DPO TO ENCOURAGE DESCRIPTIVENESS (GITHUB GIST)

Summary from the TLDR AI newsletter

A minimal code set up with TRL to tune a model to be more descriptive.