antiven0m/reverie-7b

antiven0m/reverie-7b: A Merged 7B Model for Creative and Explicit Interactions

Reverie-7b is a 7 billion parameter language model developed by antiven0m, building upon the 'finch' model merge. It aims to deliver a more coherent, intelligent, and uncensored conversational experience, often requiring a few generations to achieve its full potential.

Key Capabilities & Merged Components

This model is a merge of three distinct 7B models, utilizing the Model Stock method described in a recent research paper:

• AetherResearch/Cerebrum-1.0-7b: Contributes strong reasoning skills, making Reverie-7b one of the smarter 7B models available.
• macadeliccc/WestLake-7B-v2-laser-truthy-dpo: Enhances the model's creativity and verbosity, allowing for more elaborate and imaginative outputs.
• SanjiWatsuki/Kunoichi-DPO-v2-7B: Provides exceptional capabilities in explicit roleplay (ERP), making the model particularly 'spicy' and uncensored.

Use Cases & Strengths

Reverie-7b is particularly well-suited for:

• Creative Writing and Storytelling: Leveraging its verbose and creative components.
• Explicit Roleplay (ERP): Excelling in scenarios requiring uninhibited and detailed responses.
• Conversational AI: Offering coherent and intelligent interactions with a distinct personality.
• Reasoning Tasks: Benefiting from the Cerebrum-1.0 component for improved logical processing.

With a context length of 8192 tokens, Reverie-7b is configured with specific parameters like a temperature of 1.15 and support for Alpaca and Alpaca-Roleplay presets in instruct mode, optimizing its unique blend of capabilities.