Data Repetition Beats Data Scaling in Long-CoT Supervised Fine-Tuning - Explained Simply | ArXiv Explained