Aura 4 B

Introduction

Aura-4B is a cutting-edge roleplaying model designed to provide unique and engaging interactions. It has been fine-tuned with hundreds of millions of tokens from completion, instruction, and roleplaying datasets. The model utilizes a Kahneman-Tversky Optimization for distinctive output styles. Developed by Aura Industries, with contributions from Anthracite Org, it is licensed under Apache 2.0.

Architecture

• Model Name: Aura-4B
• Base Model: IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml
• Model Type: Chat Completions
• Prompt Format: ChatML
• Language: English
• Max Context Length: 8,192+ tokens

Training

The training process involves several configurations designed for completion, instruction, roleplaying, and KTO (Kahneman-Tversky Optimization) settings. The model has been trained with datasets including Mielikki/Erebus-87k and FourOhFour/RP_Phase. The training strategy includes gradient accumulation, cosine learning rate scheduling, and advanced plugins like Liger for optimization.

Guide: Running Locally

To run Aura-4B locally, follow these steps:

1. Environment Setup: Ensure you have Python and the necessary libraries installed.
2. Download the Model: Use the Hugging Face Hub to download the Aura-4B model.
3. Load the Model: Utilize libraries such as transformers to load and run the model.
4. GPU Requirements: For efficient performance, especially with large context lengths, consider using cloud GPUs like those provided by AWS, Google Cloud, or Azure.

License

Aura-4B is distributed under the Apache 2.0 License. For more details, visit Apache License 2.0.