Towards a data-centric RLHF: simple metrics for comparison of preference data sets
The goal of aligning language models with human preferences requires data that reveals these preferences. Ideally, time and money can ...
Tag: datacentric
Recommended.
