5 months ago
MLflow is an open-source platform to help manage ML workflows and models. There are multiple components of MLflow and it is usually self-hosted, which make it a perfect contender to run on Railway.
This bounty will be paid out when:
A high-quality MLflow template is presented
All potential template feedback has been incorporated
All requirements are met and tested
Follows all template best practices where applicable
Template requirements:
The Docker image for MLflow is deployed on Railway
The template can be used for a model registry and serving
Volume-backed storage for all databases using correct mount paths
Service dependencies should be correctly configured using proper startup order and health checks
Environment variables correctly configured for Railway domains using private networking where applicable
A few resources to get you started:
Pinned Solution
5 months ago
Alright, I've got a template working and published here
I am doing some final testing but would love some feedback while I work on buttoning everything up. Thanks!
4 Replies
5 months ago
Hey there! We've found the following might help you get unblocked faster:
If you find the answer from one of these, please let us know by solving the thread!
5 months ago
Alright, I've got a template working and published here
I am doing some final testing but would love some feedback while I work on buttoning everything up. Thanks!
mykal
Alright, I've got a template working and published hereI am doing some final testing but would love some feedback while I work on buttoning everything up. Thanks!
5 months ago
Just wanted to hop in and say I've finished testing this and am pretty happy with how it's working. I've been able to:
- Train several classical ML models using example training data from sklearn.
- Track their params/accuracy alongside versioned published models.
- Pull the previously published models down and run them as python functions.
- Build and test an essentially railway-ready (sans-auth) docker image based on one of the trained models I published that puts FastAPI in front of the trained model.
I've also documented everything I tested (with exception to the docker image part as I literally followed their docs with no changes) in an examples folder in the source repo and linked it in my template description throughout.
Looking forward to the team's feedback on Monday! have a good weekend all.
Edit on 2025-08-25: Here's the GitHub Repo for reference as well!
Status changed to Solved sarahkb125 • 5 months ago