Hello,
I’m currently working with a deep learning framework for segmentations, and I wanted to integrate wandb to automate jobs, so I followed the wandb documentation with the docker approach.
To do this, I created the docker image, then tested it locally (without wandb) and it worked very well.
Then, to ensure wandb integration, I carried out these steps:
wandb job create --project “Hippocampus_seg” --entity “ip_team” --name “training_job” image nn_mlops_image:latest
(The job has been created on my wandb UI)
As my framework needs environment variables in order to read and write to directories on my machine, I need to define volumes, so the command I used is as follows:
docker run -it --gpus all --ipc=host --name nn_mlops_container -e WANDB_PROJECT="Hippocampus_seg"
-e WANDB_ENTITY=“ip_team” -e WANDB_API_KEY="xxx"
-e WANDB_DOCKER=“nn_mlops_image:latest” -v "/c/Users/mick.jordan/dataset/aaa:/aaa"
-v “/c/Users/mick.jordan/dataset/bbb:/bbb” -v "/c/Users/mick.jordan/dataset/ccc:/ccc"
nn_mlops_image:latest
Then, on wandb I configured the queue and the agent
When I run the job on wandb: My framework which is Dockerized, it returns me an error: it cannot find the environment variables :
aaa:/aaa
bbb:/bbb
ccc:/ccc"
And on the wandb UI, I got this:
<?xml version='1.0' encoding='UTF-8'?>AccessDenied
Access denied.