EXTENSOR

Overview

EXTENSOR is an inference software package purpose-built for the AMD Ryzen™ AI Max platform that enables deployment of large language models and mixture-of-experts (MoE) models on local AI systems, bringing frontier-class AI capabilities to a broader range of developer, workstation, and edge form factors.

The innovation is captured in the name: EXTENSOR keeps each model's mixture-of-experts weights on SSD and uses adaptive, predictive prefetching to stream the right experts into memory just before they're needed. This approach significantly reduces memory requirements while maintaining high-quality model performance, enabling larger and more capable AI models to run on a single Ryzen AI Max system.