MediaTek partners with Alibaba Cloud for large models mobile chips

Source: www.ichunt.com

According to the Sci-Tech Innovation Board Daily, MediaTek, a smartphone chip manufacturer, has successfully deployed large models with 1.8 billion and 4 billion parameters on its flagship chips such as the Dimensity 9300, achieving deep adaptation of the large model on mobile chips and enabling Tongyi Qianwen to run multiple rounds of AI dialogue even in offline situations. In the future, the two parties will also adapt more large models of different sizes, including one with 7 billion parameters, based on the Dimensity chip.

Alibaba Cloud stated that it will work closely with MediaTek to provide end-side large model solutions to global mobile phone manufacturers.

Currently, MediaTek is the semiconductor company with the highest shipment volume of smartphone chips globally. According to the latest data from Canalys, it shipped over 117 million units in the fourth quarter of 2023, ranking first, followed by Apple with 78 million shipments and Qualcomm with 69 million shipments. Tongyi Qianwen is a fundamental large model developed by Alibaba Cloud. So far, it has launched versions with up to 100 billion parameters and open-source versions with 72 billion, 14 billion, 7 billion, 4 billion, 1.8 billion, and 500 million parameters, as well as multi-modal large models such as the visual understanding model Qwen-VL and the audio large model Qwen-Audio.

During MWC2024, MediaTek showcased various artificial intelligence applications, including the Dimensity 9300 and 8300 chips. It is understood that the Dimensity 9300 chip has already supported the application of Meta Llama 2's 7-billion-parameter large model overseas, and it has been implemented on vivo X100 series phones in China with a 7-billion-parameter large language model on the end side, and has also successfully run a 13-billion-parameter model in an end-side experimental environment.

The collaboration between MediaTek and Alibaba Cloud marks the first time that the Tongyi large model has achieved chip-level hardware and software adaptation. Xu Dong, the business head of Alibaba's Tongyi Lab, explained, "End-side AI is one of the important scenarios for the application of large models, but it faces many challenges such as difficulties in hardware and software adaptation and incomplete development environments. Alibaba Cloud and MediaTek have overcome a series of technical and engineering challenges related to underlying adaptation and upper-level development, truly integrating the large model into the mobile chip and exploring a new deployment model of Model-on-Chip for end-side AI."

In addition to MediaTek, Qualcomm is also actively promoting the implementation of large models on mobile devices. On March 18th, Qualcomm announced the launch of the third-generation Snapdragon 8s mobile platform, which supports large language models with up to 10 billion parameters and can also support multi-modal generative AI models, including Baichuan-7B, Gemini Nano, Llama 2, and Zhipu ChatGLM from companies such as Baidu Xiriver, Google, and META. It is reported that Xiaomi Civi 4 Pro will be the first to be equipped with the Snapdragon 8s mobile platform.

A consumer electronics industry analyst stated that the collaboration between Alibaba Cloud and MediaTek means that domestic mobile phone manufacturers now have an alternative to Baidu.

According to the reporter's understanding, Honor and Samsung have previously announced collaborations with Baidu Wenxin Yiyan. For example, Samsung's latest flagship phone, the Galaxy S24 series, integrates multiple capabilities of the Wenxin large model, including calling, translation, and smart summarization. Additionally, a source revealed that Apple is currently in contact with Baidu, hoping to use Baidu's artificial intelligence technology, and preliminary talks have already taken place between the two parties.

Lin Dahua, a leading scientist at the Shanghai Artificial Intelligence Laboratory, stated that with the exponential growth of cloud-based large models, the end-side is about to enter a golden growth period. Cloud-end collaboration will become an important trend in the future, with cloud-side computing establishing the ceiling and end-side computing supporting large-scale user adoption.

According to predictions from the consulting firm IDC, the shipment volume of smartphones in the Chinese market will reach 277 million units in 2024, with a year-on-year growth rate of 2.3%. Among them, the shipment volume of AI phones will reach 36.6 million, with a year-on-year growth rate exceeding three digits. The application of large AI models on mobile phones will become more widespread.