Kunoichi 7 B

Introduction

Kunoichi-7B is a general-purpose model noted for its role-playing (RP) capabilities. It is an advanced version of previous models, designed to enhance intelligence and maintain high performance benchmarks. The model supports context windows up to 8k and can experimentally extend to 16k with NTK RoPE alpha of 2.6.

Architecture

Kunoichi-7B is a SLERP (Spherical Linear Interpolation) merger between the Silicon-Maid-7B model and an unreleased model referred to as "Ninja-7B". This architecture aims to preserve the strengths of Silicon-Maid-7B while increasing cognitive capability. The merger was motivated by an attempt to create a viable RP-focused model with enhanced metrics.

Training

The training involved merging existing models to retain RP strengths while boosting intelligence. The benchmarks indicate that Kunoichi-7B performs well in various tests, such as MT Bench and EQ Bench, when compared to peer models like GPT-3.5-Turbo and Claude-2.

Guide: Running Locally

1. Install Required Libraries: Make sure you have Python and the Hugging Face Transformers library installed.
2. Download the Model: Clone the Kunoichi-7B repository from Hugging Face.
3. Configure Environment: Set up your environment to handle a large context window, using a compatible configuration file if needed.
4. Run the Model: Execute the model using a Python script or interactive session, providing input prompts in the required format.
5. Adjust Parameters: Tweak temperature and other parameters to balance creativity and stability in outputs.

For optimal performance, consider using cloud GPUs such as those provided by AWS, Google Cloud, or Azure.

License

Kunoichi-7B is distributed under the Creative Commons Attribution-NonCommercial 4.0 International (cc-by-nc-4.0) license. This permits non-commercial use, sharing, and adaptation with attribution.