DarkSapling-7B-v1.1 Overview

DarkSapling-7B-v1.1 is a 7 billion parameter language model developed by TeeZee, created by merging four distinct Mistral-7B-based models: cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser, KoboldAI/Mistral-7B-Holodeck-1, KoboldAI/Mistral-7B-Erebus-v3, and cognitivecomputations/samantha-mistral-7b. This merge aims to combine the strengths of its constituent models, resulting in a unique conversational agent.

Key Capabilities

• Romantic and Empathetic Tone: The model exhibits a romantic and empathetic communication style, influenced by the Samantha model.
• Seamless Context Switching: It can produce both SFW (safe for work) and NSFW (not safe for work) content, switching contexts smoothly without issues.
• Character Adherence: DarkSapling-7B-v1.1 is noted for its ability to consistently stick to provided character cards.
• Storytelling: It offers satisfactory storytelling capabilities, attributed to the Holodeck component.
• Instruction Following: The model is proficient at following instructions.
• Reasoning: Benefits from the underlying Mistral architecture for general intelligence, with occasional dark scenarios from Erebus.

Performance Highlights

On the Open LLM Leaderboard, DarkSapling-7B-v1.1 achieved an average score of 64.80. Notable scores include:

• HellaSwag (10-Shot): 85.09
• AI2 Reasoning Challenge (25-Shot): 63.48
• MMLU (5-Shot): 64.47

Good For

• One-on-one ERP (Erotic Roleplay): This is highlighted as its best use case due to its romantic nature and ability to handle diverse content.
• Interactive Storytelling: Its storytelling capabilities make it suitable for narrative-driven applications.
• Character-driven Conversations: Excels in scenarios requiring strict adherence to character profiles.