科蓝软件：锚定AI金融底座，仓颉编程+昇腾算力 双擎驱动 技术降维打击，昇腾算力定义金融AI新标准 硬核布局：#2024年8月与华为正式签署合作协议共同建设智能算力中心，深化昇腾算力集群与金融大模型领域合 作。 技术壁垒：突破传统x86算力依赖，通过昇腾AI处理器的国产化高性能计算，实现从"算力租借"到"算力自主"的跨 越。 卡位国产AI编程语言，重塑金融信创底层逻辑 #公司作为仓颉语言首批生态合作伙伴，自2022年起与华为深度共建，正式进军国产编程语言蓝海。 在"全栈信创"时代来临之际，科蓝软件不再满足于传统Java/C++的"卡脖子"风险，而是通过仓颉（Cangjie）这一华 为为鸿蒙量身打造的全场景智能应用编程语言，实现从芯片、操作系统、中间件、数据库到编程语言的全链路自 主可控。 我们将金融系统开发从"被动适配"升级为"原生重构"，解决金融机构在信创浪潮中的"底层安全"难题，构建数字金 融的原生安全宇宙。 卡位国产AI编程语言，重塑金融信创底层逻辑 #公司作为仓颉语言首批生态合作伙伴，自2022年起与华为深度共建，正式进军国产编程语言蓝海。 在"全栈信创"时代来临之际，科蓝软件不再满足于传统Java/ ...

Group 1 - The core importance of "cloud + AI" is increasingly highlighted as AI technology accelerates its large-scale deployment, shifting from model training to practical application [1] - AI cloud serves as a critical infrastructure for AI training and deployment, providing elastic computing power and efficient platforms, thus creating significant market growth opportunities [1] - Major tech companies are upgrading technologies and forming cross-industry collaborations to promote the deep integration of "cloud + AI" into sectors like robotics [1] Group 2 - Cloud service providers are transitioning from merely providing resources to becoming comprehensive AI empowerment platforms, enhancing their foundational capabilities [2] - Kingsoft Cloud has upgraded its platform to a one-stop AI training and deployment solution, covering a complete system from resource management to commercial model API services [2] - Alibaba Cloud and Huawei Cloud are collaborating with hardware manufacturers to optimize efficiency in large-scale inference scenarios, lowering the barriers for SMEs to access computing power [2] Group 3 - The industry is addressing common pain points such as fragmented R&D processes and high costs by building more open ecosystems [3] - The integration of cloud collaboration is crucial for the robotics industry, with Kingsoft Cloud launching a robot platform that supports various core functions from data collection to deployment [4] - The platform supports nearly 40 different models, enhancing the capabilities of robotics in real-world applications [4] Group 4 - Several listed companies are collaborating with cloud enterprises in the field of embodied intelligence, promoting the development and deployment of intelligent robots in various industries [5] - The trend of collaboration between terminal and cloud service companies is expected to accelerate the intelligent upgrade of the robotics industry [5] Group 5 - The AI cloud market is projected to exceed 100 billion yuan, with the demand for inference experiencing explosive growth [6] - By mid-2025, the Chinese AI cloud market is expected to reach 22.3 billion yuan, with a projected growth of 148% by 2025, reaching 193 billion yuan by 2030 [6] - The competition in the "cloud + AI" space is evolving from a focus on computing power to a comprehensive ecosystem involving computing power, models, data, and application scenarios [6]

Industry Insights - The Ministry of Commerce of China is expediting the review of export license applications related to rare earths, emphasizing the importance of maintaining global supply chain stability and security [1] - The Huawei Developer Conference is taking place from June 20 to 22, showcasing innovations in HarmonyOS and AI, with a focus on Ascend computing power and the Harmony ecosystem [2] - The photovoltaic industry is expected to see a significant production cut in Q3, with operating rates projected to decrease by 10-15%, alongside strict audits on below-cost sales [4] - The gaming industry in Beijing is receiving support through 11 new measures aimed at fostering development, indicating a high level of industry vitality [7] - UTree Technology has confirmed the completion of its Series C financing, with a pre-investment valuation exceeding 10 billion yuan, signaling growth potential in the humanoid robotics sector [8] - The International Solid-State Battery Technology Conference has commenced, highlighting advancements in materials and smart equipment, with expectations for rapid growth in the solid-state battery market by 2027 [9] Macro Insights - China's Vice Premier He Lifeng stated that the country is an ideal, safe, and proactive investment destination for multinational companies, citing its large domestic market and robust industrial system [3]

Core Viewpoint - The rapid development of large language models (LLMs) has become a cornerstone of general artificial intelligence systems, but the increase in model capabilities has led to significant growth in computational and storage demands, presenting a challenge for achieving high performance and efficiency in AI [1][2]. Group 1: Technological Advancements - Huawei's Noah's Ark Lab has developed the Pangu Ultra, a general language model with over 100 billion parameters, surpassing previous models like Llama 405B and Mistral Large 2 in various evaluations [2]. - The lab also introduced the sparse language model Pangu Ultra MoE, achieving long-term stable training on over 6000 Ascend NPUs [2]. Group 2: Key Research Presentations - A series of sharing sessions from May 28 to May 30 will cover breakthroughs in quantization, pruning, MoE architecture optimization, and KV optimization, aimed at developers and researchers interested in large models [3][4]. Group 3: Specific Research Contributions - **CBQ**: A post-training quantization framework that addresses the high computational and storage costs of LLMs, achieving significant performance improvements in ultra-low bit quantization [6]. - **SlimLLM**: A structured pruning method that effectively reduces the computational load of LLMs while maintaining accuracy, demonstrating advanced performance in LLaMA benchmark tests [8]. - **KnowTrace**: An iterative retrieval-augmented generation framework that enhances multi-step reasoning by tracking knowledge triplets, outperforming existing methods in multi-hop question answering [10]. Group 4: Further Innovations - **Pangu Embedded**: A flexible language model that alternates between fast and deep thinking, designed to optimize inference efficiency while maintaining high accuracy [14]. - **Pangu-Light**: A pruning framework that stabilizes and optimizes performance after aggressive structural pruning, achieving significant model compression and inference acceleration [16]. - **ESA**: An efficient selective attention method that reduces computational overhead during inference by leveraging the sparsity of attention matrices [18]. Group 5: MoE Model Developments - **Pangu Pro MoE**: A native MoE model with 72 billion parameters, designed to balance load across devices and enhance inference efficiency through various optimization techniques [21]. - **PreMoe**: An expert routing optimization for MoE models that allows dynamic loading of experts based on task-specific requirements, improving inference efficiency by over 10% while maintaining model capability [24]. Group 6: KV Optimization Techniques - **KVTuner**: A hardware-friendly algorithm for KV memory compression that achieves near-lossless quantization without requiring retraining, significantly enhancing inference speed [26]. - **TrimR**: An efficient reflection compression algorithm that identifies redundant reflections in LLMs, leading to a 70% improvement in inference efficiency across various models [26].