[Question] On-Policy Distillation Support + SFT Engine

[bohdan-nd]

Hey PyTorch Team,

I recently started integrating torchforge in my workflow and I've been thinking about adding a few features that are not supported yet

1. What do you think about integration of on-policy distillation? I am currently building a pipeline for my experiments, and would love to contribute

2. Why did you decide to implement a separate engine for SFT instead of using the TitanTrainer?

Thank you!