| Property | Value |
|---|---|
| Parameter Count | 3 Billion |
| Model Type | LLAMA-based Language Model |
| Architecture | LLAMA-3-Instruct |
| Author | SicariusSicariiStuff |
| Hugging Face URL | https://huggingface.co/SicariusSicariiStuff/Fiendish_LLAMA_3B |
What is Fiendish_LLAMA_3B?
Fiendish_LLAMA_3B is a specialized variant of the LLAMA architecture designed specifically for roleplay and creative writing tasks. It's characterized as the "naughty sister" to Impish_LLAMA_3B, offering edgier content while maintaining strong formatting and instruction-following capabilities. The model features a medium censorship level (4.5/10) and is optimized for short, concise responses in the style of Character.AI.
Implementation Details
The model is available in multiple quantizations including FP16, GGUF, iMatrix, EXL2, GPTQ, and specialized mobile ARM versions. It utilizes the LLAMA-3-Instruct template format and includes specific optimization for roleplay scenarios with custom temperature and sampling parameters.
- Optimized response length: 1-2 paragraphs
- Multiple quantization options for different deployment scenarios
- Customized instruction template system
- Specialized character card support
Core Capabilities
- Role-Play and Creative Writing
- General Task Assistance
- Character-based Interactions
- Low Refusal Rate for Instructions
- Consistent Formatting and Response Structure
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its optimization for roleplay scenarios with specific character formats and low refusal rates, combined with consistent short-form responses and strong adherence to character cards.
Q: What are the recommended use cases?
The model excels in roleplay scenarios, creative writing, and general task assistance. It's particularly well-suited for character-based interactions and situations requiring consistent formatting with concise responses.