Growing a stronger community of financial institutions blended with academia to nurture the creation of truly open source AI
Why does it matter for the industry?
The financial services industry currently lacks essential tools needed to foster the adoption of Large Language Models (LLMs). These include: comprehensive benchmarking to evaluate LLM performance, a secure data contribution framework, access to the actual data used for training and fine-tuning the open models, and orchestration tools that make LLMs usable as services. FINOS plays a unique role in addressing industry gaps by identifying common challenges across its membership and turning them into pre-competitive use cases, which are tackled during LLM exploration activities.
OVERVIEW
key areas of progress include:
Benchmark: Addressing a critical gap in evaluating LLM performance across diverse, real-world financial tasks. Systematically assesses LLMs on 35 datasets spanning 23 tasks, revealing their strengths and limitations in finance-specific applications like stock trading, numerical reasoning, and forecasting
CDM/DRR Use Case: Implementing retrieval-augmented generation (RAG) techniques for the CDM documentation. This enhancement significantly accelerates the process of mapping new asset classes to CDM. If scaled to “enterprise-level”, this could enable the mapping of virtually any asset class globally into the CDM standard.
Data Contribution Framework: A comprehensive data contribution framework that financial institutions can follow to contribute data securely and efficiently to an open financial dataset. Currently Validating OS-Climate Data Request to onboard mortgage data relevant for the TechSprint use cases.