[Beijing, China, March 22nd, 2024] Artificial intelligence frameworks, as foundational software technologies, have emerged as the driving force behind the rapid advancement of AI models and the growth of industrial intelligence. The MindSpore Conference 2024, centered around the theme "Upgrading for Wisdom: The Source of Thinking and Creation", took place at the Beijing National Convention Center. The summit aimed to unite the innovative energies of the AI industry, promote ongoing advancements in foundational technology, and foster the creation of a new open-source ecosystem for artificial intelligence. iSoftStone played a significant role in the summit as a co-organizer, deeply engaging in the event's proceedings.

Throughout the conference, the third closed-door meeting of the MindSpore Open Source Community Council took place. Xie Rui, Senior Vice President of iSoftStone and General Manager of Digital Infrastructure and Integration, participated as a representative of the Council members, engaging with industry partners to explore the future development trends of MindSpore.

During the meeting, Shan Jiling, Director of Digital Infrastructure and Integration Technology at iSoftStone, delivered a keynote speech titled "Empowering Enterprise Intelligence with the iSoftStone Training and reasoning Integration Platform". In his address, he emphasized how this platform facilitates customers in reaching the critical "last mile" of large model scene reasoning through an integrated machine.

The MindSpore Conference 2024 serves as a pivotal platform for industry exchanges and collaboration. At the conference, iSoftStone unveiled two groundbreaking innovations, namely the "Integrated AI Training and Reasoning Platform" and the "Tianxuan 2.0 MaaS Platform",i in the summit's exhibition area. These unveilings sparked interactive engagements and partnerships with numerous industry stakeholders.

iSoftStone's Integrated AI Training and Reasoning Platform represents the pivotal last mile of AI application

Artificial intelligence stands as a strategic technology shaping the future, profoundly catalyzing societal and industrial transformations. As a foundational software technology, the artificial intelligence framework assumes a critical role in the evolution of AI. Against this backdrop, iSoftStone introduced an Integrated AI Training and Reasoning Platform . This platform integrates components such as the Ascend AI Basic Hardware Platform and iSoftStone Enterprise OS (ISSEOS). Moreover, it incorporates the Tianxuan 2.0 MaaS platform, facilitating seamless AI development, integration of industry-specific AI models, and deep adaptation to diverse enterprise AI application scenarioses. The iSoftStone's Integrated AI Training and Reasoning Platform supports the MindSpore comprehensive AI framework. It features user-friendly design, efficient runtime, and flexible deployment capabilities in both user and operational states, providing users with a development experience that is both user-friendly and operationally efficient. Presently, the iSoftStone's Integrated AI Training and Reasoning Platform caters to a wide array of clientele, including central state-owned enterprises, scientific institutions, educational entities, and financial institutions, delivering tailored services to meet diverse customer needs.

Despite the widespread application of large models across various sectors, certain challenges persist, including accuracy limitations, inadequate integration with business systems, data security apprehensions, and high computing costs. To enhance the practical application of large models in business settings, iSoftStone partnered with Baichuan AI to introduce the " iSoftStone-Baichuan AI Large Model Integrated Machine ". Liu Ya, Vice President of iSoftStone and Deputy General Manager of Digital Infrastructure and Integration, along with Shan Jiling, Director of Digital Infrastructure and Integration Technology, attended the launching ceremony. Li Jian, General Manager of Production and Research at Baichuan AI Government, Enterprise, and Public Institutions Group, and Xie Nan, Head of Government, Enterprise, and Public Institutions Group Integrated Machine, also attended the event.

The integrated machine, an independent and innovative Integration of software and hardware , is built upon the foundational hardware platform of Ascend AI. It comes pre-loaded with Baichuan's powerful large model capability and boasts a wide array of functionalities including multi-round dialogue, report generation, content creation, code generation, knowledge retrievals, text classification, API platform integration, and enterprise knowledge base support. In numerous complex customer business scenarioses, it proves invaluable in achieving the crucial "last mile" of large model scenario reasoning.

[Tongfang AI Computing Power Infrastructure]

The iSoftStone-Baichuan AI Large Model Integrated Machine , anchored on the Ascend AI hardware foundation, offers several distinct advantages such as high energy efficiency, robust computing power, exceptional reliability, straightforward management, and seamless expandability. Users can select from two hardware options as per their requirements: the Tongfang Super K620 and the Tongfang Super A800I A2 , both of which incorporate the iSoftStone Enterprise OS (ISSEOS) and find wide applications in AI algorithm development and AI reasoning service scenarioses. Core Strengths:

■ Superb Computing Power: The integrated machine offers leading performance, accelerating AI reasoning efficiency. A single reasoning server can achieve an impressive 2.2 PFLOPS of AI computing power.

■ Ultra-High Energy Efficiency: Equipped with a built-in high-computing AI chip and a high-performance operator library, it supports multi-card high-speed interconnection and distributed storage, ensuring optimal energy utilization.

■ Flexible Expansion : Storage, computing power, and servers can be easily scaled up at any time to align with evolving business demands.

■ Secure Deployment: With an integrated intranet deployment approach, data remains within the confines of the computer room, ensuring its safety and reliability. This approach establishes a trustworthy foundation from the outset.

[Baichuan Base Large Model]

The iSoftStone-Baichuan AI Large Model Integrated Machine utilizes the comprehensive MindSpore AI development framework for full scene enhancement. With the pre-installed Baichuan large model, enterprises can swiftly engage in large-scale model scenario reasoning, fine-tune based on the reasoning outcomes, and enhance model prediction accuracy. This solution addresses challenges such as rapid deployment of large enterprise models, high computing power costs, and data security concerns.

■ Extensive Base Large Model: Equipped with built-in Baichuan2 large models ranging from 3B, 7B, 13B, to 33B parameters, enabling support for a 192K long window input, accommodating over 350,000 words at once. It also leverages cutting-edge Transformer calculation engineering optimization for superior performance.

■ Enhanced Search Capabilities + Knowledge Base: Employs enterprise knowledge base integration alongside real-time information acquisition through enhanced search functionality; Utilizes industry-leading Embedding models and achieves multi-channel recall capability through sparse retrievals + vector retrievals techniques; Enables online search and seamless integration with the internet for comprehensive data access.

■ Lifecycle Services: Offers installation and deployment, model deployment, reasoning services, and other support throughout the product’s lifecycle, ensuring high-quality and efficient technical assistance.

As artificial intelligence continues to advance and become more widespread, large model integrated machines will see increased application across various fields. Moving forward, iSoftStone, alongside partners like Huawei and Baichuan AI, will continue leveraging its strengths, driving innovation, and achieving breakthroughs to further contribute to the development of artificial intelligence technology and the industry's prosperity.