Run on CI doesn't read docker user options

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

When running as CI (run_container_on_ci), the docker options for user are not read, which means all files are created as root. See here where it uses them, but only for run_container.

2

Can you use leaverge it to apply into run_container?

3

Sorry I don’t understand you. What do you mean?

4

I mean you can modify run_container() by leveraging run_container_on_ci(). Because you mentioned that in another topic the run_container_on_ci() can work.

There are a few ways to run TAO without a TTY when using SkyPilot on GCP:

  1. Use the TAO docker directly instead of tao-launcher:
docker run --runtime=nvidia --rm nvcr.io/nvidia/tao/tao-toolkit:5.5.0-pyt dino train -e dino.yaml

This command runs the TAO container without the -it flags, avoiding the TTY requirement.

  1. Modify the tao-launcher source code:
    The tao-launcher source code is available on GitHub. You can modify the docker_handler.py file to remove the TTY requirement.

  2. Force CI mode:
    Set the CI_PROJECT_DIR environment variable before running your command. This triggers the run_container_on_ci function instead of run_container, which may work without a TTY.

export CI_PROJECT_DIR=/path/to/your/project
tao model dino train -e dino.yaml
  1. Use the undocumented tty option in .tao_mounts.json:
    Although not officially documented, you can try adding a “tty”: false option to your .tao_mounts.json file.
5

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.