This model is based on de_core_news_sm (License CC BY-SA 3.0) and has been additionally trained on NER data from the legal domain.

Training parameters:

• epochs = 4
• initial dropout rate = 0.5, linear decay
• intial batch size = 1, batch compound rate = 1.001