Update app.py

app.py

@@ -23,7 +23,9 @@ def generate(model_name, image, text):
     # print("Model initialized, now moving to cuda")
     # model_pipe.to("cuda")
     print("Generating!")
-    return model_pipe(images=[image], texts=[text])
+    with torch.autocast(device_type='cuda', dtype=torch.bfloat16):
+        result = model_pipe(images=[image_tensor], texts=[text])
+    return result
 
 iface = gr.Interface(
     fn=generate,  # function to call