As from https://github.com/RasaHQ/rasa/blob/main/rasa/nlu/components.py#L670 method, components can be cached with implementing cache_key. It looks like caching is done by ComponentBuilder class.
There is a tutorial for caching components when using rasa from pyhton API, by passing the same ComponentBuilder instance to the Interpreter.load method. http://35.196.60.7/docs/nlu/0.15.1/python/#reducing-memory-use-when-loading-multiple-models
But when using the REST API to load a model, a new ComponentBuilder instance is created by rasa each time a request is received so the caching does not work. Is there a way to make this work when using REST API?