mlem VS ZnTrack

Mlem for Lemmy has a similar look and feel to Apollo. Still missing a lot of features though. https://github.com/iterative/mlem

Here are a few things to consider: 1. You're using both DVC and MLflow to store the model artifact, why? 2. How I envision MLflow, DVC and git to work together is like this. DVC to manage the training dataset, git to manage the code, and MLflow will do the rest. About the part about "versioning" the model, MLflow has a model registry feature to "tag" a well-performing experiment. 3. Or just to do everything in DVC. DVC also has a way to do experiment tracking. Then if you need a model registry there's MLEM by the same company.

No, it's a completely separate open source tool, not directly related to DVC - https://github.com/iterative/mlem

This functionality can be used from open source tool mlem.ai and our released UI - https://studio.iterative.ai/

MLEM is a new MLOps tool to bridge the gap between ML engineers and DevOps teams by using the git-based approach that developers are already familiar with. Using MLEM, developers can store and track their ML models throughout their lifecycle: GitHub - iterative/mlem: 🐶 Version and deploy your ML models following GitOps principles

I expaned a bit on them with my own package https://zntrack.readthedocs.io/ - a general framework for building DVC pipelines through python scripts (and more). This finally brings me to the project I'm actually working on https://github.com/zincware/IPSuite which brings all of this together for the specific use case of machine learned interatomic potentials.

This looks really interesting! I have a similar scenario but haven't looked into it yet. Have you looked at dvc.org - I'm planning on using it together with slurm and what they call CML for my projects. On that context I also wrote a tool that makes DVC more pythonic https://github.com/zincware/ZnTrack altough I'm currently restructuring it a bit but having backwards compatibility in mind.