sft

Supervised fine-tuning runs, data mixtures, training settings, and lessons from instruction tuning or behavior shaping with labeled examples.