PetrolDolphin: A Roleplay-Optimized Language Model

PetrolDolphin is a specialized language model developed by Norquinal, built upon the Dolphin-2.0-Mistral-7B base model. Its core enhancement comes from the integration of the PetrolLoRA adapter, which has been fine-tuned on a unique dataset designed for rich, interactive storytelling.

Key Capabilities

• Enhanced Roleplay Generation: Specifically trained to excel in generating detailed and immersive character interactions.
• Diverse Training Data: Utilizes a blended dataset including AICG Logs, PygmalionAI/PIPPA, Squish42/bluemoon-fandom-1-1-rp-cleaned, and OpenLeecher/Teatime, totaling 2800 samples.
• GPT-4/GPT-3.5-turbo-16k Back-filling: Training data was processed and formatted using advanced GPT models to ensure high-quality prompt-response pairs.
• Specific Prompt Format: Designed to work with a structured prompt format that includes style, characters, summary, and chat_history for consistent roleplay scenarios.

Good For

• Creative Writing Applications: Ideal for generating narrative content, character dialogues, and interactive stories.
• Roleplaying Interfaces: Optimized for use with tools like Text Generation Web UI and SillyTavern UI, with specific recommendations for Last Output Sequence to guide generation.
• Descriptive and Immersive Text: Excels at producing multi-paragraph, vivid descriptions of character actions, emotions, and environments, as demonstrated by suggested system notes for better output control.