REFERENCE DPO IMPLEMENTATION

Summary from the TLDR AI newsletter

Direct preference optimization is a stable alternative to RLHF using very similar data. This repository contains an implementation for study to learn about the technique.