CLI Config File Changes

[Ali Maredia]

Hi Instructlab community,

Last week I opened up a dev doc that discusses lots of changes around the configuration file (https://github.com/instructlab/dev-docs/pull/162).

I also talked to a few of the maintainers (Jaideep, Charlie, and Nathan) about some of the changes I proposed.

In our discussions there were a few follow up items of discussion we wanted the community's input on:

1. Currently `ilab config init` downloads models, downloads the upstream taxonomy, and sets the system profile, through interactive prompting. In my proposal `ilab config init` goes away. However there's not a proposal for what an alternate workflow would look like.

I was wondering how folks felt about the interactive prompting, and taxonomy and model downloads being done through separate commands. Perhaps we go back to `ilab init` that does all of it but better than before?

2. In my discussion with Charlie, Jaideep, and Nathan we realized we had a philosophical difference about the purpose of the configuration file. 

My proposal envisions a configuration system that just throws a warning for each invalid entry in the file. The purpose of this is to be able to add and remove configuration options easily between releases as we move toward a 1.0 (configuration backwards compatibility can be revisited there). The ramifications for this are that there are no backward compatibility guarantees between configuration files before 1.0.

Other folks see the configuration file as a contract of stability for ilab. almost like an API. Invalidating and reorganizing configuration options between releases is poor user experience.

We also discussed config migration tools between versions, and a possible split between user config (that would be more stable) and system config (that would be more dynamic).

Let me know what you think. I understand that changes like this have deep ramifications and I thought this discussion would be a great way to get the community email list going.

Best,

Ali