/home/gochu/workspace/repos/llama.cpp/build/bin/llama-server \
  -m /home/gochu/workspace/models/gguf/gemma/qat/gemma-4-31B-it-qat-NVFP4-Blackwell.gguf \
  -md /home/gochu/workspace/models/gguf/gemma/draft/gemma-4-31B-it-qat-q4_0-unquantized-assistant-q8_0.gguf \
  --spec-type draft-mtp \
  --spec-draft-n-max 3 \
  --host 0.0.0.0 \
  --port 8080 \
  -c 96000 \
  -ngl 99 \
  -sm tensor \
  -b 2048 \
  -ub 128 \
  -fa on \
  -np 1 \
  --jinja \
  --chat-template-file /home/gochu/workspace/models/hf/chat-templates/gemma-4-31B-it-chat_template.jinja
  # --mmproj /home/gochu/workspace/models/gguf/gemma/mmproj/gemma-4-31B-it-mmproj.gguf \
  # --no-mmproj-offload