kukulemon-7B: Merged for Reasoning and Roleplay

This 7 billion parameter model, grimjim/kukulemon-7B, is a unique merge designed to combine robust reasoning with strong roleplaying capabilities. Created by grimjim using the SLERP merge method, it integrates two distinct base models:

Key Capabilities

• Enhanced Reasoning: Built upon two Kunoichi models known for their reasoning prowess, aiming for a "dense" encoding of logical thought.
• Roleplay Optimization: Incorporates KatyTheCutie/LemonadeRP-4.5.3, specifically targeting improved performance in roleplaying scenarios.
• Flexible Prompting: Supports Alpaca format prompts and has been tested effectively with ChatML using specific temperature and minP settings.

Technical Details

• Merge Method: Utilizes the SLERP (Spherical Linear Interpolation) merge method for combining model weights.
• Context Length: While the model claims a 32K context window, informal testing suggests optimal coherence is maintained up to 8K tokens.
• Quantization Options: Available in various quantized formats, including GGUF-IQ-Imatrix, 8.0bpw h8 exl2, and Q8_0 GGUF.

Ideal Use Cases

This model is particularly well-suited for applications requiring a balance of logical understanding and creative, engaging conversational outputs, such as advanced chatbots, interactive storytelling, and complex roleplaying simulations where both coherence and imaginative responses are crucial.