Trying out Nemotron 3.5 Lightning 30B

llama serve \
     -hf ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:Q4_0 \
     -ngl auto \
     -c 262144 -np 1 \
     -ctk q8_0 -ctv q8_0 \
     -fa on --fit on \
     --spec-type draft-mtp \
     --spec-draft-n-max 3 \
     --jinja --kv-unified \
     --temp 1.0 --top-p 0.95 --top-k 0 --min-p 0 \
     --reasoning on --reasoning-format deepseek \
     --no-reasoning-preserve \
     --host 0.0.0.0 --port 8080 \
     --alias nemotron-3.5-lightning