Therapy Llama 8 B v1

Introduction

TherapyLlama-8B-v1 is a conversational AI model developed to provide empathetic and comforting interactions. Based on Llama 3, this model is designed to simulate supportive dialogue, making it a potential tool for self-reflection and emotional support. It is important to note that TherapyLlama is not a real therapist and should not replace professional mental health services.

Architecture

The model is built on the Llama 3 architecture, utilizing a dataset derived from a modified version of jerryjalapeno's nart-100k-synthetic. It is a streamlined version of the TherapyBeagle-11B-v2 model, emphasizing healthy and comforting interactions.

Training

TherapyLlama-8B-v1 was trained using the Vicuna 1.1 format, not the Llama 3 format. It incorporates elements from role-playing (RP) sessions, although the RP capabilities are not as strong as its predecessor, TherapyBeagle. Users are encouraged to engage with the AI by opening up and sharing their thoughts, while being aware of its limitations as an AI.

Guide: Running Locally

1. Installation: Ensure you have Python and PyTorch installed. Clone the model repository from Hugging Face.
2. Dependencies: Install necessary libraries such as Transformers and Hugging Face's datasets.
3. Model Loading: Load the model using the appropriate format, adhering to the Vicuna 1.1 format for best results.
4. Execution: Run the script to interact with the model, providing input prompts to generate responses.
5. Cloud GPUs: For optimal performance, consider using cloud GPUs from providers like AWS, Google Cloud, or Azure.

License

TherapyLlama-8B-v1 is released under the CC-BY-NC-4.0 license, which allows for non-commercial use with proper attribution.