Shulou(Shulou.com)11/24 Report--

According to the latest data of OpenCompass evaluation platform, OPPO large model AndesGPT-7B ranks fifth in the comprehensive list of Chinese data sets with a comprehensive score of 62.2, and ranks first among the large models with the same scale parameter of 7 billion (7B). It is worth noting that the complete open source reproducible evaluation framework provided by OpenCompass can evaluate many capabilities of the large model at the same time, which also means that AndesGPT-7B is the first large open source model of domestic terminal manufacturers, and will continue to contribute to the open source community of the Chinese big model.

(OpenCompass Big language Model Evaluation list-Chinese dataset, 2023-10-30)

OpenCompass is an open source, efficient and comprehensive evaluation large model system and open platform, which provides data sets and evaluation methods of different dimensions. Based on language, knowledge, reasoning, discipline, understanding, five dimensions, more than 50 data sets evaluate the capabilities of large language models. Including zero sample evaluation, small sample evaluation and thinking chain evaluation, omni-directional quantification of each dimension of the model capabilities, a total of 40 + tasks cover various capabilities.

As a large language model for self-training of OPPO, AndesGPT takes "end-cloud collaboration" as the infrastructure design idea, and includes a series of large language models with different parameter scales, starting from billion to up to hundreds of billions of levels, which can support a variety of application scenarios. It is reported that OPPO has started the exploration and practice of large model related technologies since 2020, and there are more than 1400 patents related to large models. At present, it is also working with more than 10 universities to explore in-depth cooperation, related topics related to end-to-side lightweight, large model enabling ubiquitous services, intention reasoning, multi-modal generation and so on. When AndesGPT-7B performs well on the OpenCompass platform, it also contributes to the common progress of the large model open source community.

At present, the new Xiaobu assistant based on the AndesGPT model has opened a public test, demonstrating some of its cloud application capabilities. At the same time, OPPO also cooperates with MediaTek to build the end-to-side deployment scheme of lightweight large models, achieving better performance at the lower end of the accuracy effect, and jointly promoting the gradual landing of AndesGPT large language models and multi-modal large models on the end side.

AndesGPT will be used for the first time in the upcoming ColorOS 14 system on November 16, and a more diversified application scenario for AndesGPT may be revealed later.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.