| Property | Value |
|---|---|
| Parameter Count | 12.2B |
| Model Type | Text Generation |
| License | Apache 2.0 |
| Tensor Type | BF16 |
| Base Model | Mahou-1.5-mistral-nemo-12B-lorablated |
What is Mistral-Nemo-Prism-12B?
Mistral-Nemo-Prism-12B is an experimental language model that builds upon the Mahou-1.5-mistral-nemo-12B-lorablated architecture. This model represents an interesting iteration in the development process, specifically designed to address the challenge of reducing archaic language and purple prose while maintaining uncensored capabilities.
Implementation Details
The model was developed using ORPO (Optimal Reward Policy Optimization) tuning methodology, utilizing 8x A40 GPUs for 2 epochs. The training process incorporated two specialized datasets: Arkhaios-DPO and Purpura-DPO, carefully selected to achieve the desired language modernization objectives.
- Advanced ORPO tuning implementation
- Specialized training on curated DPO datasets
- Optimized for modern language patterns
- Uncensored output capabilities
Core Capabilities
- Text generation with reduced archaic language
- Conversational AI applications
- Natural language processing
- Text-inference operations
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specific focus on reducing archaic language and purple prose while maintaining uncensored capabilities, achieved through a specialized ORPO tuning process and carefully curated datasets.
Q: What are the recommended use cases?
The model is best suited for applications requiring modern language generation, conversational AI, and scenarios where natural, contemporary language output is essential while maintaining unrestricted expression capabilities.