Shulou(Shulou.com)11/24 Report--

CTOnews.com, August 24, 2023-on the morning of August 18, 2023, Intel held a technology sharing session in the Greater Bay area on Intel's latest developments in the field of AIGC (artificial Intelligence Generation Computing). At the meeting, Intel technical experts showed the participants their technical direction on AIGC and demonstrated a number of applications.

First of all, Intel introduced their optimization and support for the large language model. In our traditional understanding, to run a large language model like ChatGPT, it must be supported by a large graphics card. But this exchange has subverted our perception. Intel has built a BigDL-LLM library to enable the 12th and 13th generation core platforms to run a variety of large language models and provide a smooth experience. Through this library, you can optimize and support a variety of large language models, including some open source large language models that can run locally. The library can even run a large language model with up to 16B parameters on an Intel lightweight machine with 16GB memory. In addition, many large language models such as LLaMA / LLaMA2, ChatGLM / ChatGLM2 and so on are supported.

Next, Intel demonstrated their performance in the application of large language models. Through Demo, which integrates ChatGLM2, LLaMA2 and StarCoder, they demonstrate the performance of these models in Chinese and English applications. Through optimization and quantification operations, these large language models can generate answers at a very fast speed without affecting the reading experience. Intel has released this Demo, which can be installed and experienced directly on any computer with 12 or 13 generations of Core.

We chose a slim book certified by Intel Evo platform: Huawei MateBook 16s to test it. The processor is Core i9-13900H + 32G memory to see if you can run AIGC on the slim kernel.

Intel big language model Demo installation process is very stupid, Intel has been packaged into an EXE file, install and import the .bin format big language model can run. When we open the interface, we can see that we can choose the chat content, adjust the model parameters and view the operation delay on the left, and there is a chat box on the right.

The author first tried to ask him some questions about the world, but the answer was very good and the response speed was very fast. Such a question only takes 658.87ms to complete the response.

In the process of operation, we can see that the occupancy rate of i9-13900H has reached 100%, and the memory occupancy has reached 16G, which is also occupied by Xe kernel display. It seems that the operation is indeed carried out locally. After Intel's continuous optimization and the improvement of the computing power of the 13th generation Core processor, we have really achieved the landing of AIGC on lightweight books.

In the function of emotional analysis, we let the big language model analyze a classical prose. We can see that although the perspective of the analysis is relatively simple, but there are no logical errors, it can still be self-consistent.

In this Chinese translation function, Intel big language, the performance of the model even more surprised me, its translation quality is very high, the speed is also very fast. Even manuscripts with so many proper nouns can be translated accurately.

In terms of story creation, this big language model also shows my amazing logical and creative ability. I asked it to write a story about Guan Yu fighting Lin Daiyu and Lin Daiyu winning in the end. But AI realized that the two men were not of the same age at all, and finally wrote a drama through time and space, which was full of loopholes, but generally no problem.

The function of generating outline is a very useful function, as long as we enter the manuscript we want to write, it can help us to make a logical outline. This can play a good auxiliary role for groups who often need to write articles.

The information extraction function is very helpful for groups who often need to read reports. You can quickly extract core information from long articles. I tried to get AI to extract Chinese information from English articles, but the effect was still very good.

Finally, the author asked AI to give a Qingdao food advice and Qingdao tourism advice. Because this large language model is completely offline, the information given will be old, but it is good in terms of writing level alone.

In addition to large language model applications, Intel also demonstrated support for the AI text graph algorithm Stable Diffusion. They enabled OpenVINO acceleration and developed an AI framework that can speed up the running of the PyTorch model with a single line of code installation. With Stable Diffusion's WebUI, you can run Stable Diffusion Automatic1111 on integrated video cards and Arc stand-alone video cards. Through the demonstration, you can see the performance of Stable Diffusion on integrated graphics cards on Evo lightweight books and machines with i7-13700H processors. The powerful computing power of the 96EU version of Intel Ruiju Xe graphics card can support FP16 precision models running on Stable Diffusion software to quickly generate high-quality pictures.

At the same time, they also demonstrated the effect of running Stable Diffusion on machines with i7-13700K CPU + Arc A770, which is very fast.

In addition, Intel also demonstrated three-dimensional digital reconstruction of character movements based on Arc graphics cards. Through the powerful computing power of Arc and the optimized reasoning of AI algorithm by OpenVINO framework, we can detect and reconstruct the actions of characters in real time, and render them later. This technology does not require a smart wearable device, but only needs to connect a home camera to detect and reconstruct character movements in real time, and animate through the image of a digital human. Through real-time skeletal point information and 3D reconstruction, digital people can show flexible and diverse body movements and achieve real three-dimensional dynamic effects. At the same time, according to the needs of users, you can customize the painting style of the renderer and quickly create a variety of meta-universe applications.

In the final turbine card demonstration, Intel showed off their new turbine GPU card, the Arc A770 16G. This card has a double-slot full-length and full-height design, suitable for a variety of edge-side server chassis or equipment chassis. It adopts turbofan design to reduce the interference to the flow field in the chassis, and is more suitable for the server and alchemy furnace scenario of multi-card hybrid computing.

From the point of view of this sharing meeting, Intel has been focusing on the development of AI technology, especially in the hot field of AIGC, Intel is not absent. With the powerful computing power of 13 generations of Core and continuous algorithm optimization, even lightweight books can now support the local operation of large language models. With the blessing of Arc graphics card, the support of Stable Diffusion graphics is also being carried out continuously. At the same time, Intel is also exploring the next generation of AI application scenarios and providing more powerful solutions for enterprise users.

Through the demonstration of these technologies, Intel shows its leading position in the field of AIGC and its ability to innovate continuously. They are committed to providing users with a more intelligent and efficient computing experience and promoting the development and application of artificial intelligence technology. With the continuous progress and improvement of technology, we can expect to see more AIGC applications and solutions from Intel in the future.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.