OPS2.0 was released, and the large model went to an important "puzzle" on the end side

Since the beginning of this year, technology giants have finally set their sights on the industry side and the end side. Zhou Hongyi, founder and chairman of 360 Group, once pointed out that in specific landing scenarios, while making large models "big", making them "small" is also an important trend, so that large models can be carried on mobile phones, computers, and various Internet of Things devices, especially intelligent networked vehicles, and more large models will be equipped in 2024.

Xie Dong, chief technology officer and general manager of the R&D center of IBM Greater China, also publicly stated that for enterprises, the purpose of the application model is to hope that it can solve specific problems at a lower cost. Xie Dong said.

Indeed, it is not difficult to see from the dynamics of major manufacturers this year that compared with the "large" model parameters, manufacturers are focusing more on how to train "small" models that are easy to use in vertical industries. The application of large model capabilities on the end side such as mobile phones, computers, and automobiles has also become one of the topics focused on by technology giants.

The large model that moves to the end side needs to be "soft and hard"

In the process of the real implementation of large models on the industry side and the end side, whether it is the software level or the hardware level, it will usher in a new round of changes.

From the hardware point of view, in addition to the high requirements and requirements for many chips such as CPUs and GPUs, the large model also has different high requirements for the heat dissipation and connection of the equipment.

With the development of technology and society, as well as artificial intelligence, the digitalization of education has put forward new requirements for IT equipment, how to apply AI chips to educational equipment, and empower the development of the education industry has become a hot topic at present. AI technology can also enable teachers and students to interact more efficiently, so that teaching and learning can be mutually beneficial. With the help of AI technology, teachers can also enable the whole school to teach students according to their aptitude, so as to achieve the goal of education.

Taking the connection level as an example, when enterprises purchase equipment, they often do not purchase only one device, which puts forward higher requirements for the compatibility between devices and the connection/converter between devices. And this level has also become an important part of the ecological process of the construction industry.

It is reported that in order to meet the needs of AI large models that want to be landed on the end side, a few days ago, during the 83rd China Educational Equipment Exhibition, Intel and its partners jointly released a new OPS (Open Pluggable Specification) 2.0.

Lu Yingjie, general manager of HEC China of Intel's Network and Edge Division, told Titanium Media APP that the release of the new OPS2.0 specification provides a set of standardized operation processes and interfaces for the deployment of AI large models on the device side, which helps the collaboration between different vendors and developers and promotes the wide application of AI large models on the device side.

With the continuous development of AI technology, the traditional device-side equipment can no longer meet the needs of the current industry, taking the interactive tablet as an example, the traditional interactive tablet is more to have display functions, but with the penetration of AI large models into the education industry, the interactive tablet now needs to have more interactive and teaching assistant capabilities. For example, it is necessary to have the ability of teacher-student collaboration, that is, to see what you get, that is, to write and display, not only that, in Wang Jingjia's view, with the development of technology, the IT equipment on the end side should reserve enough expansion interfaces to provide support and capabilities for future technology and product deployment.

The just-released OPS2.0 module can be regarded as a pluggable computing unit, which has corresponding expansion capabilities in different scenarios such as adapting to CPUs and GPUs, "This is one of the biggest improvements between OPS2.0 and 1.0." Wang Jingjia emphasized.

In addition to the hardware level, the continuous iteration of the software level also plays a vital role in the implementation of large models in the industry. Also taking the education industry as an example, Peng Yuantao, deputy general manager of Shenzhen Deshengda Electronic Technology Co., Ltd., told Titanium Media APP that in the past, many interactive tablets used Android operating systems, but the use rate and convenience in actual teaching were not high. However, the upgraded OPS2.0 can be compatible with eDP, further simplifying the hardware architecture of the large screen, while also reducing the writing delay, "Now, our eDP can be directly connected to the screen through OPS, and the transmission speed is higher and the delay is lower through the form of software optimization." Peng Yuantao pointed out.

Not only that, Lu Yingjie told Titanium Media APP that at the level of software tools, Intel has also created tools including OpenVINO to meet the needs of current end users for training and inference, "Specific to the field of large models, Intel is also doing some large model exploration and project optimization with different general ISVs and industry ISVs." Lu Yingjie pointed out.

Demand is emerging, and technology is driving development

With the continuous development of AI technology, we have also seen more and more industries begin to embrace the dividends brought by AI, and more and more application scenarios will emerge. Taking the education industry as an example, the digital transformation of education has entered the stage of AI empowerment, and what teachers and students need is not only a simple display and a digital classroom, but also the digitization and intelligence of the entire teaching cycle.

For teachers, teachers need to have a more accurate and detailed understanding of each student's performance in class, data on interaction with teachers in class, and after-class learning.

For students, students are eager to feel the content of the textbook more realistically, and each student needs to have an exclusive teacher who can customize an exclusive learning plan according to their own situation;

For parents, they want to understand their students' performance and learning in school through simple and clear visual data, and can communicate with teachers at the teaching level through simple digital tools.....

In response to this, Huang Bolin, General Manager of the Commercial Display Division of CVTE BG, pointed out that at present, teachers are faced with many problems, such as the difficulty in ensuring the efficiency and quality of teaching design, the high labor cost of teaching quality evaluation, the difficulty in quantifying the results of teaching and research activities, and the lack of objective basis for classroom review.

These are just a few of the many needs facing the education industry today, and with the gradual application of AI technology in the education industry, these needs will be met one by one. In Huang Bolin's view, these are the real needs of the current education industry, and there is still a long way to go to drive the implementation of these needs with technology.

Huang Bolin told Titanium Media APP that in the future, the AI+ teaching model will be mainly displayed in four aspects: lesson preparation assistant, discussion and efficiency, classroom feedback, and learning evaluation.

In terms of lesson preparation assistant, teachers can quickly generate courseware and beautify courseware through the ability of AI large models to achieve better teaching results.

In terms of efficiency enhancement of discussions, teachers can use the ability of large language models to summarize and refine opinions, which reduces the workload of teachers in discussions and improves efficiency.

In terms of classroom feedback, teachers can capture data based on NPU calculations and collect classroom activity data, and through the capabilities of AI large models, they can deeply analyze each student's classroom performance and provide intelligent feedback on students' performance.

In terms of learning situation analysis, after class, teachers can analyze and evaluate the data of each student's class, day, week, or even semester according to comprehensive data, and use the data to truly reflect each student's learning situation and correct homework.

In Huang's view, these four aspects are the main directions for the deep integration of education and AI in the future. Of course, this is also inseparable from the coordinated development of software and hardware, as well as the construction of the entire education digital ecosystem. The iteration of OPS2.0 is an important piece of the puzzle in the process of moving large model capabilities to the device side. (This article was first published in the Titanium Media APP, by |.) Zhang Shenyu, editor | Gai Hongda)