aquif-ai

/

aquif-3.5-Max-42B-A3B

like 70

Follow

aquif AI 86

it should be a thinking only model, so you don't have to toggle reasoning like in aquif-3.5-Plus

I've been test it before. It acts like the instruct model. Temp = 0.6, repeat penalty = 1.0, llamacpp

llama-server.exe --temp 0.6 --min-p 0.0 --top-p 0.95 --top-k 20 --dry-multiplier 0.4 --dry-allowed-length 4 --ctx-size 32768 --samplers "top_k;top_p;min_p;temperature;dry;typ_p;xtc" -m ""

finally it can reason, but missing opening think tags

Now, it overthinking. Almost 12k tokens spent to solve the medium tasks in LCB.

how's the model? we want feedback to improve our next releases, like aquif-4.

I always wonder why there is gguf by unsloth that doesn't seem to work properly.