PetrolOrca: A Roleplay-Optimized Language Model

PetrolOrca is a specialized language model built upon the Mistral-7B-OpenOrca foundation, further refined by applying the PetrolLoRA. This combination targets enhanced performance in roleplay and creative text generation scenarios.

Key Capabilities

• Roleplay Generation: Designed to produce engaging and immersive roleplay dialogues and narratives.
• Descriptive Output: Encourages detailed and vivid descriptions of characters, emotions, and environments.
• Customizable Prompting: Utilizes a specific prompt format that includes style, characters, and summary fields, allowing for structured scenario setup.

Training Data

The model was fine-tuned on a curated dataset of approximately 2800 samples, comprising:

• AICG Logs (~34%)
• PygmalionAI/PIPPA (~33%)
• Squish42/bluemoon-fandom-1-1-rp-cleaned (~29%)
• OpenLeecher/Teatime (~4%)

These samples were processed and formatted using GPT-4 or GPT-3.5-turbo-16k to fit the model's prompt structure.

Integration & Usage

PetrolOrca is compatible with popular text generation UIs like Text Generation Web UI and SillyTavern UI. Users can optimize output quality by experimenting with specific 'Last Output Sequence' prompts, such as those designed to encourage multi-paragraph, descriptive responses.

Good For

• Interactive storytelling and roleplaying applications.
• Generating rich, character-driven narratives.
• Creative writing tasks requiring immersive descriptions.