Numen - AI NPCs for FNV

1 recommendations 3 thanked
LordGalactus
LordGalactus
6
9 July 2026, 07:58
tryed using unsloth/gemma-4-12b-it-qat and i get this crash:
Engine protocol runtime llama-server for yP/ze1blYcECWx0SrD/ws7tI exited before becoming healthy. exitCode=3221225477, signal=null
1 reply, 3 total
Eddoursul
Eddoursul Admin
344 37 2
9 July 2026, 08:45
How much RAM/VRAM you got? You need 32 Gb RAM + 16 Gb VRAM to run the recommended minimum, Gemma 4 12B, AND the game. You can try a smaller model, but it will be much less coherent. Without hardware, you can use cloud providers instead.
1 reply, 2 total
1 reply
Eddoursul
Eddoursul Admin
344 37 2
9 July 2026, 21:30
None, Gemma 4 12B is the minimum I found able to stay up to the task. Qwen 9B was noticeably slipping in comparison. I would gladly test other proposed options but Gemma 4 12B, released by Google in June, is absolutely the frontier at this size. Several months into the future we may have other options, though.

I have instructions in the description and in the default Agent.ini how to use free tier cloud Gemini+Gemma models, no hardware needed.

In fact, I strongly recommend to try Gemini even a bit to understand defferences in writing.