As I understand it, these instruction tuned models are released with SFT + DPO training, presumeably trained in different stages. Is there a plan to release just the SFT tuned models? This would be quite helpful for those of us working on DPO tuning.
