fix: use correct Qwen2_5_VL class for Qwen2.5-VL model

- Replace Qwen2VLForConditionalGeneration with Qwen2_5_VLForConditionalGeneration
- Install transformers from source for Qwen2.5-VL support
- Fix model loading architecture mismatch

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

app.py

@@ -7,7 +7,7 @@ import gradio as gr
 import spaces
 import torch
 from PIL import Image
-from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
+from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
 from qwen_vl_utils import process_vision_info
 
 # Qwen2.5-VL 모델 ID
@@ -35,10 +35,10 @@ def _extract_json_block(text: str) -> Optional[str]:
 def extract_text_from_image(image: Image.Image) -> str:
     """Qwen2-VL로 이미지에서 텍스트 추출"""
     try:
-        # Qwen2-VL 모델 로드
-        model = Qwen2VLForConditionalGeneration.from_pretrained(
+        # Qwen2.5-VL 모델 로드
+        model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
             MODEL_ID,
-            torch_dtype=torch.bfloat16,
+            torch_dtype="auto",
             device_map="auto"
         )
         processor = AutoProcessor.from_pretrained(MODEL_ID)

requirements.txt

@@ -1,5 +1,5 @@
 gradio>=4.0.0
-transformers>=4.37.0
+git+https://github.com/huggingface/transformers
 torch>=2.1.0
 torchvision
 Pillow