I still don't know definitively what was going on with my eye but it feels MUCH better today. The zombie-like redness and persistent itching is largely gone. Thank goodness - it was becoming concerning...
So I've been stuck without AI help for a while now... I've blown through all my FREE tokens or whatever, so I've been trying to find a permanent local solution. An AI running off my own system - so I can ask it as many questions as I want... well, I'm understanding the video card price crisis now. AI models offload themselves into GPU RAM... the more GPU RAM you have the better an AI model you can run... and some of them get HUGE 30/40G (I don't think ANY video card has anywhere near that much RAM - lol!) but for every gig you get more options... I have a paltry (by today's standards) 6G of video card RAM... It is still fine for gaming, but for AI it is really scrapping the bottom of the barrel... You need BOTH a decent RAM and a decent context size in order to do anything like what I'm trying to do. If these things don't fit in the VRAM then the AI model runs un-usably slow. I've been experimenting with all sorts of options, LM Studio, Llama, Jan... so many different models. Some work. Some don't. Most are slow. Some are unusable slow... So it really is a tedious process and repeating the core of the game over and over asking for help... Ugh... I thought the Jan app was promising but it wouldn't detect my video card... I tried all the suggestions from their help section but still nothing. I finally DID get it working but it was STILL slow or I couldn't figure out how to force VRAM usage. UGH... I figured at some point I've installed and half-assed so many solutions I needed to start over. I'm back to CachyOS and I'm trying LM Studio again. It seems to be running Gemma 4-e4b (i think the e4b is the number of parameters - 4 billion... most are MUCH higher, 16B 20B 40B etc..but those would never run on my system)... It is also about learning to use AI and understanding how to work with its limitations... which is very much an ongoing process...
UPDATE May 22 @ 12:46PM
Alrighty. *Might* have a winner....I'm running the jan.ai applimage (the .deb does not run). I had to add the nccl libraries... $sudo pacman -S nccl... I've never had a model run as fast as Jan-v3-4b-base-instruct-Q4_K_M" I'll have to work with it for a while to see if it is up to task, but it is VERY promising... Even if it is just okay, it is so fast it will likely be worth using! HUZZAH!