Ascend AI & iFlytek Spark: In-depth cooperation to discuss the "big future" of domestic large-scale models

Currently, large pre-trained models of artificial intelligence are leading a new wave of intelligence, pushing artificial intelligence from specialized fields to general applications. At the 2023 World Artificial Intelligence Conference, large models once again aroused industry attention. More than 30 large models at home and abroad competed on the same stage, and global artificial intelligence cutting-edge technologies were collectively unveiled.

On July 6, at the Shengteng Artificial Intelligence Industry Summit Forum hosted by the New Generation Artificial Intelligence Industry Technology Innovation Strategic Alliance (AITISA) and the China Artificial Intelligence Industry Development Alliance (AIIA) and hosted by Huawei, Senior Vice President of iFlytek Hu Guoping, President and Director of the National Key Laboratory of Cognitive Intelligence, announced the cooperation between iFlytek and Huawei. iFlytek Spark and Ascend AI have joined forces to create a new universal intelligence base in China. Hu Guoping said that China’s independent innovation of computing power base is the key to realizing the big future of domestic large-scale models.

Hu Guoping reviewed the development history of the iFlytek Spark cognitive large model. With many years of core technology reserves, the "1 N" cognitive intelligence large model special research project was launched on December 15, 2022. May 6, 2023 The iFlytek Spark Cognitive Model was officially released, debuting seven core capabilities including text generation, language understanding, knowledge question and answer, and logical reasoning. It continued to iterate and was upgraded again to release the Spark Cognitive Model V1.5 on June 9.

Focusing on the Spark model, Hu Guoping also shared the application of large models in education, office, automobile, medical, industry and other fields. Based on the innovation step of core technology, the Spark model has achieved from 0 in multiple industry scenarios. to 1 innovative applications.

Behind the accelerated iteration and catching up of large models is a computing power challenge that cannot be ignored.

However, the current development of large models is highly dependent on high-end AI chips, clusters and ecology. High computing performance, high communication bandwidth and large video memory have become the indispensable computing power base for large model training. The progress of single AI chips has not kept up with the demand for large computing power for large models. Computing power clustering has become an irreversible development trend. .

The key to the safety and development of my country's large models lies in relying on independently innovative hardware and software to promote rapid progress in the large model ecosystem. "Hu Guoping emphasized that the cooperation between iFlytek Spark and Ascend AI has enabled the domestic large-scale model architecture to work together on the basis of independently innovative software and hardware." "On the one hand, the iFlytek Spark cognitive large model is based on the integrated design of training and reasoning, achieving technological breakthroughs in large model sparseness and low-precision quantification. It can efficiently adapt to Ascend AI and accelerate the industrial application and iteration of large models; on the other hand, In terms of aspect, with Shengteng AI as the core, software and hardware are collaboratively optimized to build a large model training cluster with concentrated computing power, superior performance, stable supply, and data security.

In Hu Guoping's view, large models are similar to the principle layer of the brain. They are combined through more than 100 billion neurons, receive input stimulation, and then produce intelligent output. They have similar intelligent stimulation and operation mechanisms. "What the brain can do, large models can also achieve. This indicates that large models have unlimited potential. Artificial intelligence has gone through four waves. In the era of large models with the emergence of intelligence, it is possible to finally find the right solution."

Looking to the future, with more data, larger models, stronger demands, and more complex tasks, large models will continue to call for large computing power.

"We are willing to work with Shengteng AI to seize new historical opportunities for general artificial intelligence and strive to build a new base for general intelligence in my country." Hu Guoping said.

The above is the detailed content of Ascend AI & iFlytek Spark: In-depth cooperation to discuss the "big future" of domestic large-scale models. For more information, please follow other related articles on the PHP Chinese website!