dbmdz/berts

Potential publishing of TF checkpoints

Closed this issue · 3 comments

wittenator commented

Awesome work in creating another German BERT model trained on rather scientific texts, dbmdz team!
I would like to use your model with Bert-as-Service and would need the TF checkpoints for that. Do you have them somewhere laying around by any chance?

stefan-it commented

Hi @wittenator,

thanks for your interest!

The original TF checkpoints can be downloaded via:

wget https://schweter.eu/cloud/berts/bert-base-german-dbmdz-cased.tar.gz # cased model
wget https://schweter.eu/cloud/berts/bert-base-german-dbmdz-uncased.tar.gz # uncased model

Please let me know if that worked for you!

  • 👍1
wittenator commented

Sorry for the hefty delay. The checkpoints work like a charm, thanks a lot for this!
Out of curiosity, how hard would it be further train the model on more data? By chance I have a text dump of DFGs GEPRIS database at hand which could be useful for this.

stefan-it commented

Hi @wittenator ,

we haven't use your recommended corpus for future models, but we recently released larger models for German, see our paper:

German’s Next Language Model

Colab with @brandenchan and @Timoeller from deepset ❤️

So feel free to try our new models 🤗