matrix07012 @matrix@gameliberty.club

I got 27b working and it's smart enough

I liked Gemma more than than Qwen, but unfortunately 12b is too retarded to format answers correctly.

Now it's not even funny, just retarded.

new, less schizo, prompt

xAI really cooked with 4.1

Ok, it's getting stale

I turned cache off and restarted the process! Why is it doing all caps even now?

interesting though, seems like prompt caching can poison rhe context and start things like typing in all caps

qwen3-vl-8b abliterated seems to work well

Way too slow if you can't fit it in VRAM.
Letting the GPU do all the processing and letting it overflow into RAM is faster than having llama.cpp split it between CPU and GPU