AWS Brings Managed Open Source MLflow to Amazon SageMaker

AWS services available since 2017 are the foundation of today's popular generative AI models.

Amazon SageMaker was released in 2017 and has been steadily improving since then. While much of the attention in the AWS gen AI world has been focused on Amazon Bedrock over the past year, Amazon SageMaker continues to offer a significant set of features.

Amazon SageMaker is an AWS service for managing the entire machine learning lifecycle, from building and training models to deploying and managing predictive models at scale. It provides customers with a managed environment and tools to build, train, and deploy machine learning and deep learning models. Hundreds of thousands of customers use Amazon SageMaker for tasks such as training popular AI models and deploying machine learning workloads. Amazon SageMaker is used as a service to help train Stability AI's Stable Diffusion, a machine learning (ML) framework that helped power Luma's Dream Machine text-to-video generator.

AWS is further expanding the capabilities of its SageMaker service with the general availability of Managed MLflow. MLflow is a popular open-source platform for the machine learning lifecycle, including experimentation, reproducibility, deployment, and monitoring of machine learning models. With the launch of Managed MLFlow for Amazon SageMaker, AWS is giving users even more power and choice to build the next generation of AI models.

“Given the current pace of innovation in this space, our customers want to move from experimentation to production quickly and really accelerate their time to market,” Ankur Mehrotra, director and general manager of Amazon SageMaker at AWS, told VentureBeat. “So we're releasing MLflow as a managed capability within SageMaker, which enables them to set up and launch MLflow within the SageMaker development environment with just a few clicks.”

What MLflow brings to AWS users

Developers and organizations widely use the open source MLflow project for MLOps, and Mehrotra emphasized that the new managed MLflow in SageMaker service gives enterprise users more choice without replacing existing capabilities.

By offering MLflow as a fully managed service tightly integrated with SageMaker, AWS aims to provide a unified experience leveraging the capabilities of both platforms.

“As you iterate on your model and create different variations, MLflow records those metrics and lets you easily track and compare different iterations, which is a great thing about MLflow,” Mehrotra says. “You can then register those models in the model registry and easily deploy them from there.”

A key aspect of the new managed MLflow service is its tight integration with existing SageMaker components and workflows: actions taken in MLflow are automatically synced to services such as the SageMaker Model Registry.

“We built it to be integrated with other SageMaker capabilities, including training and deployment model hosting and the SageMaker Model Registry, so customers have a fully managed, seamless experience using MLflow within SageMaker,” Mehrotra explained.

AWS is already letting a few organizations try out its managed services in beta, with early users including web hosting provider GoDaddy and Toyota Connected, a subsidiary of Toyota Motor Corp.

The intersection of SageMaker and Bedrock

While Amazon SageMaker has traditionally focused on the end-to-end machine learning lifecycle, AWS has introduced new services such as Amazon Bedrock aimed at building generative AI applications.

Mehrotra shed light on SageMaker's role in this emerging AI ecosystem.

“SageMaker is essentially a service for building models, training models, and deploying models, while Bedrock is the perfect service for creating generative AI-based applications,” says Mehrotra. “Many of our customers are using multiple services, including SageMaker and Bedrock, to create generative AI solutions.”

He highlighted how developers can build models in SageMaker and then leverage serverless capabilities to deploy them into AI applications via Bedrock. The two services are complementary parts of AWS' broader generative AI stack.

Future Strategy for Amazon SageMaker

Looking to the future, Mehrotra outlined some key priorities driving the product roadmap and investments for Amazon SageMaker. He said AWS is focused on a few different areas:

One of our key areas of focus is helping you scale while optimizing costs.

“We're also focused on taking the undifferentiated heavy lifting off of our customers as they build new AI solutions,” he said. “We're going to have more capabilities coming out that make it really easy and simple for them to create these solutions and get them to market faster.”