diff --git a/README.md b/README.md
index 81ae69e..72bd59d 100644
--- a/README.md
+++ b/README.md
@@ -267,6 +267,8 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 from transformers import GenerationConfig
 from qwen_generation_utils import make_context, decode_tokens, get_stop_words_ids
 
+# To generate attention masks automatically, it is necessary to assign distinct
+# token_ids to pad_token and eos_token, and set pad_token_id in the generation_config.
 tokenizer = AutoTokenizer.from_pretrained(
     './',
     pad_token='<|extra_0|>',