BloombergGPT
lastmod 2023-04-03created 2023-04-031 edits
A large language model for finance
Features
- A 50-billion-parameter (50B) language model trained on a wide range of financial data
- Built a 363-billion-token dataset based on Bloomberg's data sources, augmented with a 345-billion-token general dataset
- No performance loss on general LLM benchmarks, and far better performance than existing models on financial tasks
- Also describes the modeling choices, training process, and evaluation methodology
- As a next step, plans to release a training log (chronicle) detailing the experience of training BloombergGPT
See also
Documentation
- BloombergGPT - A Large Language Model for Finance
- https://arxiv.org/abs/2303.17564