diff --git a/src/llamafactory/model/model_utils/visual.py b/src/llamafactory/model/model_utils/visual.py
index a8010579..ba2bf5c9 100644
--- a/src/llamafactory/model/model_utils/visual.py
+++ b/src/llamafactory/model/model_utils/visual.py
@@ -76,7 +76,7 @@ def _register_composite_model(
         model_type=model_type,
         projector_key=projector_key or "multi_modal_projector",
         vision_model_keys=vision_model_keys or ["vision_tower"],
-        language_model_keys=language_model_keys or ["language_model"],
+        language_model_keys=language_model_keys or ["language_model", "lm_head"],
         lora_conflict_keys=lora_conflict_keys or [],
     )
 
@@ -200,12 +200,12 @@ def patch_target_modules(
 
 
 _register_composite_model(
-    model_type="internvl",
+    model_type="gemma3",
 )
 
 
 _register_composite_model(
-    model_type="gemma3",
+    model_type="internvl",
 )
 
 
@@ -246,14 +246,8 @@ _register_composite_model(
     lora_conflict_keys=["audio_projection_layer"],
 )
 
-
 _register_composite_model(
-    model_type="paligemma",
-)
-
-
-_register_composite_model(
-    model_type="video_llava",
+    model_type="mistral3",
 )
 
 
@@ -264,7 +258,7 @@ _register_composite_model(
 
 
 _register_composite_model(
-    model_type="mistral3",
+    model_type="paligemma",
 )
 
 
@@ -303,3 +297,8 @@ _register_composite_model(
     else ["model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )
+
+
+_register_composite_model(
+    model_type="video_llava",
+)