| Property | Value |
|---|---|
| Parameter Count | 12.2B |
| Model Type | Text Generation |
| Architecture | Mistral-based Merge |
| Tensor Type | BF16 |
| Papers | DARE & TIES |
What is MN-12B-Mag-Mell-R1?
MN-12B-Mag-Mell-R1 is a sophisticated merged language model that combines seven different Mistral-based models using an advanced DARE-TIES methodology. Named after the Celtic Otherworld paradise, this model represents a culmination of "Best of Nemo" capabilities, specifically designed for creative and fictional use cases.
Implementation Details
The model employs a multi-stage SLERP merge approach with DARE-TIES integration, built upon the Mistral-Nemo-Base-2407-chatml foundation. It's structured into three distinct intermediate components: Hero (focusing on RP and trope coverage), Monk (emphasizing intelligence and groundedness), and Deity (specializing in prose and literary flair).
- Optimal performance with Temperature 1.25 and MinP 0.2
- ChatML formatting recommended for best results
- Supports generation up to 10K tokens with stable output
Core Capabilities
- Advanced worldbuilding comparable to legacy adventuring models
- High-quality prose generation with minimal artifacts
- Creative metaphor generation and sophisticated writing style
- Balanced combination of role-play, intelligence, and literary capabilities
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive three-part merge architecture (Hero, Monk, Deity) combined with DARE-TIES methodology creates a uniquely balanced system for creative writing and worldbuilding tasks, surpassing typical capabilities of models in its class.
Q: What are the recommended use cases?
This model excels in creative writing, fictional worldbuilding, role-playing scenarios, and generating sophisticated prose. It's particularly well-suited for narrative development and complex creative tasks requiring both intelligence and artistic flair.