diff --git a/app.py b/app.py
index b2d6fb0..5a4ddd0 100644
--- a/app.py
+++ b/app.py
@@ -17,7 +17,8 @@
 import verovio
 model_name = "ucaslcl/GOT-OCR2_0"
 tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-model = AutoModel.from_pretrained(model_name, trust_remote_code=True, low_cpu_mem_usage=True, device_map='cuda', use_safetensors=True).eval().cuda()
+#model = AutoModel.from_pretrained(model_name, trust_remote_code=True, low_cpu_mem_usage=True, device_map='cuda', use_safetensors=True).eval().cuda()
+model = AutoModel.from_pretrained(model_name, trust_remote_code=True, low_cpu_mem_usage=True, device_map='cpu', use_safetensors=True).eval()
 
 UPLOAD_FOLDER = "./uploads"
 RESULTS_FOLDER = "./results"
@@ -32,7 +33,8 @@ def image_to_base64(image):
     return base64.b64encode(buffered.getvalue()).decode()
 
 q_model_name = "Qwen/Qwen2-VL-2B-Instruct"
-q_model = Qwen2VLForConditionalGeneration.from_pretrained(q_model_name, torch_dtype="auto").cuda().eval()
+#q_model = Qwen2VLForConditionalGeneration.from_pretrained(q_model_name, torch_dtype="auto").cuda().eval()
+q_model = Qwen2VLForConditionalGeneration.from_pretrained(q_model_name, torch_dtype="auto").eval()
 q_processor = AutoProcessor.from_pretrained(q_model_name, trust_remote_code=True)
 
 def get_qwen_op(image_file, model, processor):
@@ -53,7 +55,8 @@ def get_qwen_op(image_file, model, processor):
             }
         ]
         text_prompt = q_processor.apply_chat_template(conversation, add_generation_prompt=True)
-        inputs = q_processor(text=[text_prompt], images=[image], padding=True, return_tensors="pt").to("cuda")
+        #inputs = q_processor(text=[text_prompt], images=[image], padding=True, return_tensors="pt").to("cuda")
+        inputs = q_processor(text=[text_prompt], images=[image], padding=True, return_tensors="pt")
         inputs = {k: v.to(torch.float32) if torch.is_floating_point(v) else v for k, v in inputs.items()}
 
         generation_config = {