Home  >  Article  >  Technology peripherals  >  Huawei Cloud Pangu Large Model 5.0 officially released: upgraded multi-modal capabilities, providing billions to trillions of different parameters

Huawei Cloud Pangu Large Model 5.0 officially released: upgraded multi-modal capabilities, providing billions to trillions of different parameters

WBOY
WBOYOriginal
2024-06-22 02:28:50639browse

According to news from this site on June 21, at the Huawei HDC 2024 Developer Conference held today, Huawei Managing Director and Huawei Cloud CEO Zhang Pingan announced that Pangu Model 5.0 was officially released, with three major features: full series, multi-modal, and strong thinking. Achieving upgrades.

  • Full series: Pangu Large Model 5.0 includes models with different parameter specifications to adapt to different business scenarios.

  • Multi-modal: Pangu Large Model 5.0 can better and more accurately understand the physical world, including text, pictures, videos, radar, infrared, remote sensing and other modalities. In terms of generation, Pangu 5.0 can generate multi-modal content that conforms to the laws of the physical world, allowing innovation to be arbitrary.

  • Strong thinking: Complex logical reasoning is the key for large models to become industry assistants. Pangu Large Model 5.0 deeply combines thinking chain technology and strategic search technology, greatly improving mathematical capabilities and complex task planning capabilities.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数

In terms of the full series, Pangu Large Model 5.0 includes different parameter scales such as billions, tens of billions, hundreds of billions, trillions, etc., and provides Pangu natural language large models and multi-modal large models. , large visual models, large prediction models, large scientific computing models, etc.

  • Pangu E series: 1 billion parameter level, can support mobile phones, PCs and other end-side smart applications;

  • Pangu P series: 10 billion parameter level, suitable for low-latency, low-cost inference scenarios;

  • Pangu U series: There are 135 billion parameters and 230 billion parameters, suitable for processing complex tasks, and can become the base of enterprise general large models;

  • Pangu S series: Trillion-level parameters, can handle cross-border Multi-domain tasks can help enterprises better apply AI technology in all scenarios.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数

Pangu Large Model 5.0 brings three major innovative upgrades, including major upgrades in multi-modal interaction, such as the addition of the ability to understand 10K ultra-high-resolution images.

In terms of content generation, Pangu Large Model adopts the industry's first STCG technology, focusing on multiple industry scenarios such as autonomous driving, industrial manufacturing, and construction, and can generate multi-modal content that is more in line with physical laws.

In addition, Huawei also cooperated with Leju (Shenzhen) Robot Technology Co., Ltd. to develop a large model of Pangu’s embodied intelligence. According to reports, Huawei's Pangu embodied smart large model has the ability to have one brain and multiple shapes, enabling billions of end-side devices to have independent intelligence.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数
▲ Huawei Pangu Embodied Intelligent Large Model Live Demonstration

At the meeting, Huawei also let the Pangu Embodied Intelligent Large Model perform an operation demonstration. With the help of the Pangu large model, Leju "Kuafu" can smoothly complete interactive demonstrations such as identifying objects, Q&A interactions, high-fives, and handing out water, which visually demonstrates the results of the cooperation between the two parties based on the Pangu large model.

The "Kuafu" humanoid robot can complete more than 10 steps of complex task planning, and achieve multi-scenario generalization and multi-task processing during task execution. At the same time, the Pangu large model can also generate training videos needed by the robot, allowing the robot to learn various complex scenarios faster.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数
▲ Leju “Kuafu” humanoid robot

Zhang Ping’an said that in addition to humanoid robots, the Pangu embodied intelligent large model can also empower various forms of industrial robots and service robots, allowing them to help humans To do dangerous and strenuous work. "As everyone expected, letting AI robots help us wash, cook, and sweep the floor gives us more time to read books, write poems, and paint."

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数

According to reports, in the past year By mid-year, Huawei Cloud Pangu large model has been implemented in more than 30 industries and more than 400 scenarios, in fields such as government affairs, finance, manufacturing, pharmaceutical research and development, coal mines, steel, railways, autonomous driving, industrial design, architectural design, meteorology, etc. of great value.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数

This site noticed that with the support of Pangu Big Model 5.0, Huawei has also launched a new Xiaoyi Assistant, which makes ice hockey more intelligent, smoother, and more user-friendly, and can make your game according to the weather conditions outside. Travel methods can also help you protect yourself, and Xiaoyi can also become more suitable for your usage habits through continuous learning.

华为云盘古大模型 5.0 正式发布:升级多模态能力,可提供十亿到万亿级不同参数

In addition, Xiaoyi smart body is integrated with the navigation bar, everywhere and ready to be summoned. Just feed text, pictures, and documents to Xiaoyi, and you can process text, recognize images, and analyze documents conveniently and efficiently.

Huawei HDC 2024 Developer Conference Special Topic

The above is the detailed content of Huawei Cloud Pangu Large Model 5.0 officially released: upgraded multi-modal capabilities, providing billions to trillions of different parameters. For more information, please follow other related articles on the PHP Chinese website!

Statement:
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn