Hugging Face Overview
MN-12B-Mag-Mell-R1: A Creative Powerhouse
MN-12B-Mag-Mell-R1 is a 12 billion parameter language model developed by inflatebot, specifically engineered for advanced creative and fictional applications. This model is a sophisticated multi-stage merge, inspired by hyper-merges like Tiefighter and Umbral Mind, and uses IntervitensInc/Mistral-Nemo-Base-2407-chatml as its foundation. It leverages a DARE-TIES merge method, combining three "specialist" models, each formed by layer-weighted SLERP merges of two distinct models.
Key Capabilities & Design:
- Multi-stage Merge: Combines six models into three specialists (Hero, Monk, Deity) before a final merge, optimizing for diverse creative strengths.
- Specialized Components:
- Hero: Focuses on roleplay, kink, and trope coverage (from Chronos Gold and Sunrose).
- Monk: Enhances intelligence and groundedness (from Bophades and Wissenschaft).
- Deity: Improves prose quality and expressive flair (from Gutenberg v4 and Magnum 2.5 KTO).
- Worldbuilding & Prose: Demonstrates exceptional worldbuilding capabilities and generates prose with minimal "slop" and frequently electrifying metaphors.
- ChatML Formatting: Recommended for optimal performance, based on its base model.
Recommended Use Cases:
- Fictional & Creative Writing: Ideal for generating stories, narratives, and descriptive text.
- Roleplay Scenarios: Excels in creating dynamic and engaging roleplay experiences.
- Worldbuilding: Particularly strong in developing intricate and imaginative fictional worlds.
Users should experiment with sampler settings, specifically Temp 1.25 and MinP 0.2, adjusting as needed for coherency. The model aims to provide a general-purpose "Best of Nemo" experience for creative tasks.