Our recipe for mid-training and post-training on FineDNA is fully open and we hope these lessons on breaking past the distillation ceiling help model builders do the same in new domains!
Dataset ⬇️
Part of a thread · earlier post
Our recipe for mid-training and post-training on FineDNA is fully open and we hope these lessons on breaking past the distillation ceiling help model builders do the same in new domains!
Dataset ⬇️
Lewis Tunstall · x.comPublished
Original title: Our recipe for mid-training and post-training on FineDNA is fully open and we hope these lessons on breaking past the distillation ceilin...