Issue with New Feature "switch models, settings and configs at runtime"

[byslexic]

As of the newest version, you can "switch models, settings and configs at runtime"
Could anyone address how this is any different than it was previously?
Previously, on opening Kobold Lite ("runtime"), one has always had the ability to change any model's config from the Admin panel, and the ability to prepare .kcpps configs for different layers, backends, models, etc. which were available with the Load button.
What exactly is the new functionality, and how (or why) do you use it?

[jabberjabberjabber]

Kobold Lite is the web interface that you load in a browser. KoboldCpp is the backend application that runs the inference engine and the pops open a configuration GUI when you open it. The new functionality let's you create an API endpoint that will allow you to remotely configure settings through the web UI (Lite).

[byslexic]

Thank you, I think I understand the difference now.
Love Kobold, don't love the thin documentation.

[jabberjabberjabber]

Thank you, I think I understand the difference now. Love Kobold, don't love the thin documentation.

I would argue kobold has excellent documentation. Considering how configurable it is and how technically complex each of the configurations are, it is incredibly difficult to document everything so that everyone can understand it without being immediately overwhelmed.

There are basically only a few options with the available constraints of a non-profit open source release:

1. Exclude everyone who isn't intimately familiar with how AI works and knows how to code (llama.cpp)
2. Remove all features so that everyone can use it (ollama)
3. Add as many features as people need and try your best to document it with people who volunteer to do it while upholding rigorous standards, but failing sometimes to keep up with the abilities in the releases, which come out every few weeks with sometimes game changing additions that have been requested by the community (koboldcpp)

I would pick 3 every time.

[byslexic]

Let's agree to disagree. I'm grateful for this forum and you, who chose to explain to a newbie. While the documentation is copious, it lacks detail. Although I fully support the Kobold team and highly appreciate their efforts. I simultaneously encourage them to be more thorough in the documentation. If it were thorough, I would not be here asking this question, as I couldn't find anything in the FAQ explaining this feature.

As a solo software author myself, I understand fully the constraints due to lack of resources and the challenge of writing intelligible Help files. The 3 options above are not the only choices. Thorough documentation is a choice; it expands the userbase, growing your support. It can make even the most intricate features accessible to the average user, if written well. And it is time consuming. But then, everything about software development is.

As exciting as new features are to those who already grasp their purpose and power, they are sometimes baffling to those not in-the-know, even after reading the FAQ.

[LostRuins]

Contributions are always welcome - if you spot an error or lacking information and would like to contribute, it would be appreciate

I assume you've read thru the readme and the wiki? https://github.com/LostRuins/koboldcpp/wiki

[byslexic]

@LostRuins Yes, thanks, I have read the Wiki and the latest Release Notes (Is that the ReadMe?)

I'd be very happy to contribute to the Wiki if I had access!
However, my contributions would be limited to my understanding of the program, so I'm not sure how I can help just yet. Would it be helpful to have notes in the Wiki indicating where information is lacking or could be more explicit for less knowledgeable users (like myself)?

[LostRuins]

Unfortunately, Github does not allow pull requests to the wiki directly , but if you'd like to contribute,

1. clone the existing wiki with git clone https://github.com/LostRuins/koboldcpp.wiki.git
2. Make your changes
3. simply open a Github Issue and paste your modified version there, and I will review and merge it