Xiaomi’s new AI model, MiDashengLM-7B, is making its debut, and it is already being used in practical applications like smart home systems and automobiles in China. The company has chosen to make this new AI open-source, enabling its use by developers and for commercial projects. Xiaomi aims to establish a large developer community. The AI model, trained on publicly available data, is released under the Apache 2.0 license. The AI model, MiDashengLM-7B, has been designed to function faster and handle multiple tasks better than existing voice systems. It also understands various sounds, including ambient and environmental noises, due to a comprehensive training method. The model combines Xiaomi’s Dasheng audio encoder with the Qwen2.5-Omni-7B decoder from Alibaba. The AI powers over 30 smart features, including security functions in smart homes and voice command capabilities with real-time pronunciation feedback in cars.
