NemoMix-Unleashed-12B
| Property | Value |
|---|---|
| Parameter Count | 12.2B |
| Model Type | Text Generation |
| Architecture | Mistral-based Merged Model |
| Tensor Type | BF16 |
What is NemoMix-Unleashed-12B?
NemoMix-Unleashed-12B is an advanced language model created through a sophisticated merge of multiple Mistral-based models using mergekit's della_linear method. It represents a careful balance of different model weights and densities, specifically designed to excel in roleplay and storytelling applications while maintaining reduced repetition patterns in higher context scenarios.
Implementation Details
The model utilizes a complex merging strategy incorporating five distinct models with carefully calibrated weights and densities ranging from 0.1 to 0.33. It employs the Mistral Instruct format and is optimized for both GGUF and EXL2 implementations.
- Della_linear merge method with epsilon 0.05 and lambda 1
- BFloat16 precision for optimal performance
- Customized density distribution (0.4-0.8) across merged models
- Available in multiple formats including GGUF and EXL2
Core Capabilities
- Enhanced context handling with reduced repetition
- Optimized for roleplay and creative writing
- Flexible temperature settings (1.0-1.25 recommended)
- Supports various inference configurations
- Compatible with text-generation-inference endpoints
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its carefully balanced merge of multiple high-quality Mistral-based models, resulting in improved context handling and reduced repetition, particularly in higher-context scenarios.
Q: What are the recommended use cases?
The model excels in roleplay and storytelling applications, with recommended temperature settings of 1.0-1.25 with 0.1 Top A or 0.01-0.1 Min P, and 0.8/1.75/2/0 DRY settings for optimal performance.
