DaringMaid-13B Overview
DaringMaid-13B is a 13 billion parameter instruction-tuned language model created by Kooten, built upon the Llama 2 architecture. The primary goal behind its development was to produce a Noromaid variant with improved intelligence and enhanced instruction-following capabilities. This model is a sophisticated merge of several distinct fine-tuned models, each contributing to its overall performance.
Key Merged Components
The model's architecture integrates components from various sources, including:
- sequelbox/DynamicFactor: Utilized as a base to boost general knowledge, precise communication, conceptual understanding, and technical skills.
- NeverSleep/Noromaid-13b-v0.1.1: A core component, building on the Noromaid lineage.
- Undi95/Utopia-13B: Included for its recognized quality and performance.
- tavtav/Rose-20B and CalderaAI/Thorns-l2: Incorporated to add specific characteristics.
- A merge based on Gryphe/MythoMax-L2-13b: Integrated for its strong performance characteristics.
Instruction Following and Use Cases
DaringMaid-13B is specifically designed to excel in scenarios requiring accurate instruction adherence and nuanced communication. Its development focused on creating a model that is "smarter and better at following instructions," making it well-suited for applications such as advanced chatbots, role-playing, and other instruction-driven conversational AI tasks. The model provides specific prompt templates, including an Alpaca-style format, to guide users in achieving optimal responses.