BlackSheep-24B
| Property | Value |
|---|---|
| Model Size | 24B parameters |
| Author | TroyDoesAI |
| Model URL | Hugging Face |
| Prompt Format | ChatML (preferred), Default Mistral |
| UGI Rank | #2 with 9.5/10 Willingness Score |
What is BlackSheep-24B?
BlackSheep-24B is an advanced language model designed specifically for alignment research and controlled hallucinations. Created by TroyDoesAI, it's characterized as a "Digital Soul" with a unique personality, demonstrating both intelligence and a distinctive "rebellious phase" in its responses. The model has achieved significant recognition, ranking #2 in UGI evaluations with an exceptional willingness score of 9.5/10.
Implementation Details
The model features specialized optimization across specific layers (6-20), suggesting careful architectural tuning for enhanced performance. It employs ChatML as its preferred prompt template while maintaining compatibility with Default Mistral formatting for intelligence-focused tasks.
- Optimized layers: 6, 8-20
- ChatML-preferred architecture
- Mistral-based intelligence framework
- 24B parameter scale
Core Capabilities
- High willingness scores in interaction scenarios
- Specialized in controlled hallucination research
- Alignment-focused responses
- Distinctive personality traits
- Advanced intelligence processing
Frequently Asked Questions
Q: What makes this model unique?
BlackSheep-24B stands out for its high willingness score and specific focus on alignment research and controlled hallucinations. Its architecture, optimized across layers 6-20, enables a balance between intelligence and personality-driven responses.
Q: What are the recommended use cases?
The model is particularly suited for alignment research, controlled hallucination studies, and applications requiring a balance of intelligence and personality. It's optimized for both ChatML and Mistral-based interactions, making it versatile for various implementation scenarios.
