ZTE's "Star Cloud R&D Large Model" unveiled, helping developers improve R&D efficiency and quality

At the China Mobile Global Partner Conference, ZTE showcased its self-developed "Nebula R&D Grand Model," which can assist developers in completing the entire process of requirement analysis, product design, programming, testing, and version deployment.

The "Nebula R&D Grand Model" features security and controllability, supporting functions such as whitelist mechanism, code feature recognition, sensitive word recognition, and background auditing, effectively preventing sensitive information leakage and security risks. ZTE stated that since the model was launched in April 2023, the daily active users have reached 12,000, the code adoption rate is 40% to 45%, coding efficiency has increased by 30%, and overall R&D efficiency has improved by 10%.

ZTE reportedly injected a large amount of high-quality domain data, Know-How knowledge accumulation, hundreds of thousands of technical documents in the communication field, and 100 billion tokens of wireless/core network/cloud code corpus into the grand model for incremental pre-training, using parallel training frameworks. ZTE also independently developed deployment solutions, significantly improving throughput and GPU memory utilization through dynamic batch strategy, PagedAttention technology, and lossless model quantization technology. ZTE claimed that the solution achieved a throughput of 1500 tokens/s on a single GPU (A800), and only requires four GPU cards (A800) to meet the needs of thousands of users. Compared with conventional deployment solutions in the industry, the throughput of a single GPU has increased by more than 10 times and 20 times, respectively. Meanwhile, without reducing model accuracy, the use of int4 quantization technology has reduced the model size and GPU memory usage by half.

The "Nebula R&D Grand Model" is an important innovation achievement of ZTE in the future network intelligence. It will effectively help developers improve R&D efficiency and quality, reduce R&D costs and risks. ZTE will continue to invest more innovative resources and work with global customers and partners to promote the development of communication networks towards high intelligence.