[deps] adapt to transformers v5 (#10147)

Co-authored-by: frozenleaves <frozen@Mac.local>
Co-authored-by: hiyouga <hiyouga@buaa.edu.cn>

src/llamafactory/model/loader.py

@@ -22,7 +22,6 @@ from transformers import (
     AutoModelForImageTextToText,
     AutoModelForSeq2SeqLM,
     AutoModelForTextToWaveform,
-    AutoModelForVision2Seq,
     AutoProcessor,
     AutoTokenizer,
 )
@@ -166,11 +165,9 @@ def load_model(
         else:
             if type(config) in AutoModelForImageTextToText._model_mapping.keys():  # image-text
                 load_class = AutoModelForImageTextToText
-            elif type(config) in AutoModelForVision2Seq._model_mapping.keys():  # image-text
-                load_class = AutoModelForVision2Seq
             elif type(config) in AutoModelForSeq2SeqLM._model_mapping.keys():  # audio-text
                 load_class = AutoModelForSeq2SeqLM
-            elif type(config) in AutoModelForTextToWaveform._model_mapping.keys():  # audio hack for qwen omni
+            elif type(config) in AutoModelForTextToWaveform._model_mapping.keys():  # audio-text for qwen omni
                 load_class = AutoModelForTextToWaveform
             else:
                 load_class = AutoModelForCausalLM

src/llamafactory/model/model_utils/visual.py

@@ -374,7 +374,13 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_omni_moe_thinker",
     projector_key="visual.merger",
-    vision_model_keys=["visual.pos_embed", "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"],
 )