Gemma 4 31B-it finetuned on Korean construction-safety data with Unsloth Studio, merged, and quantized to Q4_K_M GGUF for llama.cpp / Ollama / LM Studio.

Files

File	Purpose	Size
`gemma-4-31b-it.Q4_K_M.gguf`	Q4_K_M quantized weights	~18 GB
`mmproj-BF16.gguf`	Vision projector (BF16) for multimodal inputs	~1.2 GB

Usage

llama.cpp (text only)

./llama-cli -m gemma-4-31b-it.Q4_K_M.gguf -p "안전모 미착용에 대한 조치 방안은?"

llama.cpp (with vision)

bash

./llama-cli \
  -m gemma-4-31b-it.Q4_K_M.gguf \
  --mmproj mmproj-BF16.gguf \
  --image site_photo.jpg \
  -p "이 현장에서 관찰되는 안전 위험 요소를 나열하세요."

Ollama

bash

ollama create gemma4-safety -f Modelfile
# Modelfile:
#   FROM ./gemma-4-31b-it.Q4_K_M.gguf

swordKoala/Gemma-4-31B-ko-construction-safety-q4

Gemma-4-31B ko-construction-safety (Q4_K_M GGUF)

Files

Usage

llama.cpp (text only)

llama.cpp (with vision)

Ollama

swordKoala/
Gemma-4-31B-ko-construction-safety-q4