Excited to see our 🍮Flan-Palm🌴 work finally published in @JmlrOrg 2024! Looking back, I see this work as pushing hard on scaling: post-training data, models, prompting, & eval. We brought together the methods and findings of many awesome prior works, scaled them up, and evaluated the impact extensively, w/ human preference evals which were not as common or widely available then. (The key was a lot of manual test example creation and blind reviewing -- to really see the effects of different decisions.) 📜 jmlr.org/papers/v25/23-… 📽️ youtu.be/oqi0QrbdgdI?si… It was a pleasure working w/ @_jasonwei @hwchung27 & Le Hou. 1/
See @_jasonwei's reflections on the work, and how they compare to what is new: x.com/_jasonwei/stat… 2/
See @_jasonwei's reflections on the work, and how they compare to what is new: x.com/_jasonwei/stat… 2/