| Property | Value |
|---|---|
| Base Model | LLaMA 3.3 70B |
| Merge Method | SCE (Smooth Controlled Ensemble) |
| Author | zerofata |
| Model Link | Hugging Face |
What is L3.3-GeneticLemonade-Final-70B?
L3.3-GeneticLemonade-Final-70B is an advanced language model created through a sophisticated merging process of multiple high-performing LLaMA-based models. It represents the culmination of the Genetic Lemonade series, specifically designed for roleplay and creative writing applications while maintaining a balance between intelligence and creativity.
Implementation Details
The model employs a two-stage SCE merge method, first creating a strong general-purpose base using diverse multilingual models, then incorporating specific roleplay and creative writing capabilities. The implementation supports various quantization options including static and iMatrix quants, with EXL2 compatibility at 4bpw, 4.5bpw, and 6bpw settings.
- Recommended Temperature: 0.9-1.0
- MinP: 0.03-0.05
- Dry Run Settings: 0.8, 1.75, 4
- Compatible with Llama-3-Instruct-Names format
Core Capabilities
- Enhanced instruction following compared to previous versions
- Reduced censorship while maintaining ethical boundaries
- Balanced creativity and intelligence optimization
- Strong roleplay and creative writing performance
- Multi-model merge architecture for diverse knowledge integration
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its carefully crafted merge process that combines multiple high-performing models while avoiding dataset redundancy. It offers a superior balance between creative writing capabilities and logical reasoning, with reduced censorship compared to its predecessors.
Q: What are the recommended use cases?
The model is primarily designed for roleplay and creative writing applications. It excels in scenarios requiring both creative expression and coherent narrative development, while maintaining improved instruction-following capabilities.