AWS Launches Managed Open Source MLflow for Enhanced Machine Learning on Amazon SageMaker

While the spotlight in the generative AI landscape at AWS has primarily been on Amazon Bedrock over the past year, Amazon SageMaker remains a vital asset, delivering essential functionalities for machine learning.

Launched in 2017, Amazon SageMaker facilitates the entire machine learning lifecycle—from model creation and training to deployment and management at scale. It provides a comprehensive managed environment with tools that enable customers to build, train, and deploy machine learning and deep learning models. Hundreds of thousands of users rely on SageMaker for tasks like training popular generative AI models and managing machine learning workloads. Notable applications include training Stability AI’s Stable Diffusion and powering Luma’s Dream Machine text-to-video generator.

AWS is further enhancing SageMaker with the general availability of the managed MLflow service. MLflow, an open-source platform, streamlines the machine learning lifecycle, covering experimentation, reproducibility, deployment, and monitoring of ML models. By integrating MLflow as a managed service within SageMaker, AWS empowers users to create the next generation of AI models more efficiently.

“Given the rapid pace of innovation, our customers want to transition quickly from experimentation to production, accelerating their time to market,” said Ankur Mehrotra, Director and General Manager of Amazon SageMaker at AWS. "We’re launching MLflow as a managed capability in SageMaker, allowing users to set up and launch MLflow in a few clicks."

What MLflow Offers AWS Users

MLflow is widely adopted among developers and organizations for MLOps. Mehrotra emphasized that the new managed service enhances enterprise user choice without compromising existing features.

By providing MLflow as a fully managed solution integrated with SageMaker, AWS addresses the needs of users who want a seamless experience across both platforms. “As they iterate on their models, they can easily log metrics in MLflow, track, and compare different iterations,” Mehrotra explained. “They can then register these models in a model registry and deploy them with ease.”

The managed MLflow service is deeply integrated with existing SageMaker components, ensuring that actions within MLflow sync automatically with SageMaker services like the Model Registry. “We built this to integrate seamlessly with SageMaker’s capabilities, whether it's model training, deployment, or hosting, providing customers with a cohesive MLflow experience,” Mehrotra added.

Several organizations, including web hosting provider GoDaddy and Toyota Connected, a subsidiary of Toyota Motor Corporation, have already explored the managed service during its beta phase.

SageMaker and Bedrock: Complementary Services

While Amazon SageMaker focuses on the complete machine learning lifecycle, AWS recently introduced Amazon Bedrock to develop generative AI applications. Mehrotra clarified SageMaker's role within this AI ecosystem:

“SageMaker is tailored for building, training, and deploying models, whereas Bedrock excels at creating generative AI applications,” he stated. “Many customers utilize both SageMaker and Bedrock, along with other services, to develop their generative AI solutions.”

This allows developers to create models in SageMaker and deploy them into AI applications via Bedrock, leveraging its serverless capabilities—evolving these services into complementary elements of AWS’s generative AI offerings.

Amazon SageMaker's Future Focus

Looking ahead, Mehrotra shared the key priorities guiding the Amazon SageMaker product roadmap. A primary focus is on scaling and optimizing costs while simplifying the development process for customers.

“We aim to reduce the undifferentiated, heavy lifting for customers as they build new AI solutions. Expect to see more capabilities that enable customers to create and launch these solutions faster,” he concluded.

This strategic focus positions Amazon SageMaker as a pivotal player in advancing the machine learning and generative AI landscape.