Skip to main content

· 3 min read
MLflow maintainers

We are excited to announce the implementation of a release candidate process for MLflow! The pace of feature development in MLflow is faster now than ever before and the core maintainer team has even more exciting things planned in the near future! However, with an increased velocity on major feature development comes with a risk of breaking things. As the maintainers of such a widely used project, we are cognizant of the disruptive nature of regressions and we strive to avoid them as much as we can. Aside from new feature development work, our primary goal is in ensuring the stability of production systems. While we do have the aspirational goal of moving fast(er), we certainly don't want to move fast and break things. With that goal in mind, we've decided to introduce a release candidate (RC) process. The RC process allows us to introduce new features and fixes in a controlled environment before they become part of the official release.

How It Works

Starting from MLflow 2.13.0, new MLflow major and minor releases will be tagged as release candidates (e.g., 2.13.0rc0) in PyPI two weeks before they are officially released.

The release candidate process involves several key stages:

  • Feature Development Freeze: Prior to cutting the RC branch and announcing its availability, we will freeze the RC branch from feature commits. Once the branch is cut, only bug fix and stability PRs will be permitted to be merged, ensuring that unexpected, late-arriving, potentially regression-causing merges are not permitted to destabilize the forthcoming release.
  • Pre-Release Announcement: We will announce upcoming features and improvements, providing our community with a roadmap of what to expect.
  • Release Candidate Rollout: A release candidate version will be made available for testing, accompanied by detailed release notes outlining the changes.
  • Community Testing and Feedback: We encourage our users to test the release candidate in their environments and share their feedback with us by filing issue reports on the MLflow Github repository. This feedback is invaluable for identifying issues and ensuring the final release aligns with user needs (i.e., we didn't break your workflows).
  • Final Release: After incorporating feedback and making necessary adjustments, we will proceed with the final release. This version will include all updates tested in the RC phase, offering a polished and stable experience for all users.

This approach provides several benefits:

  • Enhanced Stability: By rigorously testing release candidates, we can identify and address potential issues early, reducing the likelihood of disruptions in production environments.
  • Community Feedback: The RC phase offers you, a member of the MLflow community, the opportunity to provide feedback on upcoming changes. This collaborative approach ensures that the final release aligns with the needs and expectations of our users.
  • Gradual Adoption: Users can choose to experiment with new features in a release candidate without committing to a full upgrade. This flexibility supports cautious integration and thorough evaluation in various environments.

Get Involved

Your participation is crucial to the success of this process. We invite you to join us in testing upcoming release candidates and sharing your insights. Together, we can ensure that MLflow continues to serve as a reliable foundation for your machine learning projects. Stay tuned for announcements regarding our first release candidate. We look forward to your contributions and feedback as we take this important step toward a more stable and dependable MLflow.

· 6 min read
Abe Omorogbe
Hubert Zub
Yun Park
Chen Qian

In the quickly evolving world of artificial intelligence, where generative AI has taken center stage, the landscape of machine learning is evolving at an unprecedented pace. There has been a surge in the use of cutting-edge deep learning (DL) libraries like Transformers, Tensorflow, and PyTorch to fine-tune these generative AI models for enhanced performance. As this trend accelerates, it's become clear that the tools used to build these models must rapidly evolve as well, particularly when it comes to managing and optimizing these deep learning workloads. MLflow offers a practical solution for managing the complexities of these machine learning projects.

· 7 min read
Carly Akerly

With more than 16 million monthly downloads, MLflow has established itself as a leading open-source MLOps platform worldwide. This achievement underscores the robustness of MLflow and the active community that consistently refines and improves it.

The past year marked a significant milestone for MLflow, particularly in Generative AI. Its integration and support for Large Language Models (LLMs) stood out. This strategic decision has propelled MLflow to the forefront of the AI revolution, establishing itself as the premier GenAI platform that enables users to create more intelligent, efficient, and adaptable AI models and applications.

· 16 min read
Daniel Liden

If you're looking to learn about all of the flexibility and customization that is possible within MLflow's custom models, this blog will help you on your journey in understanding more about how to leverage this powerful and highly customizable model storage format.

· 6 min read
Daniel Liden

Looking to learn more about the autologging functionality included in MLflow? Look no further than this primer on the basics of using this powerful and time-saving feature!

Robust logging practices are central to the iterative development and improvement of machine learning models. Carefully tracking metrics, parameters, and artifacts can be challenging when working with complex machine learning libraries or when experimenting with multiple different frameworks with varying APIs and selections of different objects and values to track.

· 6 min read
MLflow maintainers

The MLflow Documentation is getting an upgrade.

Overhauling the MLflow Docs

We're thrilled to announce a comprehensive overhaul of the MLflow Docs. This initiative is not just about refreshing the look and feel but about reimagining how our users interact with our content. Our primary goal is to enhance clarity, improve navigation, and provide more in-depth resources for our community.

A Renewed Focus on User Experience

The MLflow documentation has always been an essential resource for our users. Over time, we've received invaluable feedback, and we've listened. The modernization effort is a direct response to the needs and preferences of our community.