A New 7.9B Mixture of Experts Model for Tool Use and Automation
A new 7.9B parameter Mixture of Experts (MoE) model has been introduced, requiring only 1.3B active parameters per token. The model is designed for tool use and automated environment control, and is currently available for free for seven days. This development highlights the trend of creating highly efficient, smaller models that can run locally while performing complex tasks like tool use. By activating only a fraction of its parameters, it significantly reduces computational requirements without sacrificing capability. The model leverages a sparse MoE architecture to optimize inference speed and resource consumption. It focuses specifically on bridging the gap between text generation and action execution through automated environment control.
## BACKGROUND
Mixture of Experts (MoE) is a machine learning technique that divides a model into specialized sub-networks, activating only the relevant 'experts' for a given input to save compute. Tool use, or function calling, is a capability that allows language models to interact with external systems by generating structured instructions for APIs or scripts rather than just generating text.