BloombergGPT: Introducing our 50-Billion Parameter Large Language Model for Financial Purposes


Bloomberg LP has announced a ground-breaking research paper, introducing BloombergGPT, an advanced artificial intelligence (AI) model purpose-built from scratch for the financial industry. This large language model (LLM) was created to assist with a range of natural language processing (NLP) tasks within the financial domain.

This breakthrough research was led by Bloomberg’s ML Product and Research group in collaboration with the company’s AI Engineering team. For the training of BloombergGPT, the team used a mixed approach to combine data from both the finance industry and general-purpose datasets. This resulted in a training corpus of over 700 billion tokens and a 50-billion parameter decoder-only causal language model.

Michael Feeney, Head of AI Research at Bloomberg said: “The complexity and unique terminology of the financial domain require a custom AI language model. With BloombergGPT, we are excited to push the boundaries and help unlock new opportunities for our customers. This model makes it possible to bring the full potential of AI to the financial domain and to better marshal the vast amounts of information available on the Bloomberg Terminal.”

BloombergGPT outperforms existing open models of a similar size on financial tasks, while also performing on par or better on general NLP benchmarks.

Bloomberg is no stranger to the application of AI, Machine Learning, and NLP in finance. Bloomberg has long been leading the way in terms of technology, having supported a wide variety of NLP tasks, such as sentiment analysis, named entity recognition, news classification, and question answering, to name a few. Now, with the introduction of BloombergGPT, the field of AI within the finance industry has taken yet another giant leap forward.