Latest Search
Quote
| Back Zoom + Zoom - | ||||
|
XIAOMI-W Open-Sources Large Model for Sound Understanding
Recommend 17 Positive 34 Negative 15 |
||||
|
||||
|
XIAOMI-W (01810.HK) released and fully open-sourced the large model for sound understanding, MiDashengLM-7B. According to the introduction, MiDashengLM-7B achieves dual breakthroughs in speed and accuracy, including the first token delay for a single sample is only one-fourth of similar models, and its efficiency under the same memory is more than 20 times that of industry-leading models. It has set new records for the best SOTA performance on 22 public evaluation sets in multimodal large models. MiDashengLM harnesses the Xiaomi Dasheng audio encoder as its core component, marking a significant upgrade for the Xiaomi Dasheng series models. Based on the current version, Xiaomi has begun further upgrades to the computational efficiency of this model, aiming for offline deployment on terminal devices and enhancing more comprehensive functions such as sound editing based on user natural language prompts. AASTOCKS Financial News Website: www.aastocks.com |
||||
Copied to Clipboard
 |
||||
