LitGPT demystifies LLM training/finetuning with zero abstractions. Seeing models like Llama 3 and Gemma implemented from scratch, optimized for performance and memory, is incredibly valuable for deeper understanding and customizability.
[Link to repo below]