Artificial intelligence (AI) start-up DeepSeek has acknowledged the significant role that a technical solution from Tencent Holdings played in addressing a persistent issue in its system, highlighting the local open-source community’s contribution to China’s AI advances.
DeepEP, a code library for inter-chip communications, attained a “huge speed boost” by using a technical solution from Tencent’s network platform department, according to Zhao Chenggang, an AI infrastructure engineer at DeepSeek.
Tencent’s solution not only enhanced the robustness of the DeepEP project, but also “elevated communication performance”, Zhao wrote in a recent note published on a GitHub repository, where DeepSeek hosts its open-source models and projects.
DeepEP is one of eight open-source projects that Hangzhou-based DeepSeek unveiled during a week-long initiative in February, when the company revealed the technical details behind the development of its low-cost, high-performance foundation and reasoning AI models.
02:51
South Korea says DeepSeek sent data to ByteDance-owned servers in China without consent
South Korea says DeepSeek sent data to ByteDance-owned servers in China without consent
Tencent said in a statement that the technical solution builds on its extensive experience in developing and optimising data centres and communication between graphics processing units. These technological advances, which have been used to train the company’s Hunyuan AI models, have been open-sourced for external users such as DeepSeek.