How R1-style reasoning models were trained — Teach a Small AI to Think | unicodesnwrites