feat: tokenized details

README.md

@@ -27,4 +27,11 @@ size_categories:
 
 From the Frontier Research Team at **Takara.ai**, we present **MicroPajama** — a dataset made for distillation and feature extraction without wikipedia from the larger SlimPajama.  
 
----
+---
+
+This dataset contains **253,636,240* tokens using the BAAI/bge-large-en-v1.5 wordpiece tokenizer, you can reproduce this result in scripts/tokenize/main.py.
+
+---
+For research inquiries and press, please reach out to research@takara.ai
+
+> 人類を変革する

scripts/tokenize/main.py

@@ -0,0 +1,13 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer
+
+
+def main():
+    ds = load_dataset("takara-ai/micropajama", split="train")
+    tok = AutoTokenizer.from_pretrained("BAAI/bge-large-en-v1.5")
+    lens = ds.map(lambda b: {"len": [len(x) for x in tok(b["text"], add_special_tokens=False).input_ids]}, batched=True, remove_columns=ds.column_names)
+    print(sum(lens["len"]))
+
+
+if __name__ == "__main__":
+    main()