cache: bool = False
     mm_enable_dp_encoder: bool = False
     mm_process_config: Optional[Dict[str, Any]] = None
+    limit_mm_data_per_request: Optional[Union[str, Dict[str, int]]] = None
 
     # For checkpoint decryption
     decrypted_config_file: Optional[str] = None
@@ -2485,13 +2485,6 @@ class ServerArgs:
             action="store_true",
             help="Enable the multimodal functionality for the served model. If the model being served is not multimodal, nothing will happen",
         )
-        parser.add_argument(
-            "--limit-mm-data-per-request",
-            type=json.loads,
-            default=ServerArgs.limit_mm_data_per_request,
-            help="Limit the number of multimodal inputs per request. "
-            'e.g. \'{"image": 1, "video": 1, "audio": 1}\'',
-        )
         parser.add_argument(
             "--revision",
             type=str,
@@ -4403,6 +4396,13 @@ class ServerArgs:
             default=ServerArgs.mm_enable_dp_encoder,
             help="Enabling data parallelism for mm encoder. The dp size will be set to the tp size automatically.",
         )
+        parser.add_argument(
+            "--limit-mm-data-per-request",
+            type=json.loads,
+            default=ServerArgs.limit_mm_data_per_request,
+            help="Limit the number of multimodal inputs per request. "
+            'e.g. \'{"image": 1, "video": 1, "audio": 1}\'',
+        )
 
         # For checkpoint decryption
         parser.add_argument(