Xiaomi open sources the Large-Scale Voice Understanding Model MiDashengLM-7B.

Xiaomi's self-developed large-scale sound understanding model MiDashengLM-7B was officially released today and is completely open-sourced. According to Xiaomi officials, MiDashengLM-7B has achieved double breakthroughs in speed and accuracy: the delay of the first token for a single sample is only 1/4 of that of similar models, and the concurrent processing is over 20 times faster under the same memory. It has also set the best performance for large multimodal models on 22 public benchmark datasets.