Preferred Networks, Inc. began offering a derivative model of the large-scale language model (LLM) “PLaMo,” called “PLaMo-fin-base,” which enhances Japanese financial knowledge, to financial institutions. PFN’s financial team will provide total support to financial institutions, from identifying issues, verifying technologies and use cases, to developing and operating applications, in order to improve the accuracy of various processes that require natural language processing technology and increase operational efficiency.
The PFN Group developed PLaMo from scratch, starting with pre-learning, without using any existing LLMs. The strength of this is that it can flexibly design systems appropriate for business processes, develop additional learning and functions required for LLMs specialized in specific fields, using the LLM development know-how and models gained from this.
PLaMo-fin-base is based on PLaMo, which has world-class Japanese language capabilities, and PFN’s finance team has additionally trained it with a large amount of Japanese language data from the Japanese financial sector, further improving its ability to answer tasks that require domestic financial knowledge. For example, by using PLaMo-fin-base as the LLM at the core of the AI agent, it can be used to streamline and enhance a wide range of operations at banks and securities companies, such as drafting proposals based on sales daily reports, creating investment and loan approval documents, role-playing counter and corporate sales, analyzing companies based on IR information, and summarizing information issued by regulatory authorities.
PLaMo-fin-base has received high praise in the Japanese Language Model Financial Evaluation Harness, a Japanese language benchmark in the financial field, and has achieved high performance in tasks that test the required skills of financial professionals such as securities analysts and accountants. is possible to use and link internal data and know-how, and it is possible to use unique internal files such as daily work reports, training materials, manuals, and customer databases through RAG (Search Augmentation Generation). In addition, it is possible to reflect unique business know-how, such as criteria for valuation in investments, in the model itself through additional learning.
It is designed for use in financial institutions that require extremely high security standards, and can be used in an on-premise environment where no data is allowed outside the company. PFN’s financial team, which has been highly praised by customers and at academic conferences both in Japan and overseas, will support practical application. In addition, the company has an extensive track record of developing and providing solutions based on cutting-edge machine learning and deep learning technologies, such as parameter tuning through additional learning of LLM.
PFN explains that it is continuing to develop LLMs specialized for the financial sector, aiming to develop lightweight models that can run on edge devices such as PCs, and to improve safety and performance. It also explains that it is developing AI agents that utilize these, and will further enhance its LLM utilization support services for financial institutions.
