Since the Open Artificial Intelligence Research Center (OpenAI) in the United States released the chatbot program ChatGPT and its GPT series of big language models, the trend of AI big models has continued to spread. Global technology companies are competing to announce the development or launch of their own big models, and how to commercialize the rapidly growing number of big models has become a focus of industry attention. In addition to the business model adopted by companies such as OpenAI and Google that provides cloud based large model application programming interfaces (APIs), another highly anticipated commercialization path is to sell "software hardware integrated" large model training and promotion all-in-one machines. This business model is highly popular in the Chinese market, especially with many small and medium-sized technology enterprises launching large model training and promotion integrated machine products. This type of "out of the box" artificial intelligence product can be localized and deployed within enterprises, enabling the latest artificial intelligence technology to be easily applied in industrial scenarios, achieving the transformation from artificial intelligence technology to productivity, and accelerating the deep integration of artificial intelligence and industry. Large model training and inference integrated machine generally refers to a "full stack" artificial intelligence device that integrates large model training and inference functions. It usually includes software and hardware components such as central processing unit (CPU), graphics processing unit (GPU), memory, operating system, deep learning models, etc. It can master data processing and prediction abilities through large-scale data training. Compared to using large model functionality through application programming interfaces, enterprises have multiple advantages in deploying large models through a training and promotion integrated machine. Firstly, the training and promotion all-in-one machine can provide customized large model solutions that are ready to use out of the box for specific business needs of enterprises, and improve performance through software and hardware collaborative optimization, reducing the threshold for large model deployment, business solution construction, and adaptation. For example, Jiadu Technology has launched the "Jiadu Zhixing Transportation Industry Big Model Training and Promotion Integrated Machine" based on Huawei's "Ascend AI" basic software and hardware platform, which can provide intelligent, efficient and real-time solutions and decision support for key smart city construction fields such as rail transit, urban transportation, and urban safety. Secondly, due to the fact that the training and promotion all-in-one machine concentrates all computing resources on one device, it can significantly reduce data transmission time, improve computing speed and model training efficiency in large-scale data training, enabling enterprises to use and manage resources more quickly. In addition, the integrated training and promotion function deployed locally ensures data privacy and model security to the greatest extent possible, meeting the needs of enterprises for data security supervision. The Financial Times recently reported that the deployment of large model training and promotion integrated machines within companies is impacting the public cloud based artificial intelligence services provided by some large technology companies. The report states that the recent exposure of security vulnerabilities in artificial intelligence enterprises has intensified user concerns about data security. For example, ChatGPT shares user search history with others, and Samsung employees in South Korea accidentally leak company trade secrets while using ChatGPT to handle work. The report quotes Liu Qingfeng, Chairman of iFlytek in China, as saying that institutions need to be able to protect their data, and establishing a private cloud is a way to prevent valuable data leakage. But there are also