Preventing Model Reinitialization in Label Studio ML Backend on Each Annotation


I’m utilizing a Label Studio machine learning backend for Named Entity Recognition (NER), and I’m experiencing long load times because the model gets reinitialized after every annotation. Is there a way to avoid this and reduce the time between annotations?


Certainly! To prevent your ML model from reinitializing with each annotation, you should disable the active learning feature in Label Studio, which is usually active by default and causes the model to be retrained with new annotations.

To disable active learning:

  1. Access your project settings in Label Studio.
  2. Click on the Machine Learning settings.
  3. Find your ML backend configuration.
  4. Turn off the “Allow version auto-update” option, which prompts the model to retrain.

By disabling this feature, your model will no longer retrain and reinitialize automatically after annotations. You can choose to train the model manually whenever necessary.

Additionally, verify if you have any webhooks configured to initiate training on annotation as they could also trigger reinitialization. You can check and amend webhook settings in the Webhooks section of your project settings.