zerofata/MS3.2-PaintedFantasy-v4.1-24B

Overview of MS3.2-PaintedFantasy-v4.1-24B

MS3.2-PaintedFantasy-v4.1-24B is a 24 billion parameter uncensored language model developed by zerofata, specifically designed for creative character-driven roleplay (RP) and erotic roleplay (ERP). This iteration, v4.1, focuses on addressing and reducing repetition in generated assistant messages, a common challenge in conversational AI. The model leverages a 32768 token context length, allowing for extended and coherent narrative generation.

Key Capabilities

• Enhanced Creative Roleplay: Optimized for generating dynamic and engaging character interactions in both SFW and NSFW contexts.
• Reduced Repetition: Incorporates advanced filtering and rewriting techniques during training to minimize repetitive outputs, improving narrative flow.
• Robust Training Methodology: Built using a Supervised Fine-Tuning (SFT) phase on approximately 25 million tokens (17.5 million trainable) followed by Direct Preference Optimization (DPO). The DPO phase expanded to include non-creative datasets like cybersecurity and general assistant/chat preferences to stabilize the model's overall logic.
• Flexible Output Formats: Recommended for use with plaintext actions, "quoted dialogue," and asterisked thoughts for structured roleplay.

Good For

• Developers and enthusiasts seeking a specialized model for high-quality, creative, and uncensored roleplay scenarios.
• Applications requiring long-form narrative generation with a focus on character depth and varied responses.
• Experimentation with advanced fine-tuning techniques aimed at mitigating common LLM issues like repetition in conversational outputs.