SwanLab

Telegram ​

If you wish to receive immediate Telegram notifications upon training completion or errors, the Telegram Notification plugin is highly recommended.

Improve the Plugin

SwanLab plugins are open-source. You can view the GitHub source code. Suggestions and PRs are welcome!

Preparation ​

References

  1. In the Telegram search bar, enter @BotFather

  2. In the chat window, send the command /newbot. Follow the prompts to input the name and username respectively. Notes:

  • The name is used to display the bot’s name in the message window.
  • The username is for redirecting to Telegram channels, must end with bot, and cannot be duplicated with other bots.

You will receive a message as shown below. Copy and save the token.

  1. In the Telegram search bar, enter @getidsbot. Send the command /start in the chat window to obtain the chat_id. The chat_id is the unique identifier of the user.

Basic Usage ​

Using the Telegram notification plugin is straightforward. Simply initialize a TelegramCallback instance：

python
from swanlab.plugin.notification import TelegramCallback

telegram_callback = TelegramCallback(
    bot_token="<YOUR_TELEGRAM_BOT_TOKEN>",
    chat_id="<YOUR_TELEGRAM_CHAT_ID>",
    language="en"
)

Then pass the telegram_callback object into the callbacks list parameter of swanlab.init:

python
swanlab.init(callbacks=[telegram_callback])

In this way, you will receive a Telegram notification when training completes or an error occurs (triggering swanlab.finish()).

Custom Notifications ​

You can also use the send_msg method of the TelegramCallback object to send custom Telegram messages.

This is particularly useful for notifying you when certain metrics reach specific thresholds!

python
if accuracy > 0.95:
    # custom messages
    telegram_callback.send_msg(
        content=f"Current Accuracy: {accuracy}",
    )

Limitations ​

• The training completion/error notification of the Telegram notification plugin relies on the on_stop lifecycle callback of SwanKitCallback. Therefore, if your process is abruptly killed or the training machine shuts down unexpectedly, the on_stop callback will not be triggered, and no Telegram notification will be sent.

• A more robust solution will be available with the launch of SwanLab's Platform Open API. Stay tuned!