On April 27, the China Academy of Information and Communications Technology announced via its official WeChat account that it has officially launched the domestic adaptation testing for DeepSeek V4. This initiative, carried out in collaboration with the Artificial Intelligence Software and Hardware Collaborative Innovation and Adaptation Verification Center, aims to promote deep adaptation between DeepSeek V4 and domestic software and hardware, accelerate technological co-optimization, and drive industrial application deployment.
The adaptation testing leverages key laboratories such as the Key Laboratory of Artificial Intelligence Large Models and Software/Hardware Evaluation under the Ministry of Industry and Information Technology, as well as the AISHPerf (Performance Benchmarks of Artificial Intelligence Software and Hardware) benchmarking system and testing tools. The tests cover a wide range of AI software, hardware products, and systems, including chips, servers, integrated machines, clusters, development frameworks and toolchains, and intelligent computing facilities and platforms.
The testing encompasses the full series of DeepSeek V4 models, with a focus on inference, fine-tuning, and other processes. Evaluations are conducted across multiple dimensions, including adaptation ease-of-use, functional completeness, optimization effectiveness, performance metrics, and cost. In light of the model’s technical characteristics, new evaluation directions have been added, such as long-sequence processing, code generation and comprehension, successful agent invocation rates, and task decomposition capabilities, forming a comprehensive three-dimensional evaluation system to thoroughly assess adaptation readiness.
Recently, DeepSeek V4, the next-generation model from DeepSeek, was officially released and simultaneously open-sourced. On the day of its release, multiple domestic hardware manufacturers carried out “0-day adaptation,” marking a new phase of synchronized iteration and seamless integration for China's AI software and hardware ecosystem. This development further underscores the importance of collaborative software-hardware optimization.