Update app.py

app.py

@@ -6,7 +6,7 @@ Optimized for Hugging Face deployment with efficient resource usage
 
 import gradio as gr
 import torch
-from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
+from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
 from qwen_vl_utils import process_vision_info
 from PIL import Image
 import json
@@ -39,9 +39,9 @@ def load_model_cached(model_name: str = "Qwen/Qwen2.5-VL-3B-Instruct"):
         
         try:
             # Load with memory optimization for HF Spaces
-            MODEL = Qwen2VLForConditionalGeneration.from_pretrained(
+            MODEL = Qwen2_5_VLForConditionalGeneration.from_pretrained(
                 model_name,
-                torch_dtype=torch.bfloat16 if DEVICE == "cuda" else torch.float32,
+                torch_dtype='auto',
                 device_map="auto" if DEVICE == "cuda" else None,
                 trust_remote_code=True,
                 attn_implementation="eager",  # More stable for HF Spaces