I have seen this question asked only once before but unanswered - we have noticed our RASA run with SpacyNLP as component in config.yml loads the Spacy model twice and this results in a very big memory footprint (2.5GB whereas the Spacy model is only 500mb):
2024-04-05 15:08:25 DEBUG rasa.engine.graph - Node 'provide_SpacyNLP1' loading 'SpacyNLP.load' and kwargs: '{}'.
2024-04-05 15:08:25 INFO rasa.nlu.utils.spacy_utils - Trying to load SpaCy model with name 'nl_core_news_lg'.
2024-04-05 15:08:27 DEBUG rasa.engine.graph - Node 'run_SpacyNLP1' loading 'SpacyNLP.load' and kwargs: '{}'.
2024-04-05 15:08:27 INFO rasa.nlu.utils.spacy_utils - Trying to load SpaCy model with name 'nl_core_news_lg'.
Is this expected? I have 1 Sanic worker and the only relevant thing in config.yml is
- name: "SpacyNLP"
model: "nl_core_news_lg"
How do I reduce the memory footprint of RASA?