Update README.md

README.md

@@ -11,6 +11,8 @@ Baseline for continuous high-quality updates.
 
 This dataset is meant to make small models more nuanced on their instruction-following skills.
 
+It is also useful for byte-level models since a great part of it requires granularity bpe models usually lack.
+
 It followed human-curated decontamination and syntax-level dedup.
 
 The final dataset contains around 205k examples and we'll continue expanding it.