Aryanne/WestSenzu-Swap-7B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Mar 1, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Aryanne/WestSenzu-Swap-7B is a 7 billion parameter experimental merged language model, created using a task_swapping method with NeuralNovel/Senzu-7B-v0.1-DPO as the base and senseable/WestLake-7B-v2. This model is primarily optimized for role-playing scenarios, leveraging the characteristics of its merged components. It achieves an average score of 67.28 on the Open LLM Leaderboard, with notable performance in HellaSwag and Winogrande.

Loading preview...

Model Overview

Aryanne/WestSenzu-Swap-7B is an experimental 7 billion parameter language model developed by Aryanne, created through a novel task_swapping merge method. This model integrates the strengths of two distinct pre-trained models: NeuralNovel/Senzu-7B-v0.1-DPO as the base, and senseable/WestLake-7B-v2.

Key Capabilities

  • Role-play Optimization: Designed with a focus on enhancing performance in role-playing applications, leveraging the combined characteristics of its constituent models.
  • Merged Architecture: Utilizes a unique task_swapping merge method, allowing for experimental blending of model capabilities.

Performance Highlights

Evaluated on the Open LLM Leaderboard, WestSenzu-Swap-7B demonstrates competitive performance for its size:

  • Average Score: 67.28
  • HellaSwag (10-Shot): 85.70
  • Winogrande (5-shot): 82.48
  • MMLU (5-Shot): 64.14

Ideal Use Cases

This model is particularly well-suited for:

  • Creative Role-playing: Its experimental merge design aims to provide robust and engaging responses for interactive narrative and character-driven applications.
  • Exploratory Merging Research: Developers interested in the task_swapping merge method and its effects on model behavior may find this model a valuable case study.