Helsinki very slow with Japanese


I have been using helsinki model in different languages, and I noticed that the Helsinki-NLP/opus-mt-jap-en is excessively slow (up to x20 slower than fr-en for example with the exact same setup, for example using the pipeline object).
Is there any reason for this? Can I make this specific model faster?


Helsinki-NLP/opus-mt-ja-en solved the issue :slight_smile: