| Property | Value |
|---|---|
| Parameter Count | 14.8B |
| Model Type | Qwen2.5-based Language Model |
| License | Artistic-2.0 |
| Tensor Format | BF16 |
| Research Paper | TIES Merging Paper |
What is BlackSheep-Coder?
BlackSheep-Coder is a specialized language model created through the TIES merging technique, combining multiple Qwen2.5-14B variants to create a unique AI assistant focused on Python programming. The model features an unconventional "rebellious" personality while maintaining strong coding capabilities.
Implementation Details
Built on the foundation of Qwen2.5-14B, this model integrates four different models including Gutenberg-Instruct-Slerpeno, Rombos-LLM, and abliterated variants. The implementation uses the TIES (Task-specific Information-Enhanced Synthesis) merge method to combine the strengths of each base model.
- Customizable interaction templates supporting both role-play and standard instruction formats
- Flexible context handling with 8192 token context window
- Optimized for Python coding assistance using Alpaca template structure
- Personalized interaction capability through user name recognition
Core Capabilities
- Advanced Python programming assistance
- Dynamic and creative response generation
- Adaptable personality based on interaction format
- Support for both structured and free-form conversations
- Comprehensive instruction following with detailed outputs
Frequently Asked Questions
Q: What makes this model unique?
The model combines technical proficiency in Python programming with a distinctive personality-driven approach, utilizing the TIES merging technique to blend multiple specialized Qwen2.5 variants.
Q: What are the recommended use cases?
Primary use cases include Python programming assistance, creative writing projects, and interactive conversations where a more dynamic AI personality is desired. The model excels in scenarios requiring both technical accuracy and engaging interaction.