| Property | Value |
|---|---|
| Parameter Count | 70B |
| Base Architecture | LLaMA 3.3 |
| Merge Method | SCE (Select, Calculate, and Erase) |
| Model URL | https://huggingface.co/Steelskull/L3.3-Mokume-Gane-R1-70b-v1.1 |
What is L3.3-Mokume-Gane-R1-70b-v1.1?
L3.3-Mokume-Gane-R1-70b-v1.1 is an advanced language model inspired by the Japanese metalworking technique Mokume-gane. Built on the DS-Hydroblated-R1 foundation, it combines multiple specialized components to create a unique model focused on creative expression while maintaining technical precision. The model is part of a three-model experimental series, representing the creative-focused variant.
Implementation Details
The model employs a sophisticated architecture that integrates multiple components through the SCE merge method. It builds upon the L3.1x3.3-DS-Hydroblated-R1-70B-v4.1 base model and incorporates elements from EVA-LLaMA-3.33, Euryale-v2.3, Cirrus-x1, Hanami-x1, Anubis-v1, and Negative_LLAMA.
- Utilizes SCE merge methodology for component integration
- Implements enhanced reasoning capabilities through structured prompting
- Features specialized sampler settings for optimal performance
- Incorporates bias reduction through Negative_LLAMA integration
Core Capabilities
- Enhanced creative expression and scene comprehension
- Strong character adherence and natural dialogue flow
- Advanced reasoning capabilities with step-by-step thinking patterns
- Balanced response generation with detailed scene descriptions
- Unique output generation differentiating it from standard models
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its ability to generate creative and unexpected outputs while maintaining technical precision, achieved through its unique combination of components and the SCE merge method. It excels particularly in character adherence and creative expression.
Q: What are the recommended use cases?
The model is particularly well-suited for creative writing, character-based interactions, and scenarios requiring both innovative thinking and logical reasoning. It performs best when using structured prompts and appropriate sampler settings (Temperature: 1-1.05, Min P: 0.03).