Skip to content
/ lmvp Public

Open source neural network versioning system that separates model management and training operations.

License

Notifications You must be signed in to change notification settings

dsc-umass/lmvp

Repository files navigation

License GitHub commit activity GitHub stars

LMVP

Lightweight Model Versioning Platform

Contributing

Please read the entire guide here.

About

An open source neural network versioning system designed to separate model management functions and computationally expensive training operations.

Getting Started

Docker Compose is the recommended way to run the API server. After cloning or downloading the repository, run export DB_PASSWORD=arbitrarypassword and then docker-compose run web python3 manage.py migrate to build the initial tables in thedatabase. Then run docker-compose up to start LMVP. From there, you can navigate to the server in a web browser to see the Django Rest Framework interface, or use another tool to make API requests.

Features

  • Fetch, update, and publish models from anywhere
  • Track metrics
  • Manage model history on the web
  • Add team members, make contributions

Example commands

>>> localModel = lmvp.download(modelName)
>>> lmvp.showHistory(project)
>>> lmvp.update(modelname, localModel)

Task Groups

Milestones for each group are available in the Projects tab. Please make sure to assign each issue to the correct milestone.

Group Features
Python module Allow team members to manage their models from Colab or another environment. Makes API requests to the central server.
Central server Provides a REST API for versioning, organization, and model storage. Built using Django REST Framework.
Web Interface The web administrator console to allow project management. Makes API requests to the central server.

Comparison with other platforms

Traditional Cloud Worlflow Local Training Workflow

Docker Hub

You can use Docker to pull an autobuild image from Docker Hub. Right now it is available here. Be aware that they seem to only work for the x86 architecture, so if you want ARM or something else you will need to build it yourself.

Troubleshooting starting Docker container

Here are some problems you might run in to when trying to start the docker container:

psql: FATAL: password authentication failed for user "postgres"

This will happen if you tried to start the docker container without exporting the database password first. To fix this, run these commands

$ docker kill $(docker ps -q)
$ docker rm $(docker ps -a -q)
$ docker rmi $(docker images -q)

then export the password

$ export DB_PASSWORD=arbitrarypassword

and then run these commands to build and run the container

$ docker-compose run web python3 manage.py migrate
$ docker-compose up

Cannot connect to the Docker daemon. Is 'docker daemon' running on this host?

I ran into this problem while running Docker on a VM with WSL 2. Try installing docker-desktop on Windows first, and then starting docker on your Linux VM with

$ sudo service docker start

To make sure Docker automatically starts on boot, run

$ sudo systemctl enable docker

this shouldn't be necessary, but you may as well run it just to be safe.

About

Open source neural network versioning system that separates model management and training operations.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published