medusa-ml-template

a template project for machine learning experiments, using dockerized pytorch environments and data downloaders

TODO

• Update pytorch docker image
• jupyter setup guide
• Integrate Joe R Playlist examples
• LoRa fine tuning example

Requirements

• docker
• a bash shell

Optional Dependencies

• for GPU users nvidia container toolkit

Getting Started

• Build the docker image by using sh build-docker-image.sh
• start your training environment by running sh run-training-environment.sh or sh gpu-environment.sh and follow the link to the jupyter server
• Use a downloader to download a dataset into the data folder
• Run a training notebook to fit your model
• Export your model to the saved models folder
• Shrink, optimize, and deploy your model, see deploy for examples

Frequently Asked Questions

Will this run on Windows or macOS?

Not really, but you can try if you must.

How to I shut down the notebook?

Go to the terminal where you ran the run-training....sh and press CTRL+C, WTF

What the hell is this notebook stuff?

Try this tutorial to learn more tutorial

Can I deploy this in a cloud environment via Docker? or run it on my big ML rig with many nvidia GPUs?

Hell yes you can! I'll eventually write a more detailed guide here on how to do that, but the setup is almost identical to running locally.