aifeifei798/llama3-8B-DarkIdol-2.3-Uncensored-32K: Roleplay and Uncensored Text Generation

This model is an 8 billion parameter Llama 3-based language model, specifically designed for uncensored text generation and specialized role-playing scenarios. It is a merge of several pre-trained models, with winglian/Llama-3-8b-64k-PoSE serving as the base, providing a theoretical context length of 64K (tested up to 32K).

Key Capabilities

• Uncensored Responses: Designed to provide unrestricted output, suitable for a wide range of creative and sensitive topics.
• Specialized Roleplay: Excels in various role-playing scenarios, adapting to diverse character types and narrative demands.
• Scholarly and Detailed Output: Capable of generating extensive and detailed responses, akin to a thesis in depth.
• Efficient Quantization: Features custom quantizations (f16 for output/embed tensors, 15_k, q6_k, or q8_0 for others) to minimize degradation while reducing model size, enabling 3-6 t/sec on CPU-only systems.
• Vision Functionality: Supports multimodal capabilities with vision input when used with compatible tools like Koboldcpp and a specified mmproj file (e.g., Nitral-AI/Llama-3-Update-3.0-mmproj-model-f16).
• Quick Response: Optimized for rapid text generation.

Good For

• Creative Writing & Roleplay: Ideal for generating detailed narratives, character interactions, and immersive role-playing experiences without content restrictions.
• Offline Use: Quantized versions are suitable for running on CPU-only devices using tools like llama.cpp, Koboldcpp, LM Studio, and Backyard AI.
• Exploratory Content Generation: Useful for developers and users seeking models that offer broad and uncensored conversational capabilities.