KoboldAI-Client
KoboldAI-Client copied to clipboard
Add a TUI frontend
As KoboldAI can use a lot of RAM VRAM etc. It would be interesting if you could use it in a TUI mode since web browsers require a lot of RAM, this would also allow you to just run it in a TTY which is very minimal
Also just a question, but could further optimizations be made? I remember being able only generate 512x512 on stable diffusion, but with more optimizations I am able to do 1024x1024 and even higher comfortably
note: take a look at charm.sh for an kick ass TUI framework.
KoboldAI isn't RAM/VRAM intensive because of the web server. A TUI mode would be cute, but is certainly not going to be a performance win. A TUI mode would also require writing an interface for configuration, and is generally less useful than having a web interface, so I don't see this being added (unless you wanted to submit a PR).
https://github.com/rothgar/awesome-tuis#libraries
- https://github.com/louipc/turses is built on urwid for the TUI: https://github.com/urwid/urwid
- https://github.com/Textualize/textual
jupyter_consoleis the IPython REPL with non-ipykernel Jupyter kernels https://github.com/jupyter/jupyter_console- re: a [sensorcraft, Minecraft,] jupyter kernel https://news.ycombinator.com/item?id=32657051
- https://jupyter-client.readthedocs.io/en/latest/kernels.html#making-kernels-for-jupyter
- https://github.com/microsoft/vscode-jupyter/wiki/Connecting-to-a-remote-Jupyter-server-from-vscode.dev
- https://research.google.com/colaboratory/local-runtimes.html
import IPython; IPython.display.display??
There is one already: https://github.com/atisharma/koboldterm
But changes in the API seem to have broken it.
From https://news.ycombinator.com/item?id=35767901 :
Could Frogmouth display Markdown cells in Jupyter Notebooks on the CLI? FWIW, IPython is built with Python Prompt Toolkit and Jupyter_console is IPython for non-python kernels;
conda install -y jupyter_console; jupyter kernelspec -h. But those only do%logstart -o example.out.py; Markdown in notebooks on the CLI is basically unheard-of.
Frogmouth does CLI Markdown output in Python and is built with textual; https://github.com/Textualize/textual :
Textual adds interactivity to Rich with an API inspired by modern web development.
On modern terminal software (installed by default on most systems), Textual apps can use 16.7 million colors with mouse support and smooth flicker-free animation. A powerful layout engine and re-usable components makes it possible to build apps that rival the desktop and web experience.