Thanks, but I wonder about if using NFS, when a agent need to read a dataset from cache folder, it need to copy dataset from NFS server => local machine, so it's basically same as downloading from clearml server unless NFS has a better bandwidth, faster or smth like that. Do you know if Is there any faster way to do that?

@<1523701070390366208:profile|CostlyOstrich36> Thanks, I know about editing from webUI but for some arg like network and ipc I want to set to default for clearml-agent so our enduser don't need to worry about it. When changing agent.default_docker.arguments in clearml-agent to

["--network=host", "--ipc=host"] 

Then when our user init a tasks with a custom image (different than agent.default_docker.image) => I check the console log and see that in the docker run command, there is n...

Here is the clearml.conf

    default_docker: {
        # default docker image to use when running in docker mode
        image: "python3.10-cuda12.2:latest"

        # optional arguments to pass to docker image
        arguments: ["--network=host", "--ipc=host" ]
    }

And here is the code that user used to run task (with python3.8 image)

from clearml import Task
task = Task.init(project_name='my_project', task_name='my_remote_task')

task.set_base_docker('python3.10-cuda12....