Search
Xiaomi Unveils Open-Source Large Voice Understanding Model MiDashengLM-7B
2025-08-04 /
Read about 0 minute
Author：小编
Xiaomi has launched and fully open-sourced the extensive voice understanding model, MiDashengLM-7B. This groundbreaking model has established 22 new state-of-the-art (SOTA) benchmarks in the evaluation of multimodal large models, underscoring its exceptional voice comprehension abilities. Notably, its first token delay for single-sample inference (TTFT) is merely one-quarter of that of industry-leading models. Furthermore, under the same video memory constraints, its data throughput efficiency surpasses industry-leading models by more than 20 times. Xiaomi is actively engaged in enhancing the MiDashengLM-7B model to enable offline deployment on terminal devices.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
