TL;DR - A ~4M example, ~3B token CoT dataset designed around helping small models think more concisely, accurately and reliably. Hi all! For the past few months I have been working on a dataset designed around improving small model performance through a structured framework (or Scaffold) for thinkin
