Flo Health fine-tunes open models with AI judges and Databricks AI Runtime
Overview
Flo Health describes a post-training pipeline for fine-tuning open models on Databricks AI Runtime using serverless NVIDIA H100 GPUs, according to a Databricks blog post.
The pipeline uses 26 AI judges that score answers on medical safety and usefulness, and it generated more than 15,000 synthetic questions to train the models without human-annotated responses.
Flo's team estimates that tailored fine-tuned models can cut inference costs by an order of magnitude compared with frontier model APIs. That estimate comes from Flo and Databricks, and the post does not include measured cost figures.
Written by AI from the articles below · updated Oct 12, 12:52 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Databricks BlogOfficialFlo Health fine-tunes open models with Databricks AI Runtime for safer health answers
AIFlo Health built fine-tuned open models using AI judges, privacy-first synthetic data, and Databricks AI Runtime on serverless NVIDIA H100 GPUs. The team estimates tailored fine-tuned models can cut inference costs by an order of magnitude compared with frontier model APIs. Flo's pipeline uses 26 judges across medical-safety and usefulness criteria, and it generated more than 15,000 synthetic questions to train models without human-annotated responses.
Heat trend
Not enough continuous observations to show a trend yet.