Update app.py

app.py

@@ -15,10 +15,13 @@ use_triton = False
 
 tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
 
+device="cuda" if torch.cuda.is_available() else "cpu"):
+
 model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
         use_safetensors=True,
         #device="cuda:0",
-        device_map="auto",
+        device=device,
+        #device_map="auto",
         use_triton=use_triton,
         quantize_config=None)