Case Study: GumGum accelerates LM fine-tuning with SuperAnnotate and Databricks

How GumGum fine-tunes LMs with SuperAnnotate and Databricks

GumGum, a leader in contextual intelligence for digital advertising, needed a faster, more reliable way to label complex multimodal data and keep its language models up to date as taxonomies and market conditions changed. The company used SuperAnnotate alongside Databricks to improve data annotation quality and streamline the workflow from labeling to model training.

With SuperAnnotate’s pre-labeling, human review, and Explore tools integrated into GumGum’s Databricks-based pipeline, the team could curate data more efficiently, support active learning, and fine-tune models faster. SuperAnnotate helped GumGum achieve a 10-point increase in F1 score for labeled data and significantly reduced the time needed to prepare training datasets, improving overall iteration speed and model performance.

GumGum

Iris Fu

Director of Engineering