On November 1st, at the 2023 vivo Developer Conference, vivo announced that it will be the first to open-source the large language models BlueLM with 7 billion parameters (7B) and its corresponding fine-tuning framework.
In response to this, vivo has introduced the BlueKit large model development kit, allowing developers to focus on innovative product experiences without having to consider usage costs. In addition, vivo has also announced the launch of the BlueHeart Developer Incentive Program, providing one billion yuan (approximatly ＄14 million) worth of resources support in terms of technology, operations, and branding.
Previously, vivo announced the launch of five BlueLM with parameter levels of one billion, ten billion, and one hundred billion. These models comprehensively cover core scenarios. Among them, the 7B model with a level of 7 billion is designed for both mobile and cloud applications. Vivo claims that it performs exceptionally well in language understanding and text creation scenarios.
According to reports, in order to build its own large model system, Vivo has organized a large-scale AI expert team, and since the establishment of the Artificial Intelligence Mapping Research Institute in 2018, it has accumulated 13,000T of multi-modal data and 2,800T of high quality Chinese text data. Their engineers have continued to enrich the training data and strengthen the algorithm from multiple dimensions, while solving the problem of arithmetic power and continuously improving the technical security capacity through the application of information technology innovation.