TUI client for LLM inference? #888
magikRUKKOLA
started this conversation in
Ideas
Replies: 1 comment
-
|
Yes, indeed, this stupid charmbracelet/mods terminal happily eats away my CPU: 30 seconds of CPU time while outputting a few thousands of tokens in decode. This is pretty crazy. Its over a minute of CPU time by now. lool [EDIT]: [EDIT2]: |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
I had been using charmbracelet/mods until I'd noticed that sometimes its lagging behind the llama server while eating up a few CPU cores. charmbracelet/mods#635
I've done some test and yeah, its extremely slow. For example, in case I would want to output 20k ctx into the terminal it takes:
Lets compare it to some regular stuff like hightlight program with a custom nested rules for the markdown:
And here is the results for the same 20k ctx:
So its about ten times faster than mods.
And mods is all I found which is working more-or-less stable.
I mean, all the TUI client does is syntax highlighting and conversation history management. It shouldn't be that resourse-intensive, right? But what that actually mean? Is there any decent and stable TUI or not? Because I can't find it.
Beta Was this translation helpful? Give feedback.
All reactions