Unveiling the Secret Recipe: A Guide For Supervised Fine-Tuning Small LLMs