Was distillation removed? housekeeping issue or deeper stuff going on?
3 Likes
Likely developer engagement was approaching zero.
Distillation is good for using bulk data to produce submodels. For curated synthetic data, where both a high-quality model is used and custom prompting to generate that, inputs not employed in inference in a final specialized fine-tuning model, it didn’t fit the bill.
1 Like
It was intentionally removed, because the content had gone out of date, and the practice is already implicitly covered in other documentation on the site. We should have redirected this though instead of letting this be a 404, we will fix.
Thanks!
7 Likes
