Huawei Cloud Ascend AI Cloud Service Officially Launched, Allowing Thousands of Industries to Achieve Inclusive Benefits of Large Models

On September 21, at the Huawei Full Connection Conference 2023, Huawei Cloud CEO Zhang Pingan announced that the Ascend AI cloud service has been officially launched. Enterprises can now access powerful AI computing power with one click, allowing all walks of life to benefit from the popularity of large-scale models. At the same time, the "Hundred Models and Thousands of Modes" area of Shengteng AI Cloud Service has also been officially launched, providing enterprises and developers with faster creation of large model applications, allowing them to maintain their leading position in the era of large models

Zhang Pingan pointed out that in the era of full intelligence, the demand for AI computing power has increased dramatically. However, the power density of AI servers far exceeds that of general servers. The power consumption of a single cabinet is 6-8 times that of the past, and a specialized liquid cooling system is required to dissipate heat. At the same time, the training of large-scale models requires the support of hundreds of calories, thousands of calories, or even ten thousand cards. The interconnection, delay, reliability, and operation and maintenance between servers are very complex. Self-built AI data centers face many challenges

In order to better support "hundreds of models and thousands of states", Huawei Cloud has deployed three large-scale artificial intelligence computing centers in Gui'an, Ulanqab, and Wuhu to provide enterprises with powerful artificial intelligence computing power. Enterprises can use artificial intelligence cloud services at any time with just one click. They no longer need to spend time and energy building or transforming traditional data centers

To protect user data, Huawei Cloud Ascend AI cloud services use data Multiple technologies such as transmission encryption, data storage encryption, data security clearing, data access control, and data watermark leakage prevention ensure the security of the entire life cycle of data trained on the cloud.

At the same time, the "Hundred Modes and Thousands of States" zone of Huawei Cloud Ascend AI Cloud Service has been launched. In addition to supporting the Pangu large model, it will also adapt to hundreds of mainstream open source large models in the industry, such as LLAMA, GLM, etc. Enterprises can use Pangu's L0 and L1 layer capabilities to build their own large model applications, or they can use the industry's open source large model technology to create exclusive large models

In addition, Huawei Cloud Ascend AI Cloud Service also provides The efficient, long-term and stable large-model training environment and complete tool chain enable 30 consecutive days of uninterrupted Kcal training, and task recovery time is less than 30 minutes; and more than 30 visual tuning and deployment tools can help enterprises process data more efficiently. processing, model fine-tuning, prompt engineering, etc., to complete the end-to-end development of an industry model with 100 billion parameters, which can be shortened from the past 5 months to the current 1 month, and the overall speed has been increased by 5 times.

The above is the detailed content of Huawei Cloud Ascend AI Cloud Service Officially Launched, Allowing Thousands of Industries to Achieve Inclusive Benefits of Large Models. For more information, please follow other related articles on the PHP Chinese website!