Mahou-1.3-mistral-nemo-12B Overview

Mahou-1.3-mistral-nemo-12B is a 12 billion parameter model developed by flammenai, optimized for generating short, conversational messages. It is particularly adept at casual conversation and character roleplay, making it suitable for interactive dialogue applications. The model was fine-tuned using the ORPO method on a Google Colab A100 for one epoch, building upon the Mistral-Nemo architecture.

Key Capabilities

• Conversational AI: Designed to produce natural and engaging short messages in dialogue.
• Character Roleplay: Excels at adopting and maintaining character personas, including specific formatting for actions and speech.
• ChatML Format: Trained to utilize the ChatML format for structured conversations.
• ORPO Fine-tuning: Leverages the ORPO (Odds Ratio Preference Optimization) method for enhanced performance in its target use cases.

Recommended Usage

This model is ideal for applications requiring dynamic and engaging conversational agents or character-driven interactions. For optimal performance, users should configure their inference environment to use ChatML for context templating and enable instruct mode. Specific settings for platforms like SillyTavern are provided, including custom ChatML Instruct and Sampler presets, along with recommended additional stopping strings like "\n", "<|", and "</" to improve response quality.