DL Workbench enables performing multiple experiments to analyze performance and accuracy data. To always have access to your data in the DL Workbench, use additional parameters when you run the DL Workbench that enable DL Workbench state preservation.
NOTE: It is highly recommended to enable DL Workbench state preservation.
Why you need to preserve DL Workbench state:
Ignore state preservation if:
Even if you do not enable state preservation, your data is secured as long as there is a container on your machine. When you stop a container, you still have access to your data when you resume the container as described in the Pause and Resume Docker Container section of Work with Docker Container. However, if you remove a container that you ran without enabling state preservation, all your data is lost with that removed container.
To restore the DL Workbench data:
To enable DL Workbench state preservation, mount a host directory to a Docker container. The example below shows how to mount a local folder while running a Docker container on Linux* with a CPU enabled as a profiling target and in detached mode. To learn how to run the application on different targets, operating system, or in a different mode, see Install DL Workbench from Docker Hub.
start_workbench.shscript as described in the Install DL Workbench from Docker Hub* on Linux* OS section of Install DL Workbench from Docker Hub.
All your data is placed in the mounted directory once you mount it and run the DL Workbench:
Due to problems of mounting a local folder to a Docker container on Windows, the best way to preserve the state on Windows is to use Docker volumes:
All your data is placed in the mounted volume once you mount it and run the DL Workbench:
ASSETS_DIR contains sensitive data such as a token, models, and datasets. Share this data only in a trusted environment. DL Workbench supports a scenario when you share only system files with profiling data, and not models and datasets.
Choose instructions for your operating system:
When you share only profiling data, the DL Workbench marks models, datasets, and projects as Read-only. Read-only means that it is not possible to run optimizations, profiling, or measurements on removed assets, while you can continue with importing new models and datasets.
Read-only artifacts on the Create Project page:
Read-only project on the Projects page:
~/.workbenchfolder, so that it contains only the
~/.workbenchfolder to another machine and import it with the following command:
When importing assets, DL Workbench validates their consistency. If any assets have different checksum to what the DL Workbench stores, these artifacts are considered as threatening security of the DL Workbench. Remove these assets and try to run the DL Workbench again.
DL Workbench fails to start if the provided assets cannot be imported due to aforementioned versioning policy. In that case, create new assets directory and mount it instead of the existing one.