KCL-BMEIS/runai_custom_conda_docker

Runai with Conda

This sample project builds a custom Docker image with conda environment that is passed to the runai GPU distribution layer.

Docker

The Dockerfile builds on top of an existing PyTorch docker image. It then takes the dependencies from the environment file to generate a toy environment in the custom image. The image can be built by calling

sh build_docker.sh

Submit a Job

A job can be submitted via runai submit. The complete call is safed inside the submit.sh bash script, call

sh submit.sh

Run a Python Script

The submit shell script also calls the run.sh shell script via the --command flag. This script activates conda and the toy environment, and executes the toy.py Python script.