Greetings, I am working on containerizing our BraTS submission and wanted to confirm whether it's acceptable to use a -devel base image, specifically pytorch/pytorch:2.2.2-cuda11.8-cudnn8-devel This is needed to compile a CUDA-dependent libray (mamba-ssm) used by one of our models. Just wanted to make sure this doesn't violate any submission constraints. Thanks in advance! Best regards, Luke Hsu

Created by To-Liang Hsu LukeHsu1999
Hi @vchung, Thank you for the clarification! Yes, I have tested locally with the recommended commands: ``` docker run \ --rm \ --network none \ --gpus=all \ --volume /PATH/TO/INPUT:/input:ro \ --volume /PATH/TO/OUTPUT:/output:rw \ --memory=16G --shm-size 4G \ docker.synapse.org/PROJECT_ID/IMAGE_NAME:TAG ``` Best regards, Luke Hsu
Hi @LukeHsu1999 , In case it will help, here are the resource specs of the AWS cloud compute we will run your containers on: * 1 NVIDIA V100 GPU with 16 GiB VRAM * Driver version: 575.57.08 * CUDA version: 12.9 * **No** network access * Containers to be run with 16 GB memory and 4GB shared memory (shm-size)
Hi Mehdi, Thank you! I will make sure to test the image locally before pushing to Synapse. Best regards, Luke Hsu
Hi @LukeHsu1999, Thanks for reaching out. I suggest running your built Docker image locally first, to ensure reproducibility (among others), before pushing it to the Synapse registry. Best, /Mehdi

Question about base image for Docker Submission page is loading…