IRP-ICBC（Intelligent Research Platform of Industrial and Commercial Bank of China）
Employees in financial institutions who are engaged in industry research and strategic planning often face issues such as scattered information resources, complex research tools, and data silos. Common general Large Language Models(LLMs) on the market cannot be directly applied to the finance sector, which is knowledge-intensive and requires high professional standards. Therefore, this project has trained a specialized LLM tailored for the financial research field and established the ICBC Research Intelligence Platform. The platform assists research analysts by applying LLMs technology to scenarios such as research report analysis and writing, industry risk assessment, data-driven decision-making, and business management planning. It helps users quickly access information, improve research and decision-making efficiency, and acts as a "second brain" for financial analysts.
Platforms And End-to-End Solutions For Government And Public Administration
Intelligent decision-making systems
Two years
Within China and some foreign countries
1. The first LLM application in the industry specifically for financial research in large commercial banks.This project is based on the company's self-developed, fully controllable, trillion-parameter-scale financial large model. Leveraging years of accumulated financial research data and business data, it creates a specialized large model dedicated to financial research. 2. Comprehensive coverage of financial data. The platform aggregates major research outputs from leading research institutions in the market, integrates 40 years of TB-scale data accumulated within the company, and includes high-quality financial industry data (such as news, research reports, announcements, and structured data). The pre-training financial corpus reaches hundreds of billions of tokens. It can comprehensively understand financial terminology, accurately grasp user intentions, and provide responses that are both professional and compliant. 3. Optimal use of AI computing power. A heterogeneous cluster with thousands of GPUs has been constructed to support model training, capable of completing secondary training of trillion-parameter models within six weeks. At the inference level, it can support over 1,800 query requests per second. 4. Content safety and compliance. The platform employs advanced information technology to ensure user data security and protect user privacy. By enhancing the security access mechanism of foundational large models and uniquely creating an automated large model security evaluation tool, it ensures that training datasets meet graded standards through sensitive word detection mechanisms, improving the security of large model outputs and continuously enhancing model safety. 5. Customized services. The platform can provide personalized information and research reports based on user subscriptions, making it the most understanding and considerate assistant for user needs.
N.A.
N.A.
