ZTE launches 'Nebula R&D Big Model”: AI programming assistant, 100 billion tokens make shocking debut

IT House reported that from October 11 to 13, 2023, during the China Mobile Global Partner Conference, ZTE demonstrated their "Nebula R&D Large Model", which is designed to help developers conduct demand analysis, Product design, programming, testing and version deployment

According to reports, the "Nebula R&D Large Model" supports a whitelist mechanism, which can effectively control the scope of use. At the same time, it can also effectively identify sensitive code fragments through code feature value recognition, and monitor and intercept sensitive content in real time through the sensitive word recognition mechanism. In addition, the model also has a background audit mechanism that can completely trace back security events

ZTE stated that in April 2023, the "Nebula R&D Large Model" was launched. As of now, the number of daily active users has reached 12,000, the code adoption rate has reached 40%~45%, the coding efficiency has increased by 30%, and the overall R&D has improved. Effective 10%.

According to the official announcement, IT Home learned that ZTE will inject domain data, knowledge accumulation, a large number of technical documents in the communication field, and 100 billion wireless/core network/cloud code corpus into large-scale models for increment. Pre-training and using parallel training framework

ZTE claims: “Our self-developed deployment solution uses dynamic batch processing strategy and PagedAttention technology, combined with lossless model quantification, which greatly improves throughput. The throughput of a single GPU (A800) reaches 1500 tokens/second , using only 4 GPU cards (A800) can meet the needs of more than a thousand people. Compared with the conventional deployment scheme in the industry, the throughput of a single GPU has been increased by more than 10 times and more than 20 times respectively; at the same time, combined with int4 quantification technology , without reducing model accuracy, the model size and video memory usage are reduced by half.”

The above is the detailed content of ZTE launches 'Nebula R&D Big Model”: AI programming assistant, 100 billion tokens make shocking debut. For more information, please follow other related articles on the PHP Chinese website!