firefly.social
Continue reading on Firefly.Social
@AdamConwayIE wrote #2 in Dec: 120B on 24GB sounds like a cheat code.
The catch: experts live in RAM, and the author suggests 16–32K context if needed: useful for a voice assistant.
A different test than I need: hours of code, tools, tests and review.
https://www.xda-developers.com/running-120b-local-llm-24gb-vram-smart-home/