Enabling a Myriad of Models and Modalities to Help Countless Industries Go Intelligent Faster
Produits, solutions et services pour les entreprises
Akıllı Telefonlar, Bilgisayar&Tabletler, Giyilebilirler ve Daha Fazlası
In his book Homo Deus: A Brief History of Tomorrow, Yuval Noah Harari wrote, "The single greatest constant of history is that everything changes." In Sapiens: A Brief History of Humankind, he detailed the Cognitive Revolution, during which homo sapiens came into being. The Second Cognitive Revolution, powered by advanced technologies like AI and big data, is right around the corner.
We are currently bidding farewell to the digital world and standing at the threshold of an intelligent world. AI-Generated Content (AIGC) was on the list of top 10 tech buzzwords for 2022. AIGC and foundation models are advancing faster than ever, thus taking AI further and expanding its role beyond perception and comprehension and towards content generation. We see AI applications everywhere, in living rooms and on factory floors, at the edge and in core systems, and in both consumer and business scenarios. AI is bringing a seismic change to almost all sectors of society.
Huawei's customers have implemented a wide array of solutions based on AI this year. If you open the website for the European Centre for Medium-Range Weather Forecasts (ECMWF), you can find 10-day weather forecasts that are generated in just seconds. These forecasts are made using Huawei Cloud's Pangu Weather Model. The AI gives new depth to weather forecasts, including fine-grained meteorological features such as geopotential height, humidity, wind speed, temperature, and sea level pressure. This model is already being applied to predict the paths of typhoons.
Another example is Baidu's Ernie Bot foundation model. Version 3.5 of this model, launched this year, comes packed with new fine-tuning technologies, stronger knowledge point management, and augmented logical and inference capabilities. The synergy between the PaddlePaddle deep learning platform and the Ernie foundation model makes broader adoption of this model possible. Baidu decided to join hands with Huawei Ascend AI to build a hardware ecosystem based on PaddlePaddle and Ernie.
China Railway Zhengzhou Group has employed Huawei's Pangu-powered Train Freight Detection System (TFDS) Solution. The solution leverages the pretrained foundation model to greatly improve the 5T inspection workshop in Zhengzhou North Rolling Stock Depot. This has tripled detection efficiency, increased the fault detection ratio to 99.3%, and led to 50% faster fault detection.
Another direct beneficiary of Pangu models is the counter clerks in the Guangdong Rural Credit Union (GDRC). The bank now uses the Pangu optical character recognition (OCR) foundation model for the finance sector to automatically recognize hand-written information, such as dates, account numbers, account names, account-holding banks, and amounts. Not only does this reduce workloads, but it improves field accuracy from 83.9% to 91.0%.
Finally, the Shandong Energy Group has built an AI training center using Huawei's Pangu model. The technological innovations they are making have been attributed to the extensive use of AI across Shandong Energy Group's coal mines, which has led to more efficient, safe, and sustainable coal mine production operations and management. This has also reduced the number of miners needed to work in hazardous conditions below ground.
All of these are success stories from Huawei's customers in 2023. Pioneering businesses across all industries are welcoming AI with open arms by converging industry know-how and innovative ideas with foundation model capabilities. The use of AI is not just limited to common work scenarios like marketing, design, and search. It is expanding into industrial production scenarios including public utilities, transportation, finance, and energy.
AI has already reshaped the tools of productivity. It is set to change the way society functions for the foreseeable future and become a new driver of production that will spur a qualitative change for society at large.
Computing power is crucial to broad AI adoption. At Huawei, we are aiming to make Ascend a leading AI computing platform with an innovative architecture that helps remove the barriers to large-scale computing power. To this end, we have integrated our strengths across multiple tech domains, from computing and data storage to networking, cloud, and digital power. Our Ascend AI cluster is the culmination of our integrated strengths. It delivers great performance and industry-leading reliability, and can support months of stable training. The Ascend AI cluster is already being used in some AI computing centers, which have even reached up to 400 petaflops.
Gone are the days when we had to rely on small, fragmented AI models that were trained on massive amounts of discrete data. An exciting era of foundation models, with a myriad of models and modalities, is right on the horizon.
Huawei has decided to pursue foundation models, providing support for their entire lifecycle, from training and development to inference and deployment. Our open Ascend computing platform is designed to do this by working with mainstream operators, frameworks, and acceleration libraries. On this platform, partners and developers can customize operators more efficiently – an ability that will be vital for speedy native innovation in foundation models.
At present, Ascend AI natively supports more than 30 L0 foundation models, including the well-known models like GPT-3, Llama 2, GLM, and iFlytek Spark. Huawei also provides a one-stop migration tool for existing foundation models. For a model with over 10 billion parameters, this migration tool can help developers migrate the model's scripts in just two days, and complete precision and performance optimization within two weeks. Huawei's Ascend inference engine can also connect to different frameworks' inference APIs and service systems to support online, distributed inference and fast application launch.
In tandem with our partners from the finance, telecom, Internet, and energy industries, to name just a few, we have also hatched a number of L1 industry-specific models and L2 scenario-based models. Moving forward, we will remain committed to openness and help customers develop their own industry-specific models and scenario-based models, turbocharging their intelligent upgrade.
A robust AI industry can only be built upon open architectures. Huawei has proposed a reference architecture for the intelligent transformation of industries. This open architecture embodies the extensive experience Huawei has gleaned from different industries, and stands out thanks to its synergies, openness, agility, and trustworthiness. This architecture can bolster the powerful AI platforms that smart applications need to power industrial intelligent transformation.
Additionally, a prosperous AI industry cannot exist without a dynamic ecosystem. This is why we at Huawei are working closely with industry partners and outside developers to create an open AI ecosystem for shared success.
A prime example of these efforts is our Ascend Partner Program. This program has brought together 1,200 partners (including more than 30 hardware partners) and 1.8 million developers to create over 2,500 industry-specific AI solutions. Going forward, we will redouble our efforts to foster a plethora of new AI professionals and promote collaboration between industry and academia, bridging research and application.
Returning to Mr. Harari, he also said, "Looking back, humanity will turn out to be just a ripple within the cosmic data flow." While many see this sentiment as disheartening, I don't believe it should be. History has proven time and again that we have what it takes to do better.
We are on the cusp of greatness with AI. It's time to look over the garden wall and try to catch a glimpse of what the future holds. At Huawei, our goal is crystal clear: becoming a preferred intelligent infrastructure partner for all industries in the AI era. To that end, we will enable a myriad of models and modalities, build an intelligent foundation, and foster a dynamic ecosystem to expedite intelligent transformation across industries.
