diff --git a/README.md b/README.md
index d1a9127..9b9130d 100644
--- a/README.md
+++ b/README.md
@@ -1,2 +1,3 @@
 # dolly
 Dolly is a custom LLM built from scratch
+only for personal usage
diff --git a/tokenizer_part/train_tokenizer.py b/tokenizer_part/train_tokenizer.py
new file mode 100644
index 0000000..a072183
--- /dev/null
+++ b/tokenizer_part/train_tokenizer.py
@@ -0,0 +1,21 @@
+from datasets import load_dataset
+import os
+os.environ["HF_DATASETS_CACHE"] = "./data/"
+wikipedia = load_dataset("wikipedia", language="zh")
+bookcorpus = load_dataset("bookcorpus")
+openwebtext = load_dataset("openwebtext")
+cc100 = load_dataset("cc100", language="en")
+
+wikipedia_text = wikipedia['train']['text']
+bookcorpus_text= bookcorpus["train"]["text"]
+openwebtext_text = openwebtext['train']['text']
+cc100_text = cc100['train']['text']
+combined_text = wikipedia_text + openwebtext_text+ bookcorpus_text
+print("finished")
+from tokenizers import Tokenizer, models, pre_tokenizers, trainers
+tokenizer = Tokenizer(models.BPE())
+tokenizer.pre_tokenizer = pre_tokenizers.Whitespace()
+trainer = trainers.BpeTrainer(vocab_size=30000, min_frequency=2)
+tokenizer.train_from_iterator(combined_text, trainer=trainer)
+tokenizer.save("path_to_save_your_tokenizer.json")
+