Aryanne/Open-StarLake-Swap-7B Overview

Aryanne/Open-StarLake-Swap-7B is a 7 billion parameter language model developed by Aryanne, utilizing a unique task_swapping merge method via a custom branch of mergekit. It is built on the senseable/WestLake-7B-v2 as its foundational base model.

Key Capabilities & Merge Details

This model integrates the strengths of three distinct pre-trained language models:

• berkeley-nest/Starling-LM-7B-alpha
• NousResearch/Nous-Hermes-2-Mistral-7B-DPO
• openchat/openchat-3.5-0106

The merge configuration involved specific layer ranges and weighting parameters for each contributing model, aiming to combine their respective proficiencies. The model has a context length of 4096 tokens.

Intended Use Case

Open-StarLake-Swap-7B is specifically optimized for verbose and descriptive role-play conversations. The developer provides a recommended prompt format, including start and end sequences, to facilitate engaging and detailed interactive storytelling, particularly within environments like Koboldcpp. This focus on rich, character-driven dialogue distinguishes it from more general-purpose instruction-tuned models.