gemma-4-12b-it-abliterated-GGUF

gemma-4-12b-it-abliterated-GGUF (Ollama / llama.cpp)

Abliterated Gemma 4 12B for Ollama / llama.cpp (CPU, CUDA, or Intel Arc via Vulkan/SYCL). Built from Google's official QAT q4_0. Refusals removed (base 3/3 → 0/3).

• gemma-4-12b-it-abliterated-Q4_0.gguf — text model, ~7 GB
• mmproj-gemma-4-12b-it-abliterated.gguf — vision projector (~175 MB), enables image input
• Audio is NOT available in Ollama/llama.cpp — that's a runtime limitation, not the model. Audio only works in transformers/vLLM (see the -nf4 / bf16 repos).

Run in Ollama

hf download smashingtags/gemma-4-12b-it-abliterated-GGUF --local-dir ./g4ab
cd ./g4ab
ollama create gemma4-ablit -f Modelfile
ollama run gemma4-ablit "your prompt"          # text
ollama run gemma4-ablit "describe this ./pic.jpg"   # text + image

The included Modelfile references both the text GGUF and the mmproj so vision works. Gemma 4 vision is new in llama.cpp — make sure your Ollama is current (ollama -v, update if old).