Update app.py

app.py

@@ -2,14 +2,28 @@ import spaces
 import os
 import time
 import torch
-from transformers import AutoProcessor, AutoModelForVision2Seq
 import gradio as gr
 from threading import Thread
 from PIL import Image
 
-# Model and processor initialization
-processor = AutoProcessor.from_pretrained("Qwen/QVQ-72B-Preview")
-model = AutoModelForVision2Seq.from_pretrained("Qwen/QVQ-72B-Preview").cuda().eval()
+# Install required packages
+import subprocess
+subprocess.run('pip install --upgrade transformers', shell=True)
+subprocess.run('pip install accelerate', shell=True)
+
+from transformers import AutoProcessor, AutoModelForCausalLM
+
+# Model and processor initialization with trust_remote_code=True
+processor = AutoProcessor.from_pretrained(
+    "Qwen/QVQ-72B-Preview",
+    trust_remote_code=True
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    "Qwen/QVQ-72B-Preview",
+    trust_remote_code=True,
+    device_map="auto"
+).eval()
 
 # Footer
 footer = """
@@ -26,9 +40,12 @@ def process_image(image, text_input=None):
     
     # Prepare inputs
     if text_input:
-        inputs = processor(text=text_input, images=image, return_tensors="pt").to("cuda:0")
+        inputs = processor(text=text_input, images=image, return_tensors="pt")
     else:
-        inputs = processor(images=image, return_tensors="pt").to("cuda:0")
+        inputs = processor(images=image, return_tensors="pt")
+    
+    # Move inputs to the same device as the model
+    inputs = {k: v.to(model.device) for k, v in inputs.items()}
     
     # Generate output
     outputs = model.generate(**inputs, max_new_tokens=1000)