SakanaAI/Llama-3-8B-Instruct-CycleQD-CS

Model Overview

SakanaAI/Llama-3-8B-Instruct-CycleQD-CS is an 8 billion parameter agentic language model developed by Sakana AI, built upon the Llama-3-8B-Instruct architecture. This model is distinguished by its creation using the CycleQD method, which combines the capabilities of several specialized expert models:

• SakanaAI/Llama-3-8B-Instruct-DB-Expert
• SakanaAI/Llama-3-8B-Instruct-OS-Expert
• SakanaAI/Llama-3-8B-Instruct-Coding-Expert

This integration aims to enhance the model's agentic skills across diverse domains. The development process and detailed methodology are described in the associated paper.

Key Characteristics

• Agentic Capabilities: Designed to acquire and demonstrate agent-like skills, particularly in database interaction, operating system commands, and coding.
• Modular Training: Leverages a novel CycleQD method to combine domain-specific expert models.
• Research-Oriented: Primarily intended for research and development purposes, serving as an experimental prototype.

Training Data

The model's training incorporated several datasets, including:

• Agent-FLAN
• Magicoder-Evol-Instruct-110K
• Magicoder-OSS-Instruct-75K

Intended Use

This model is provided for research and development only. It is not intended for commercial use or deployment in mission-critical environments, and its performance is not guaranteed. Users should be aware of the experimental nature of this prototype.