uncased Distillbert model for German

[neufang]

Hi, I am trying to use Transformers for German question answering. I used distillbert-cased for fine-tunning QA model, which gave worse performance than other pre-trained models. I also found out the the performance with uncased models is generally better than cased models.

• Is there any tips for parameter tuning with fine tuning models?
• Do you have plan to train a Distillbert-uncased for German?