On-premises Valohai Enterprise installation allows users to use local directory mounts.
How this technically works is that specified local directories are mounted inside the Valohai execution containers as Docker volumes. This allows reading and writing to the host machine or local network disks.
You define the mounts on per-step basis in valohai.yaml:
- step: name: mount-example image: tensorflow/tensorflow:1.13.1-gpu-py3 command: - ls -la /my-data mounts: - source: /path/to/directory/outside/container destination: /my-data readonly: false
You can try this out with:
vh exec run --adhoc mount-example(if not committed and pushed)
vh exec run mount-example(if committed, pushed and fetched)