Update app.py

app.py

@@ -98,9 +98,8 @@ class Qwen3VLAutoThinkDemo:
         self.model = Qwen3VLForConditionalGeneration.from_pretrained(
             model_path,
             dtype="bfloat16",
-            device_map="auto",
             attn_implementation="sdpa",
-        ).eval()
+        ).to('cuda').eval()
 
         self.processor = AutoProcessor.from_pretrained(model_path)
         self.tokenizer = AutoTokenizer.from_pretrained(model_path)
@@ -200,8 +199,8 @@ class Qwen3VLAutoThinkDemo:
         Returns:
             Dictionary containing response and metadata
         """
-        if self.model.device.type != "cuda":
-            self.model.to("cuda")
+        # if self.model.device.type != "cuda":
+            # self.model.to("cuda")
         
         # Prepare message
         message = [{"role": "system", "content": self.system_prompt}]