Databricks AI Runtime supports tabular foundation model experimentation and fine-tuning
Understand this story
Databricks has published a technical article stating that its AI Runtime now supports experimentation and fine-tuning for tabular foundation models (TFMs), and provides managed GPU infrastructure, integrated development tools, and model governance. According to Databricks, TFMs are pretrained on millions of synthetic datasets and make predictions via in-context learning in a single forward pass, without training on a specific dataset. The code repository accompanying the article demonstrates zero-shot inference running multiple TFMs on AI Runtime, as well as a complete fine-tuning example using KumoTabular for classification and regression.
AI overview based on source reports · Updated 39 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Databricks BlogTabular foundation models on Databricks AI Runtime
This article explains how tabular foundation models (TFMs) are pretrained on millions of synthetic datasets so they can make predictions on new tables via in-context learning without task-specific training, covering tokenization approaches, the row-compression architecture used by TabICLv1 and later models, and a comparison of recent models by scale and license. It discusses GPU memory and provisioning challenges for TFM inference and describes how Databricks AI Runtime provides managed serverless GPU infrastructure, MLflow tracking, Unity Catalog governance and Model Serving for TFM experimentation and fine-tuning.
Activity for this story
Not enough continuous observations to show a trend yet.