chargen v2

Introduction

CharGen v2 is a model designed for generating character descriptions for role-playing games. It creates character details based on user prompts in a dialogue format, allowing for step-by-step character creation and reducing repetition compared to its predecessor, CharGen v1.

Architecture

CharGen v2 is based on the Mistral 7B model and builds on the Good Robot model to enhance its character generation capabilities without the GPT-style verbosity. The model outputs character details field-by-field, facilitating a more dynamic and editable creation process.

Training

The model was trained on a manually curated dataset comprising over 140,000 character cards, filtered and refined to 16,000 cards through pre-filtering, manual review, and grammar correction. The dataset included content from Chub, Venus, and JanitorAI character cards. The training leveraged a combination of Coedit and Llama2 models for optimal grammar correction.

Guide: Running Locally

1. Setup Environment: Ensure you have Python and necessary libraries installed.
2. Download Model: Obtain CharGen v2 from Hugging Face.
3. Configure GPU: Use a local machine with a compatible GPU or opt for cloud-based GPUs like AWS EC2 or Google Cloud.
4. Run Inference: Use the model with input prompts to generate character descriptions.

Cloud GPUs: Consider using Nvidia's cloud-based solutions for heavy workloads to enhance performance.

License

CharGen v2 is distributed under the Apache 2.0 License, allowing for wide-ranging use and modification with appropriate citation.