During the highly anticipated opening ceremony of the 2024 vivo Developer Conference, the company officially unveiled its latest self-developed technological innovation—the Blue Heart Large Model Matrix. This groundbreaking initiative marks a significant advancement for vivo in the artificial intelligence sector, comprehensively enhancing the capabilities of both language and edge-side large models, and introducing self-developed voice, image, and multimodal large models.
The Blue Heart Large Model Matrix encompasses various dimensions, including language, edge-side, voice, image, and multimodal aspects. The newly introduced 3B Blue Heart Edge-Side Large Model, featuring 3 billion parameters, demonstrates exceptional performance in dialogue generation, summarization, and information extraction, rivaling industry models with 7B-9B parameters. Compared to the Blue Heart 7B, the 3B model achieves a 300% performance improvement while optimizing power consumption and memory usage by 46% and 63%, respectively. The 3B model reaches an impressive word generation speed of 80 words per second, operates at a system power consumption of just 450mA, and reduces memory usage to 1.4GB. In comprehensive evaluations using SuperCLUE and Equal Eval, the Blue Heart 3B topped the rankings and successfully obtained Thiel Laboratory’s comprehensive capability certification for edge-side large models, as well as AIIA’s security prevention capability certification.
In addition to the significant advancements in edge-side large models, vivo also introduced its self-developed Blue Heart Voice Large Model. This model precisely interprets natural language semantics, simulates human voice with emotional expression, and supports simultaneous interpretation, providing users with a more convenient and efficient voice interaction experience.
Vivo has also conducted comprehensive upgrades in the image and multimodal large model domains. The Blue Heart Image Large Model particularly enhances the integration of Chinese characteristics and Eastern aesthetics in its generation capabilities, earning recognition as one of the most contextually aware image models in Chinese settings. Unlike other large models, the Blue Heart Image Large Model supports traditional Chinese ink painting styles and allows the incorporation of Chinese character-related creative elements during image generation. The Blue Heart Multimodal Large Model has been optimized for visual perception and understanding, enabling deeper comprehension of smartphone screen content and delivering smoother and more natural performance in real-time video stream conversations.
The Blue Heart Voice, Image, and Multimodal Large Models have ranked prominently across various professional listings, fully showcasing vivo’s profound expertise and innovative capabilities in the AI technology field.
The launch at the vivo Developer Conference not only presented vivo’s latest achievements in artificial intelligence but also signals the company's continued dedication to AI technology, aiming to deliver more intelligent, convenient, and efficient product experiences to users.
