A WIP dataset using evalutaion of prompts as distillation and several high quality open models to answer said evaluated prompts.
The quality is not great. Some instructions are clearly just a single line from a bigger prompt (likely a recipe). It would be helpful to have corresponding responses to be "More info please" and not just assume that a partial instruction is actionable. Future me should, maybe, look into that?
