DS-Archive/ds-brew-13b: A Llama 2-based Roleplaying Merge

DS-Archive/ds-brew-13b is a 13 billion parameter language model built upon the Llama 2 architecture. It is a unique creation, resulting from a SLERP (Spherical Linear Interpolation) merge of five distinct Llama 2-based models:

• jondurbin/airoboros-l2-13b-2.2
• elinas/chronos-13b-v2
• NousResearch/Nous-Hermes-Llama2-13b
• lemonilia/limarp-llama2-v2
• PygmalionAI/pygmalion-2-13b

Key Capabilities & Differentiators

This model's primary strength lies in its specialization for roleplaying. By merging models known for their conversational and narrative generation abilities, ds-brew-13b is designed to excel at:

• Character-driven interactions: Generating responses that adhere to specified character personas.
• Scenario-based storytelling: Engaging in dynamic narratives based on user-defined scenarios.
• Interactive chat: Facilitating roleplaying conversations where the model acts as a specific character.

Usage and Limitations

While various prompt formats may work due to its merged nature, the model is explicitly tested with an Alpaca-style instruction format, similar to LIMARP v2, which includes sections for Character's Persona, User's Persona, Scenario, and Input. Users should be aware that the model may exhibit biases present in niche roleplaying communities, in addition to those inherited from its base Llama 2 architecture. It is not intended for factual information retrieval or providing advice.