Update app.py

app.py

@@ -17,7 +17,9 @@ print(device)
 print("********************")
 
 
-model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(model_id, device_map="cuda:0", config=config)
+
+#model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
 model = model.to(device)
 
 tokenizer = AutoTokenizer.from_pretrained(model_id)