samuelcardillo

/

Carnice-Qwen3.6-MoE-35B-A3B

like 24

Hi

I'm testing this model for Hermes agent tasks and have two quick questions:

1. Thinking mode: Does it need to be enabled?
2. Sampling params: Any officially recommended settings?

I've seen the usual temp 0.6 / top-p 0.95 / top-k 20 / min-p 0 / repetition_penalty 1.0 in base model cards, but want to confirm if it's still optimal for this variant.

Current setup: reasoning mode: on
Tested params: same as above.

Any configs or tips would be helpful! Thanks