Parsing a Single German Sentence Requires More Than 12GB GPU RAM

[Yong Zheng-Xin]

The issue as mentioned in the title. I am wondering if this is normal since English Constituency Parsing requires way fewer GPU resources (can be done with less than 12 GPU RAM).

My code (following https://spacy.io/universe/project/self-attentive-parser)

import spacy
from benepar.spacy_plugin import BeneparComponent
!python -m spacy download de_core_news_sm
import de_core_news_sm
nlp = de_core_news_sm.load()

nlp.add_pipe(BeneparComponent('benepar_de'))
doc = nlp("Guten Morgen. ")