Update app.py

app.py

@@ -11,18 +11,24 @@ import subprocess
 subprocess.run('pip install --upgrade transformers', shell=True)
 subprocess.run('pip install accelerate', shell=True)
 
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoConfig, PreTrainedModel, AutoTokenizer
 
 # Model and tokenizer initialization
 model_name = "Qwen/QVQ-72B-Preview"
 
+config = AutoConfig.from_pretrained(
+    model_name,
+    trust_remote_code=True
+)
+
 tokenizer = AutoTokenizer.from_pretrained(
     model_name,
     trust_remote_code=True
 )
 
-model = AutoModelForCausalLM.from_pretrained(
+model = PreTrainedModel.from_pretrained(
     model_name,
+    config=config,
     trust_remote_code=True,
     device_map="auto",
     torch_dtype=torch.float16
@@ -42,14 +48,30 @@ def process_image(image, text_input=None):
         # Convert image to PIL format
         image = Image.fromarray(image).convert("RGB")
         
-        # Prepare prompt
+        # Prepare inputs
         if text_input:
-            prompt = f"<image>Please describe this image and answer: {text_input}</image>"
+            messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {"image": image},
+                        {"text": text_input}
+                    ]
+                }
+            ]
         else:
-            prompt = "<image>Please describe this image in detail.</image>"
+            messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {"image": image},
+                        {"text": "Please describe this image in detail."}
+                    ]
+                }
+            ]
         
         # Generate response
-        response = model.chat(tokenizer, prompt, history=[], images=image)
+        response = model.chat(tokenizer, messages=messages)
         
         return response
     except Exception as e: