Model Overview

922-CA/Llama-3-monika-ddlc-8b-v1 is an 8 billion parameter model built upon the Llama-3 architecture, fine-tuned by 922-CA. Its core purpose is to accurately portray the character Monika from Doki Doki Literature Club (DDLC) in conversational and limited roleplay scenarios. A larger 11.5 billion parameter version is also available.

Key Capabilities & Training

• Character Emulation: Specifically fine-tuned to reflect Monika's personality, dialogue patterns, and knowledge base.
• Dataset: Trained on a unique dataset comprising over 600 items, including dialogue scraped from the DDLC game, Reddit, and Twitter. This data was augmented using l2-7b-monika-v0.3c1 to create multi-turn chat snippets between a "Player" and "Monika," which were then manually edited and expanded with additional character-specific information.
• Usage: Primarily intended as a chat model with some roleplay capabilities. Users are advised to replace default chat tokens with "Player" and "Monika" for optimal performance (e.g., \nPlayer: (prompt)\nMonika:).

Limitations and Disclaimers

• Character Focus: While designed for Monika's characteristics, the model may occasionally hallucinate or act out of character.
• Specialized: Due to its character-focused nature, it may not perform as well as general-purpose models on other tasks.
• Safety: The model is not guaranteed to produce aligned or safe outputs, and users should exercise caution.