FactSet moves Text-to-Formula model fine-tuning to Databricks AI Runtime, cutting setup to hours

FactSet says it cut the time to set up, train and deploy its custom Text-to-Formula model from days to hours by running fine-tuning on Databricks AI Runtime's serverless GPU infrastructure.

The account comes from a first-party Databricks blog post, published October 14, 2026. Databricks says engineers can choose compute for each training run without first provisioning and configuring the GPU environment.

FactSet retrains its models roughly every two to three weeks, according to the post, and has worked with Llama-based models while evaluating newer open models. Training runs inside its existing Lakehouse workflow, with MLflow tracking experiments and governed data.