Snowflake Data Clean Rooms ML Jobs Now Generally Available
Snowflake's ML Jobs feature is now generally available within Data Clean Rooms, allowing data scientists to run complex Python ML workloads with distributed training and GPU compute. This addresses previous limitations of SQL-only or single-node Python in clean rooms, enabling multiparty model training and scoring without raw data leaving an organization's account. The feature is available to engineers and data scientists who work with sensitive, cross-organizational data in secure collaboration environments.
- →Multiparty ML model training and scoring in Data Clean Rooms
- →Use case: Incrementality measurement without data intermediaries
- →Foundation for training AI agents across distributed data
- →Improved accuracy through multi-source features for ML models
- →Simplified ML workflow development and deployment
Features (3) ›
- Multiparty ML model training and scoring in Data Clean Rooms
ML Jobs in Snowflake Data Clean Rooms is now generally available, enabling data scientists to bring standard Python ML stacks, including distributed training, custom packages, and GPU compute, directly into multiparty collaborations. Models can train on combined data from multiple organizations without raw records leaving their respective accounts, and pipelines run automatically.
- Use case: Incrementality measurement without data intermediaries
ML Jobs enables advertisers to measure the true sales lift from advertising by joining ad exposure data with purchase outcomes across party lines within Data Clean Rooms. This removes the historical need for a neutral third party to hold combined data or extensive custom infrastructure, making cross-party data analysis more accessible.
- Foundation for training AI agents across distributed data
The ML Jobs infrastructure supports the training of AI agents whose effectiveness relies on signals distributed across multiple organizations. This includes combining behavioral data, conversion history, and demographic enrichment within a governed clean room for fine-tuning runs, representing a future direction for multiparty AI.
Enhancements (2) ›
- Improved accuracy through multi-source features for ML models
By combining distinct signals about consumer behavior from multiple organizations, such as purchase history, transaction patterns, and engagement data, ML Jobs allows for the creation of more predictive models than those trained on single data silos. This facilitates richer feature sets for tasks like propensity scoring, leading to more accurate results.
- Simplified ML workflow development and deployment
Data scientists can use their standard Python development environments and IDEs, with scaling to multiple nodes or GPUs managed via parameter changes. The process avoids manual infrastructure provisioning, container registry configuration, or Docker image building. Workloads can be tested outside clean rooms before deployment and can be scheduled or triggered by upstream events for operational pipelines.
https://www.snowflake.com/content/snowflake-site/global/en/blog/ml-jobs-snowflake-data-clean-rooms