Training framework with a goal to explore the frontier of sample efficiency of small language models - View it on GitHub