Gemma‑4‑Gembrain‑31B‑it‑uncensored‑heretic Model Released on HuggingFace

The Gemma‑4‑Gembrain‑31B‑it‑uncensored‑heretic model has been released on HuggingFace in both Safetensors and GGUF formats, providing a 31 B‑parameter uncensored finetune of Gemma‑4. The model was created by llmfan46, with the original finetune author Nimbz, and is designed to boost logical and lateral thinking, increase swipe variety, and improve creative prose. The release includes a KLD of 0.0186 and 13 % refusal rate, indicating a relatively low divergence from the base model and moderate safety filtering. The model is available at https://huggingface.co/llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic and its GGUF counterpart at https://huggingface.co/llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic-GGUF.

Users can also request additional formats such as GPTQ or NVFP4 if needed. The model is intended for applications requiring uncensored, high‑reasoning capabilities while maintaining a manageable refusal rate. The release demonstrates continued interest in fine‑tuning large language models for specialized reasoning tasks. The HuggingFace repository also hosts documentation and example usage scripts for integrating the model into pipelines.