Xiaomi has unveiled its MiDashengLM-7B AI model, which is already deployed in smart home systems and cars in China. The open-source nature of the model allows developers and companies to use it for various projects. Xiaomi hopes to build a strong developer base. The AI model is trained using publicly available data. It is released under the Apache 2.0 license. MiDashengLM-7B outperforms other voice systems in terms of speed and multitasking. The model can understand ambient sounds, background music, and environmental noise due to its unified training strategy. This training utilizes Xiaomi’s Dasheng audio encoder and Alibaba’s Qwen2.5-Omni-7B decoder. The AI is integrated into over 30 smart features across Xiaomi’s product line. It provides 24/7 sound monitoring and alerts for smart homes and supports voice commands and real-time pronunciation feedback in cars.
