---
base_model: Carlosian/Gemma-4-12b-it-Abliterated
base_model_relation: quantized
library_name: gguf
pipeline_tag: text-generation
tags:
- gguf
- quantized
- abliterated
- uncensored
- llama.cpp
license: other
---

# Gemma-4-12b-it-Abliterated — GGUF quant ladder

Quantized GGUF builds of [`Carlosian/Gemma-4-12b-it-Abliterated`](https://huggingface.co/Carlosian/Gemma-4-12b-it-Abliterated) for llama.cpp / Ollama / LM Studio.

| Quant | Typical use |
|-------|-------------|
| Q2_K | smallest / extreme low VRAM |
| Q3_K_S / Q3_K_M | low VRAM |
| Q4_0 / Q4_K_S / Q4_K_M | **recommended default (Q4_K_M)** |
| Q5_K_S / Q5_K_M | higher quality |
| Q6_K | near-lossless |
| Q8_0 | highest quality quant |
| F16 | full intermediate (large) |

Built with llama.cpp on Thunder Compute (A100). Generated 2026-07-20.

## Acknowledgments & method references

The refusal-removal method used to build the base checkpoint draws on:

- Herring, S., Naviasky, J., Malhotra, K. (2026). *Targeted Neuron Modulation via Contrastive Pair Search.* Nous Research. https://huggingface.co/papers/2605.12290
- Nous Research (2026). *llm-abliteration* — toolkit for abliteration (norm-preserving & biprojected variants, multi-architecture). github.com/NousResearch/llm-abliteration.