Alternative approaches for text classification task

I am trying to classify every single feed from a stream of news feeds.
I tried zero-shot classification with bart-large-mnli however it was very slow (I have about 200 labels at the moment).

  1. How can I speed up the task with zero-shot classification? It will be deployed on a server without GPU
  2. Are there other, better suited, approaches for this task (many small texts with many labels)?

Many thanks :pray: