Update app.py

app.py

@@ -32,7 +32,7 @@ m = AutoModelForCausalLM.from_pretrained(
     torch_dtype=torch.bfloat16,
     device_map={"": 0}
 )
-tok = LlamaTokenizer.from_pretrained(model_name)
+tok = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf")
 tok.bos_token_id = 1
 
 stop_token_ids = [0]