In an age where smartphones have become almost an extension of the human body, a quiet revolution is quietly taking place - the rapid advancement of AI (artificial intelligence) technology is beginning to redefine the way we interact with our phones.

IDC (International Data Corporation) predicts that there will be a significant turning point in the next few years: in 2024, global shipments of AI smartphones will reach 170 million units, and by 2027, shipments of AI phones in the Chinese market alone will reach 150 million units. It is obvious that AI is no longer a distant future technology, but a real force that has been integrated into our lives.

Faced with the booming market, mobile phone manufacturers around the world have begun to take action and pressed the fast-forward button in the AI ​​large-scale model research and development competition. Mobile phones have become the first battlefield for the implementation of AI.

Challenges also follow.

How to build an AI phone? In the past, large models were deployed in the cloud and completely relied on cloud computing power to run, but this approach often failed to meet users' needs for instant response and high personalization.

In fact, the challenges hidden behind this technological evolution also include how to run increasingly complex AI models on mobile phones with limited resources, and how to solve problems such as network latency and data synchronization.

Finding a complete solution has become the current consensus in the communications industry.

At this critical moment, Alibaba Cloud and MediaTek recently jointly announced the end-cloud collaboration technology, successfully deploying the "Tongyi Qianwen" large model on the SoC, realizing the deep adaptation of the large model on the mobile phone chip for the first time.

This innovation not only ensures that users can still enjoy smooth AI services even when the network is disconnected, but also indicates that driven by end-cloud collaborative technology, the user experience of AI smartphones has leapt into the "kingdom of freedom", that is, AI can be used anytime and anywhere, meeting users' needs for AI functions in complex scenarios.

Behind the innovation lies the deep cooperation results between Alibaba Cloud and MediaTek in model slimming, tool chain optimization, inference acceleration and other aspects. It has overcome multiple challenges from the underlying chip to the upper-level operating system and application development, and demonstrated the huge potential of deep integration of software and hardware.

This exploration of edge AI technology not only opens up a new channel for the commercialization of Model on Chip, but also provides new solutions for global mobile phone manufacturers, indicating that the smartphone industry is about to enter a new AI era.

New Trend——Model on Chip

Optimizing and deploying large artificial intelligence models on mobile phone processors overcomes the traditional reliance of mobile phones on cloud computing power. This innovation not only enables advanced AI functions such as natural language processing and image recognition to run directly on mobile phones, but also significantly improves execution efficiency, while achieving a qualitative leap in user experience and data privacy protection.

Xu Dong, business manager of Tongyi Lab, pointed out: "As the key to the implementation of large model applications, end-side AI faces challenges such as hardware and software adaptation and insufficient development environment. The cooperation between Alibaba Cloud and MediaTek has successfully overcome these difficulties. Through comprehensive technical research from underlying hardware to upper-level applications, large models have been effectively deployed on mobile phone chips, leading a new trend in the deployment of end-side AI in Model on Chip."

It is understood that the open source large model of "Tongyi Qianwen" with 1.8 billion parameters surpasses previous models and has made significant progress in performance and resource utilization efficiency. It can handle reasoning of up to 2048 tokens using only 1.8G of memory, demonstrating its low cost, easy deployment and extremely commercial-friendly characteristics.

Observers believe that this technological innovation has broken through previous limitations and made it possible to deeply adapt large models to mobile phone chips. The technical principles behind it are largely due to Alibaba Cloud's product-oriented thinking.

First, before entering a small mobile phone, the large model needs to be "slimmed down". The model size is reduced through a variety of technical means such as quantization, parameter pruning and knowledge distillation.

Quantization is the process of converting the model's floating-point parameters into more efficient low-width integer forms, reducing the need for storage and computing resources; parameter pruning is the process of reducing the size of the model by removing non-core parameters; knowledge distillation is the use of a small but efficient model to imitate the behavior of a complex large model, which can both make the model lightweight and maintain its performance.

Secondly, although the quantized model is smaller in size, it may suffer from performance loss. Therefore, various optimizations and careful adjustments are needed to further improve the model performance and ensure its efficient operation on mobile phones.

Finally, the optimized large model needs to further enhance its potential, gradually strengthen the model capabilities, and better adapt to the specific needs of terminal applications.

Alibaba Cloud's end-cloud collaboration technology combines the advantages of cloud computing and edge computing, and compared with traditional methods, it brings five major advantages: low latency, privacy protection, offline capabilities, bandwidth saving, and real-time processing. This technology is crucial for future application scenarios such as autonomous driving and smart manufacturing.

End-cloud collaboration reshapes mobile AI

It is foreseeable that the end-cloud collaboration technology has completely changed the computing efficiency of AI mobile phones. For example, the intelligent voice assistant can quickly process simple requests, such as weather queries, while complex tasks rely on the powerful computing power of the cloud to complete, thus providing an efficient and flexible user experience.

In terms of privacy protection, the end-cloud collaboration technology greatly improves data security and privacy by processing sensitive data on the device side. For example, in health monitoring applications, personal health data will first be encrypted locally, and only necessary information will be securely transmitted to the cloud, effectively reducing the risk of leakage.

According to the latest data forecast, with the continuous advancement of end-side AI models and hardware technology, the market types and quantity of AI smartphones are expected to grow significantly by 2024. The market penetration rate is expected to reach 6.4%, and is expected to double to 19% in 2025.

This trend was clearly reflected at the 2024 Mobile World Congress, where Honor, OPPO, vivo and other mobile phone brands showcased their latest AI phones, highlighting the many applications of AI technology in improving user experience. With the decentralization of Alibaba Cloud's cloud collaboration technology, the competitive landscape of the smartphone market will inevitably undergo major changes - who can refuse a mobile phone that can run large AI models locally?

At the same time, the end-cloud collaboration technology also provides a powerful driving force and a broad platform for the innovation of mobile applications. Developers can use this technology to develop more innovative applications to meet the personalized needs of users, thereby accelerating the vitality of the mobile application market and the prosperity of the ecosystem.

Conclusion

This revolution driven by end-cloud collaboration technology not only heralds an unprecedented new era for mobile phone performance and applications, but also symbolizes that the deep integration of humans and AI has taken another step forward. As a result, we are also standing on a new high-speed road of technological development, witnessing the gradual evolution of mobile phones from traditional communication and entertainment tools to all-round partners that integrate AI intelligence. It is foreseeable that with the further integration with AI technology, the future of mobile phones will also change into infinite possibilities.

As a winner of Toutiao's Qingyun Plan and Baijiahao's Bai+ Plan, the 2019 Baidu Digital Author of the Year, the Baijiahao's Most Popular Author in the Technology Field, the 2019 Sogou Technology and Culture Author, and the 2021 Baijiahao Quarterly Influential Creator, he has won many awards, including the 2013 Sohu Best Industry Media Person, the 2015 China New Media Entrepreneurship Competition Beijing Third Place, the 2015 Guangmang Experience Award, the 2015 China New Media Entrepreneurship Competition Finals Third Place, and the 2018 Baidu Dynamic Annual Powerful Celebrity.