From 9640f79ae5bb7515d3b44ec45d711b149d564dab Mon Sep 17 00:00:00 2001
From: Jewon Lee <105219284+je1lee@users.noreply.github.com>
Date: Tue, 27 Jan 2026 17:33:01 +0900
Subject: [PATCH] [fix] add visual.pos_embed to Qwen3-VL visual model keys
 (#10139)

---
 src/llamafactory/model/model_utils/visual.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/llamafactory/model/model_utils/visual.py b/src/llamafactory/model/model_utils/visual.py
index 955c52291..80f2e187b 100644
--- a/src/llamafactory/model/model_utils/visual.py
+++ b/src/llamafactory/model/model_utils/visual.py
@@ -356,7 +356,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_vl",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
+    vision_model_keys=["visual.pos_embed", "visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
     language_model_keys=["language_model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )
@@ -365,7 +365,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_vl_moe",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
+    vision_model_keys=["visual.pos_embed", "visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
     language_model_keys=["language_model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )
@@ -374,7 +374,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_omni_moe_thinker",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list", "audio_tower"],
+    vision_model_keys=["visual.pos_embed", "visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list", "audio_tower"],
     language_model_keys=["model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )