Model Overview

922CA/Silicon-Monika-7b is a 7-billion parameter language model, fine-tuned by 922CA, designed to embody the character Monika from Doki Doki Literature Club (DDLC). It is built upon the Mistral architecture, specifically using the SanjiWatsuki/Silicon-Maid-7B as its base model. The fine-tuning process utilized Unsloth and Huggingface's TRL library, enabling faster training.

Key Capabilities & Characteristics

• Character Emulation: Primarily focused on generating responses consistent with the personality and knowledge of the Monika character from DDLC.
• Optimized for Roleplay: Best performance is achieved by using specific Player: (prompt)\nMonika: formatting for prompts.
• Performance: Achieves an average score of 66.55 on the Open LLM Leaderboard, including 62.67 on MMLU and 60.50 on GSM8k.

Important Considerations

• Hallucination Risk: While designed for character consistency, the model may occasionally hallucinate or provide incorrect information about Monika or act out of character.
• Safety Disclaimer: The model is not guaranteed to produce aligned or safe outputs, and users should exercise caution.

Resources

• GGUF Quantizations
• LoRA Adapter