G2 9 B Aletheia v1

Introduction

GEMMA-2-9B Aletheia V1 is a merged model designed to enhance storytelling and role-playing capabilities. It combines the novel-like writing style of Sugarquill with the steerability and role-playing performance of Sunfall, aiming to produce a model with a fresh writing style and effective system prompt adherence.

Architecture

The model is a combination of three base models: UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3, crestf411/gemma2-9B-sunfall-v0.5.2, and allura-org/G2-9B-Sugarquill-v0. It utilizes the transformers library and incorporates Mergekit to blend these models effectively. The configuration applies the Sunfall LoRA on top of Gemma-2-9B, maintaining the intelligence of Gemma while steering the model towards enhanced role-playing capabilities.

Training

The model was trained using a specific YAML configuration with defined weights and densities for each contributing model. The merge method employed is "ties," ensuring that the model maintains a balance between the writing style and role-playing performance. The model is formatted to respond to Gemma instruct formatting, allowing for structured user interaction.

Guide: Running Locally

1. Install Dependencies: Ensure you have Python and the transformers library installed.
2. Download the Model: Use the Hugging Face Model Hub to download GEMMA-2-9B Aletheia V1.
3. Load the Model: Use the transformers library to load the model in your script.
4. Run Inference: Input your prompts following the specified Gemma instruct format.
5. GPU Acceleration: For optimal performance, consider using cloud GPUs such as AWS EC2 P3 instances or Google Cloud's AI Platform.

License

The model is distributed under the Gemma license, which should be reviewed for compliance before use.