German UD syntax models

Add de-ud 2021 models

This is the final set of models. Move from TensorDot S3 to GitHub.

Dutch UD syntax model, 6 layers, 384 hidden units

Dutch UD syntax model:

• Universal tags
• Lemmas
• Morphology
• Universal dependencies

Distilled from finetuned XLM-RoBERTa large, into a transformer with 6 hidden layers, 384 hidden units, and 12 attention heads.

Dutch UD syntax model, 12 layers, 384 hidden units, ALBERT architecture

Dutch UD syntax model:

• Universal tags
• Lemmas
• Morphology
• Universal dependencies

Distilled from finetuned XLM-RoBERTa large, into a transformer with 12 hidden layers, 384 hidden units, and 12 attention heads. The model uses the ALBERT architecture with 6 layer groups and 128 dimensional piece embeddings.

Dutch UD syntax model, 12 layers, 384 hidden units

Medium Dutch UD syntax model:

• Universal tags
• Lemmas
• Morphology
• Universal dependencies

Distilled from finetuned XLM-RoBERTa large, into a transformer with 12 hidden layers, 384 hidden units, and 12 attention heads.