Arm and Alibaba Partner to Enhance On-Device Multimodal AI with KleidiAI

Arm Holdings plc (Nasdaq: ARM), hereinafter referred to as "Arm," today announced a new collaboration with Alibaba’s Taotian Group on its lightweight deep learning framework, MNN. Through the integration of Arm KleidiAI, the two companies have successfully enabled multimodal artificial intelligence (AI) workloads to run on Arm CPU-powered mobile devices using Alibaba’s instruction-tuned Qwen2-VL-2B-Instruct model. This version of the Qwen multimodal model is specifically designed for on-device image understanding, text-to-image reasoning, and multimodal generation across multiple languages. This collaboration significantly enhances the performance of multimodal AI workloads at the edge, delivering an entirely new user experience.

Stefan Rosinger, Senior Director of Product Management at Arm’s Client Business, stated: “We are in the midst of an AI revolution, witnessing the rise of multimodal AI models that can process and understand diverse data types, including text, images, audio, video, and sensor data. However, deploying these advanced multimodal models on edge devices presents significant challenges due to hardware constraints such as power consumption and memory limitations, as well as the complexities of handling multiple data modalities simultaneously.”

Arm Kleidi provides an ideal solution to address these challenges, delivering seamless performance optimization for AI inference workloads running on Arm CPUs. KleidiAI is a lightweight, high-performance, open-source collection of Arm routines designed specifically for AI acceleration. It has been integrated into the latest versions of major on-device AI frameworks, including ExecuTorch, Llama.cpp, LiteRT (via XNNPACK), and MediaPipe. This enables millions of developers to achieve significant AI performance gains automatically—without requiring any additional effort.