Reproduce data generation process

[sachit-menon]

Hi, thanks for the great work! I was trying to reproduce the instruction generation process described in section 3.1 of https://arxiv.org/pdf/2305.15334.pdf and have not seen much success, likely due to differences arising from the particular examples, inference settings, etc used; I'm hoping to eventually try with e.g., Llama-2. I saw some discussion of this and a link to the generated training data in #75 from last year, but no conclusion on the data generating process itself. Have I missed this somewhere? It would be great if that code can be provided to make it possible for others to build on the innovations you've contributed here!