remove cuda

app.py

@@ -40,7 +40,7 @@ models = {
     "Qwen/Qwen2-VL-7B-Instruct": Qwen2VLForConditionalGeneration.from_pretrained(
         "Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto"
     )
-    .cuda()
+    .cpu()
     .eval()
 }
 
@@ -95,7 +95,7 @@ def run_example(image, text_input=None, model_id="Qwen/Qwen2-VL-7B-Instruct"):
         padding=True,
         return_tensors="pt",
     )
-    inputs = inputs.to("cuda")
+    inputs = inputs.to("cpu")
 
     # Inference: Generation of the output
     generated_ids = model.generate(**inputs, max_new_tokens=1024)