The Shangtang model is open to the public for discussion, with a total test score ranking second globally
Today, the natural language model "SenseChat" developed by Shangtang Technology is open to users for service. Its base model is the "Scholar Pu Language InternLM-123B" jointly released by Shangtang and multiple domestic research institutions, with 123 billion parameters. It ranks second in the world in terms of total test scores on a collection of 300000 questions in 51 well-known evaluation sets, and surpasses GPT-4 in 12 major evaluations, ranking first.
SenseChat, officially launched in April this year, is one of the earliest large-scale language models with billions of parameters released in China, and has been continuously iterated and updated since then. At present, it is at the forefront of the industry in terms of language, knowledge, understanding, reasoning, and disciplinary abilities. It can process various types of text and information, becoming a portable comprehensive knowledge base, efficient text editor, mathematical calculator, and easy-to-use programming assistant.
It is reported that the "discussion" is based on SenseCore, a large artificial intelligence device of Shangtang. The number of GPUs launched has increased from 27000 at the end of March this year to around 30000, and the computing power has increased by 20%, reaching 6 ExaFLOPS, which can effectively support the training, upgrading, iteration, and service of language models.
In terms of training data, Shangtang can produce approximately 2 trillion tokens of high-quality data per month to support base model training. It is expected that by the end of this year, the reserve of high-quality data will exceed 10 trillion tokens. At the same time, the company has also invested hundreds of servers equipped with computing resources of thousands of GPUs. Using algorithms combined with manual methods, the original corpus data is finely cleaned and classified to ensure that the quality, security, and values of the data meet the requirements.
The performance of "Scholar Pu Language InternLM-123B" in the main evaluation collection
![The Shangtang model is open to the public for discussion, with a total test score ranking second globally](https://a5qu.com/upload/images/7f12e91afb76a1137d2224d5dbb07f29.jpg)
As of now, "Negotiate" has established cooperative relationships with over 500 customers in various industries such as finance, healthcare, automotive, real estate, energy, media, and industrial manufacturing. Through various flexible API interfaces and services, we provide customers with various artificial intelligence technologies and services for large models, and achieve various generative AI applications with low threshold and high efficiency.
It is reported that "Discussion" is one of the large model systems and generative AI product series of Shangtang's "Nissin SenseNova", and other products include "Instant Painting", "Ruying", "Qiongyu", and "Gewu". They correspond to five mainstream generative AI applications, including natural language interaction, AI cultural graphics, digital humans, 3D scene reconstruction, and 3D small object generation.
In the future, this enterprise will rely on a powerful base model and technological accumulation based on computing power, data, and algorithms to continuously upgrade various generative AI products under the "Daily New" big model system.
Publicly accessible https://chat.sensetime.com After completing the registration, you can use "Discuss SenseChat" or experience the effect of "Discussing" with AI models to solve problems by visiting the official website of Shangtang.