The accuracy evaluation for models is available on the spacy website. E.G. https://spacy.io/models/de#de_dep_news_trf-accuracy . Which test data has been used to compute the accuracy evaluation? I want this information for all german models.
Evaluation details for english models is at https://spacy.io/usage/facts-figures/#section-benchmarks .
According to https://github.com/explosion/spaCy/discussions/8338#discussioncomment-872225 in the case of a dutch model the sources of training data were also used for evaluation. The training/test split is not documented, as far as I can see. I assume the same to be true for german models.