Someone on linux with more than one gpu please try running
llama-cli -m model.gguf -p 'test' -bs --samplers 'top_k;temperature' -c 1000 --no-warmup
and see if it segfaults.
You can try with https://huggingface.co/Qwen/Qwen3-0.6B-GGUF/tree/main in case it's model dependent but everything I tried crashes.