started · updated
AMD partners with Unsloth to support XDNA2 NPU architecture
AMD has partnered with Unsloth to integrate its XDNA2 NPU architecture with Unsloth Desktop, marking the first NPU platform to support this capability. The integration utilizes Lemonade and FastFlowLM to enable Large Language Models (LLMs) to run locally on the NPU.
This partnership allows for efficient agentic AI inference, offloading workloads from the CPU and GPU to preserve battery life and system resources. The ecosystem supports over 100 AI agents and workflows via the Unsloth API, including tools such as Hermes Agent, OpenClaw, Open WebUI, Cline, LangChain, and Visual Studio Code.
AMD demonstrated the technology using a Qwen3.6 35B-A3B model, noting that while the NPU handles inference, specific memory requirements apply. The architecture is designed for power-efficient local AI, providing a low-power execution option for developers and users working on Windows 11 or Linux x64 systems.
Entities
AMD · Unsloth · Unsloth Desktop · Windows 11 · XDNA2