MN-12B-Mag-Mell-R1 Uncensored
fully unlocked version of the creative language model
2 min read
This model is a completely uncensored version of the popular creative merge MN-12B-Mag-Mell-R1 (12.25B parameters), built on Mistral-Nemo. The original model combines three specialized directions: prose and style (Deity), plot development and roleplay (Hero), and logic and reasoning (Monk). The new version has been processed using abliteration (removal of restrictions and censorship at the model weights level while preserving cognitive capabilities) via norm-biprojected preservation. It is designed for unrestricted fiction, deep lore, and complex creative tasks without artificial refusals from the assistant.
Key features
- Complete removal of built-in refusals and restrictions (Uncensored) without degrading the model's logical and creative abilities.
- Preservation of the unique literary style and strong skills in crafting vivid metaphors and prose from the original Mag Mell base.
- Use of the norm-biprojected preservation method during abliteration to prevent cognitive degradation.
- High stability in creative scenarios and roleplay thanks to the balanced architecture.
What you can do
- Write a piece of fiction, a fantasy novel, or a script with complex plot twists and worldbuilding.
- Run an interactive text roleplaying game (RP) with deep atmospheric immersion and variable event progression.
- Have a conversation on any open topic without triggering safety filters and template safety refusals.
Example
When using, the ChatML format, a temperature around 1.25, and a MinP parameter of 0.2 are recommended for maximum text expressiveness.
Read more
DetailsBase model: inflatebot/MN-12B-Mag-Mell-R1
Parameter count: 12.25B (transformers library)
Abliteration settings: dimension 29, scale 1.5 across all layers
Setup features: full compatibility with GGUF quantization (including recipes for Q8_K_XL)










Comments