iFLYTEKThe group’s official WeChat account published a blog post stating that at the results conference held in Beijing yesterday (August 28),China National Petroleum Corporation33 billion parameters releasedKunlun Large Model,It is the first large model to pass the registration in China's energy and chemical industry.
Introduction to Kunlun Model
According to a press release, the Kunlun Big Model was jointly built by China National Petroleum Corporation, China Mobile, Huawei and iFlytek, and a cooperation and construction agreement was signed in May this year. In accordance with the "Five Ones" action plan, 8 big models were trained and 18 application scenarios were developed.
Kunlun Big Model has passed the national generative artificial intelligence service registration on August 23.It is the first large model to pass the registration in China's energy and chemical industry.
The top-level design has three major characteristics:
- First, an innovative four-layer architecture is proposed to support the needs of massive data modeling in professional fields such as oil and gas new energy, refining and chemical new materials;
- The second is to coordinate the management of the large model architecture, formulate large model construction standards, and guide the orderly development of large model construction;
- The third is to open up and share the Kunlun big model capabilities, empower the vast number of practitioners in the energy and chemical industry, and lead the intelligent development of the industry.
In addition, the model has the following main advantages:
- Large model training, trained and released 8 large models of different levels, types and sizes to meet the needs of different business scenarios.
- Industry big model, released language models with 13 billion parameters and 33 billion parameters, as well as a visual model with 300 million parameters, effectively improving professional capabilities such as industry knowledge question answering, concept understanding, paper abstract generation, and industrial visual understanding;
- Professional large model, released two large models with professional characteristics: 5 billion parameter seismic interpretation and 100 million parameter well logging processing interpretation, and achieved significant results in intelligent application;
- Scene large model, released three large models: 13 billion parameter intelligent query, 300 million parameter device identification, and 16 billion parameter customer marketing, to support business needs such as intelligent operation query and image and text generation.