Offline Dashboard

Note

The offline dashboard is a legacy feature of SwanLab and is currently maintained with minimal updates. It is no longer actively developed.

For private deployment needs, we recommend using the Docker version.

The offline dashboard is a lightweight web-based dashboard that operates similarly to tensorboard in offline mode.

GitHub: https://github.com/SwanHubX/SwanLab-Dashboard

Installation

Starting from SwanLab version 0.5.0, the offline dashboard is no longer included by default and must be installed via the dashboard extension.

To use the offline dashboard, install the dashboard extension for swanlab:

pip install swanlab[dashboard]

Offline Experiment Tracking

To track experiments offline, configure the logdir and mode parameters in swanlab.init:

...

swanlab.init(
  logdir='./logs',
  mode="local",
)

...

• Set the mode parameter to local to disable syncing experiments to the cloud. • The logdir parameter is optional and specifies the directory where SwanLab log files are saved (default is the swanlog folder). • Log files are created and updated during experiment tracking, and the offline dashboard relies on these files for operation.

All other functionalities remain consistent with the cloud version.

Launching the Offline Dashboard

Open a terminal and use the following command to start a SwanLab dashboard:

swanlab watch ./logs

Mnemonic: Use SwanLab to watch the files in ./logs.

After execution, a backend service will be launched, and SwanLab will provide a local URL (default is http://127.0.0.1:5092).

Access this URL in your browser to view experiments using the offline dashboard.

How to Set Port and IP