Overview

MarinaraSpaghetti/NemoRemix-12B is a 12 billion parameter merged language model, developed by MarinaraSpaghetti, focusing on creating a smart and universal roleplaying experience. It was constructed using the della_linear merge method, combining several NeMo-based models such as Mistral-Nemo-Instruct-2407, Gryphe_Pantheon-RP-1.5-12b-Nemo, shuttle-2.5-mini, MN-12B-Lyra-v1, and magnum-12b-v2. The primary goal behind this merge was to achieve stability and enhanced performance, particularly at higher context lengths, making it suitable for extended roleplaying scenarios.

Key Capabilities

• Universal Roleplaying: Designed to be a versatile model for various roleplaying applications.
• High Context Stability: Aims for stable performance even with context lengths up to 32768 tokens.
• ChatML Format Support: Utilizes the ChatML instruction format, with theoretical compatibility for Mistral Instruct.

Good For

• Character-driven Interactions: Excels in scenarios requiring detailed and consistent character roleplay.
• Extended Conversational Contexts: Suitable for long-form interactions where maintaining context over many turns is crucial.
• Customizable Inference: Recommended inference parameters include Temperature 1.0-1.2 with 0.1 Top A or 0.01-0.1 Min P, and 0.8/1.75/2/0 DRY, offering flexibility for different output styles.