Kaiju 11 B G G U F

Introduction

Kaiju-11B-GGUF is a quantized model derived from the original Himitsui/Kaiju-11B model, using llama.cpp. This model aims to reduce common biases and 'GPT-isms' found in other models. It leverages a mix of popular models to enhance performance and applicability across various use cases.

Architecture

The model architecture is based on a combination of several models:

• Fimbulvetr-11B-v2-Test-14
• KuroMitsu-11B
• Fimbulvetr-10.7B-v1
• SOLAR-10.7B-Instruct-v1.0-uncensored
• Solstice-11B-v1

These models are integrated using Gryphe's MergeMonster, adjusting probabilities to reduce unwanted biases and improve model robustness.

Training

The training process involved merging multiple models to address specific issues such as positivity bias and verbosity that occur in role-play scenarios. By carefully adjusting the weightings and probabilities, the model aims to provide more nuanced and accurate responses.

Guide: Running Locally

To run Kaiju-11B-GGUF locally:

1. Clone the repository from Hugging Face.
2. Ensure you have the necessary environment set up for running the model, including Python and required libraries.
3. Download the model files using the provided download link.
4. Load the model using a compatible inference framework, such as llama.cpp.

For optimal performance, especially for intensive tasks, using cloud GPUs from providers like AWS, Google Cloud, or Azure is recommended.

License

Kaiju-11B-GGUF is licensed under the Creative Commons Attribution-NonCommercial 4.0 International (cc-by-nc-4.0). This allows for personal and non-commercial use with appropriate attribution.