Amazon SageMaker has officially launched a fully managed MLflow capability, making it easier for machine learning teams to manage the entire ML lifecycle. This new release allows customers to effortlessly configure and manage MLflow tracking servers, streamlining the process and increasing productivity.
MLflow is a popular open-source tool that data scientists and machine learning developers use to track multiple attempts to train models, compare runs, evaluate models, and record the best models in a Model Registry. With Amazon SageMaker’s fully managed MLflow capability, ML administrators can quickly establish secure and scalable MLflow environments on AWS.
The core components of Managed MLflow in SageMaker include the MLflow Tracking Server, the MLflow backend metadata store, and the MLflow artifact store. These components allow users to efficiently track machine learning experiments, maintain metadata related to experiments, and store artifacts securely.
By using Amazon SageMaker with MLflow, users can streamline their machine learning workflows, monitor experiments across various platforms, leverage all of MLflow’s capabilities, and unify model governance. Users can also efficiently manage tracking servers, improve security, and effectively monitor activity on these servers to support governance.
To get started with MLflow in SageMaker, users need to create a SageMaker Studio domain, configure the IAM execution role, and create the MLflow tracking server. Tracking runs, comparing training runs, and registering candidate models with MLflow are essential steps in the process. Users can also clean up tracking servers to optimize costs.
The SageMaker with MLflow capability is now generally available in all AWS Regions where SageMaker Studio is available, except China and US GovCloud regions. Users are encouraged to explore this new capability to experience the improved efficiency and control it brings to their machine learning projects.
For more information, users can visit the SageMaker Developer Guide and provide feedback to AWS re: Publishing or their AWS support contacts.
Article Source
https://aws.amazon.com/blogs/aws/manage-ml-and-generative-ai-experiments-using-amazon-sagemaker-with-mlflow/