At the opening speech of the 2024 vivo Developer Conference, vivo announced the launch of its self-developed BlueCore large model matrix.
In this release, vivo will showcase its latest 3 billion parameter BlueCore edge-side large model 3B. It is claimed that this model can rival industry models with 7B to 9B parameters in capabilities such as dialogue writing, summary summarization, and information extraction. Compared to the previous BlueCore 7B model, the new BlueCore 3B has improved performance by 300%, with a 46% reduction in power consumption under balanced mode, a 63% decrease in memory usage, achieving an extreme word output speed of 80 words per second, system power consumption of only 450mA, and memory usage of just 1.4GB.
vivo will also introduce a new self-developed BlueCore voice large model, which can accurately understand natural language semantics, simulate human voices and express emotions, while supporting simultaneous interpretation functions.
Additionally, vivo will upgrade the BlueCore image and multimodal large models. The BlueCore image large model will enhance the integration of Chinese characteristics and oriental aesthetics, officially recognized as one of the most Chinese-context-aware image models in the country, supporting traditional Chinese ink painting styles and incorporating Chinese character creation in image generation.
The BlueCore multimodal large model has upgraded visual perception and understanding capabilities, enabling deeper comprehension of the smartphone screen and providing a smoother, more natural experience in real-time dialogue within video streams.