Aryanne/MixSwap: A Merged Model for Role-Play

Aryanne/MixSwap is a 7 billion parameter language model developed by Aryanne, utilizing a custom branch of mergekit for its creation. This model was produced using the task_swapping merge method, building upon Aryanne/Open-StarLake-Swap-7B as its base.

Key Capabilities

• Specialized for Role-Play: The model is specifically designed and optimized for generating detailed and verbose role-play conversations.
• Merged Architecture: It integrates components from three distinct Mistral-based models:
  • cognitivecomputations/dolphin-2.2.1-mistral-7b
  • teknium/Mistral-Trismegistus-7B
  • l3utterfly/mistral-7b-v0.1-layla-v4-chatml
• Custom Merge Configuration: The merge process involved specific diagonal_offset, random_mask, and weight parameters for each contributing model, indicating a fine-tuned approach to combining their strengths.

Good For

• Interactive Storytelling: Ideal for applications requiring dynamic and descriptive conversational outputs in a role-playing context.
• Character Interaction: Excels at generating detailed responses for character-driven narratives.
• Customizable Prompting: Supports a specific prompt format for role-play, including start and end sequences, and an 'In Memory' section for character descriptions and conversation setup.