Update app.py

app.py

@@ -50,7 +50,7 @@ h1 {
 
 # Load the tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained("OnlyCheeini/greesychat-v3")
-model = AutoModelForCausalLM.from_pretrained("OnlyCheeini/greesychat-v3", device_map="auto")  # to("cuda:0") 
+model = AutoModelForCausalLM.from_pretrained("OnlyCheeini/greesychat-v3",offload_folder="offload", device_map="auto")  # to("cuda:0") 
 terminators = [
     tokenizer.eos_token_id,
     tokenizer.convert_tokens_to_ids("<|eot_id|>")