Gemma 4 12B Uncensored
HauhauCS’s Balanced uncensored build of Google’s QAT Gemma 4 12B: a 6.9 GB GGUF with reduced refusals for 8–12 GB GPUs and 16 GB Macs. A community-modified checkpoint, not a Google release.
What do you want to do?
Choose a task. Get the software, files, settings, and steps to run it.
Open a guide for its software, hardware requirements, and setup steps.
At a glance
- Parameters
- 12B
- Architecture
- Dense multimodal transformer · community-modified QAT checkpoint
- Context length
- 256,000 native (Google base model) · the agent guide starts at 65,536
- License
- Gemma terms, as tagged on the repository · see the linked page
- Software
- llama.cpp, LM Studio, Jan, koboldcpp, Ollama
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Run a local Hermes agent that reads and writes files with your approval.
- Private chat with fewer refusals on an 8–12 GB GPU or a 16 GB Mac.
- Ask about images by loading the vision file alongside the model.
Tradeoffs
- “0 refusals on 465 prompts” is HauhauCS’s own result; a finite test does not guarantee every request is answered.
- Fewer refusals do not make answers more accurate. Check important facts yourself.
- The bundled MTP speculative drafter was only tested by the publisher in llama.cpp; a user reports LM Studio leaves it disabled.
- Not an official Google release. Google’s Gemma terms still apply.
Ways to run it
llama.cpp · Windows, macOS, Linux
Used by the Hermes agent guide on this page.
llama-server -hf HauhauCS/Gemma4-12B-QAT-Uncensored-HauhauCS-Balanced:Q4_K_M --jinja -c 65536Ollama · Windows, macOS, Linux
Listed on the model’s Hugging Face page; not used by the guide here.
ollama run hf.co/HauhauCS/Gemma4-12B-QAT-Uncensored-HauhauCS-Balanced:Q4_K_MLM Studio / Jan / koboldcpp · Windows, macOS, Linux
HauhauCS lists these GGUF apps as compatible.
Official instructionsWhat to expect
The pick for 8–12 GB of memory. Start with the Q4_K_M file in llama.cpp; a 12 GB GPU or a 16 GB Mac runs it with room for an agent-sized context.
Explore its uses
Sources and files
Primary sources and files behind this model page.
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
An open-weight model for reasoning, coding, agents, and image understanding. DeepSeek V4.1 Flash combines a 552B backbone with sparse activation and up to one million tokens of context. Choose the original checkpoint or a community uncensored variant below.
Qwen3.8
Explore 4 versions of Qwen3.8. Compare model sizes, packages, hardware requirements, and available setups.