SenseTime Group Inc, a leading Shanghai-based artificial intelligence software company, is doubling down on its development efforts on SenseCore, its AI infrastructure, to drive innovation and implementation of AI tasks in enterprises.

SenseCore integrates AI model development, generation and application, empowering a new paradigm in AI production characterized by high efficiency, low cost and scalability. It has developed nationwide network-unified scheduling and expanded new computing nodes in cities such as Shanghai; Guangdong province's Shenzhen and Guangzhou; Fuzhou, Fujian province; Jinan, Shandong province; and Chongqing, said the company.

Currently, the total computing power of SenseCore exceeds 12,000 petaflops, which are used to measure the performance of a processor's floating point unit (FPU).

SenseCore represents the most advanced domestic infrastructure for training large-scale models, with maximum connectivity of 10,000 GPU (graphics processing unit) clusters with a sustained acceleration efficiency of 90 percent.

SenseTime has also optimized training stability, demonstrating the capability of uninterrupted stable training for over 30 days, with a recovery time of only 30 minutes in case of interruptions.

In terms of large-scale model inference services, SenseTime has improved its optimization, resulting in major progress in terms of cost-effectiveness over the past year.

SenseTime announced on Wednesday the launch of SenseNova's Cantonese large model, including the SenseChat (Cantonese) Large Language Model and Sense-Chat (Cantonese) Multimodal Large Model. It was also announced that the application programming interface (API) was available to enterprise users. Meanwhile, SenseChat (Cantonese) — including the web version and mobile app — was introduced for the first time as a free service to consumers.

The major advancements in the new version focus on knowledge, mathematics, reasoning and coding capabilities.

Qi Qi, head of LLM (large language model) solutions at Sense-Time, said SenseNova currently offers various large-scale model capabilities, including natural language processing, image generation, automated data annotation and custom model training, enabling the application of LLM AI technology across industries at low cost and high efficiency.

Moreover, the company's Code Racoon inference integrated machine, a stand-alone LLM product tailored for software development, provides software development and hardware integration solutions that enhance research and development efficiency while ensuring secure operations, said Qi.

SenseTime also launched the edge-side full-stack LLM product matrix, which includes the Sense-Time Edge-side LLM for terminal devices, and the SenseTime Integrated LLM edge device that can be applied to fields such as finance, coding, healthcare and government services.

In the past year, SenseTime has focused on three core business segments: generative AI, traditional AI and smart autos. The company has placed a strong focus on developing the generative AI business while also ensuring that the traditional AI and the smart auto business segments remain competitive. Its 2023 generative AI business achieved revenue of 1.2 billion yuan ($165.7 million).