English

简体中文

English

简体中文

Appearance

FAQ ​

Why can't I input the API Key during login? ​

Refer to this answer: Link

How to start multiple experiments from a single script? ​

Add swanlab.finish() between multiple experiment creations.

After executing swanlab.finish(), executing swanlab.init() again will create a new experiment.
If swanlab.finish() is not executed, subsequent swanlab.init() calls will be ignored.

How to disable SwanLab logging during training (for debugging)? ​

Set the mode parameter of swanlab.init to 'disabled' to prevent experiment creation and data logging.

python
swanlab.init(mode='disabled')

The local training has ended, but the experiment is still shown as running on the SwanLab UI. How to change the status? ​

Click the stop button next to the experiment name to change the status from "Running" to "Interrupted" and stop receiving data uploads.

stop

On the same machine, multiple people are using SwanLab. How should it be configured? ​

After completing the login with swanlab.login, a configuration file will be generated on the machine to record the login information, so that there is no need to log in again next time. However, if multiple people are using the same machine, care must be taken to avoid logs being transmitted to someone else's account.

There are two recommended configuration methods:

Method 1: Add swanlab.login(api_key='Your API Key') at the beginning of the code.

Method 2: Before running the code, set the environment variable SWANLAB_API_KEY="Your API Key".

How to view local details of the line chart? ​

Zoom in on the line chart by holding down the mouse and dragging over the target area to magnify that region.

details

Double-click the area to restore the original view.

Internal Metric Names ​

Metric names refer to the key part of the dictionary passed into swanlab.log(). Some keys are internally used by SwanLab to transmit system hardware metrics, so it is not recommended to use them.

Internal metrics include:

  • __swanlab__.xxx

Experiment Status Rules ​

An experiment can be in one of three states: Completed, Running, or Crashed.

  • Completed: The training process has ended naturally, or swanlab.finish() was manually executed.
  • Running: The training process is ongoing, and swanlab.finish() has not been executed.
  • Crashed: The training process was abnormally terminated due to bugs, machine shutdown, Ctrl+C, etc.

Some users may encounter the following situation: Why does my training process seem to be ongoing, but the SwanLab chart shows it as crashed?

This is because SwanLab has a hidden rule for determining crashes. If no logs (including automatically collected system metrics) are uploaded within 15 minutes, the experiment is marked as crashed. This is to prevent the experiment from remaining in the "Running" state indefinitely if the training process is unexpectedly killed and cannot trigger the status upload logic in the SwanLab SDK.

Therefore, if your machine experiences network issues for more than 15 minutes, the experiment status will be displayed as "Crashed."

Command Line Logging and Truncation ​

SwanLab records the standard output stream of the process after swanlab.init() is called, which can be viewed in the "Logs" tab of the experiment.

If a line of command-line output is too long, it will be truncated. The current limit is 500 characters.

If you wish to remove this restriction, you can modify the MAX_UPLOAD_LEN parameter in swanlab/log/console.py in the SwanLab package source code. This will ensure that the corresponding command-line output is stored in the swanlog log file, and the offline dashboard will display it accordingly. However, this change will not affect the rules of the cloud version.