SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Dec 21, 2023License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Cold

SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE is a 7 billion parameter merged language model built upon the Mistral-7B-v0.1 architecture, utilizing the DARE ties method. It integrates components from Starling-LM-7B-alpha, rwitz/go-bruins-v2, chargoddard/loyal-piano-m7, Undi95/Toppy-M-7B, and NeverSleep/Noromaid-7b-v0.1.1. This model is specifically optimized for engaging roleplay (RP) with strong character card adherence and general intelligence, demonstrating solid performance on logic tests.

Loading preview...

Model Overview

SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE is a 7 billion parameter language model designed for highly engaging roleplay (RP) and intelligent conversational capabilities. It is built on the Mistral-7B-v0.1 base model and created using the DARE ties merging method, combining several specialized models to achieve its unique characteristics.

Key Capabilities & Features

  • Enhanced Roleplay (RP): Optimized for immersive RP experiences with strong adherence to character cards, integrating diverse RP datasets from models like chargoddard/loyal-piano-m7 (PIPPA, rpbuild, LimaRP) and NeverSleep/Noromaid-7b-v0.1.1.
  • General Intelligence: Incorporates Starling-LM-7B-alpha, known for its strong performance in chatbot arenas, and rwitz/go-bruins-v2 (a MetaMath-Cybertron-Starling derivative), contributing to its "smart cookie" capabilities.
  • Creativity: Leverages Undi95/Toppy-M-7B, a model recognized for its creativity and high ranking on OpenRouter among 7B models.
  • Merging Method: Utilizes the DARE ties method with specific weight and density parameters, as detailed in the MergeKit GitHub discussions, to blend the strengths of its constituent models effectively.

Performance & Use Cases

This model has demonstrated strong performance in internal evaluations, ranking highly on personal RP unit test benchmarks and achieving a solid score of 20 on lilblam's LLM Logic Test. It is particularly well-suited for applications requiring:

  • Interactive Storytelling: Generating dynamic and consistent narratives based on character profiles.
  • Character-driven Chatbots: Creating chatbots that maintain distinct personalities and conversational styles.
  • Creative Content Generation: Tasks benefiting from a model with high creativity and adherence to specific prompts.

Prompt Format

The model supports a custom Noromaid template for optimal SillyTavern results, with configuration files provided. Additionally, it is designed to be compatible with the Alpaca prompt format.

Finetunes
5 models