Docker Image Setup

[Amaan Kazi]

Dear Challenge Team,

For my current docker setup, I am using nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu20.04 which requires the NVIDIA driver, nvidia-cuda-toolkit, and nvidia-container-toolkit to run properly. I was wondering if the instance or machines you run the code on has these toolkits installed?

Additionally, for the image generation, would you be able to run the commands to generate the images (e.g., run a bash command) or should I have this incorporated within the Dockerfile or train_model code?

Thank you for your time.

Best,

Amaan

[PhysioNet Challenge]

Dear Amaan,

Yes, I expect you can run this container on our system. I tested this base image (nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu20.04) on one of the machines and the CUDA installation looks good. Go ahead and submit, and we can troubleshoot any issues that come up.

We can’t support custom instructions like manually running a bash command before training (the pipeline is automated), so the best option is to call the image generator during the train_model code. I wouldn’t put it in the Dockerfile: the system where we build the images doesn’t have the same resources as the system where we run the training code.

Best,

James