Update app.py

app.py

@@ -9,13 +9,11 @@ model_name = "unsloth/Meta-Llama-3.1-8B-Instruct"
 tokenizer = AutoTokenizer.from_pretrained(model_name, load_in_4bit=True)
 
 # Load the base model with adapters, ensuring it's loaded in 4-bit
-model = AutoAdapterModel.from_pretrained(model_name, load_in_4bit=True).to("cpu")
+model = AutoAdapterModel.from_pretrained(model_name, load_in_4bit=True)
 
 # Load the adapter
 model.load_adapter("Braszczynski/Llama-3.2-3B-Instruct-bnb-4bit-460steps")
 
-# Free up unused memory
-torch.cuda.empty_cache()
 
 def respond(message, history, system_message, max_tokens, temperature, top_p):
     # Combine system message and chat history