Xiaomi has introduced a new AI model, MiDashengLM-7B, signaling its foray into the competitive AI landscape. This isn’t a mere lab experiment; the model is already operational in real-world devices such as smart home systems and vehicles within China. Xiaomi’s new AI is available as open source, opening up possibilities for developers and commercial projects. The company’s aim is to build a broad developer base, recognizing that a robust developer ecosystem could be key to success in the current AI race. MiDashengLM-7B is designed to perform exceptionally well in terms of speed and multitasking compared to other voice systems. The model’s capabilities extend to understanding ambient sounds, background music, and environmental noise within a single system. This is a result of a unified training strategy, utilizing Xiaomi’s Dasheng audio encoder with the Qwen2.5-Omni-7B decoder, initially developed by Alibaba. According to Xiaomi, the model currently powers over 30 smart features across its product range, including 24/7 sound monitoring and alerts for unusual sounds in smart homes, and voice command support with real-time pronunciation feedback in cars.
