Fine-tuning LLMs from scratch and exploring various methods with evals - View it on GitHub