Adding GermanQuAD and GermanDPR to HF datasets

[Timoeller]

Lets add our data to HF datasets with info from our landing page and links to haystack

[julian-risch]

@Timoeller could you please review this so that we can close this issue?

Both datasets are uploaded, have dataset cards with links to our landing page and to haystack, and are linked to the models we trained on them. Main effort was to code a generate_examples method that yields records from the datasets, e.g., here: https://huggingface.co/datasets/deepset/germanquad/blob/main/germanquad.py

• GermanQuAD is uploaded now here https://huggingface.co/datasets/deepset/germanquad
• GermanDPR is uploaded now here https://huggingface.co/datasets/deepset/germandpr

Usage after installing HF datasets via pip install datasets:

from datasets import load_dataset
dataset = load_dataset("deepset/germanquad", split="train")
dataset[0]
{'answers': {'answer_start': [146], 'text': ['britischen Common Laws']}, 'context': "Recht_der_Vereinigten_Staaten\n\n=== Amerikanisches Common Law ===\nObwohl die ...", 'id': 51870, 'question': 'Von welchem Gesetzt stammt das Amerikanische ab? '}

Limitations
There are two different ways to share datasets on HF datasets: community provided datasets and canonical datasets
I chose the first option so that the dataset is identified under the namespace of our organization: deepset/germanquad
However, it seems that the dataset viewer/explorer is limited to the second option: https://huggingface.co/datasets/viewer/?dataset=deepset/germanquad does not work
See details here: https://huggingface.co/docs/datasets/share_dataset.html

[Timoeller]

Lets have the datasets under our namespace. I love the summaries.

2 optional improvements could be: add our germanquad picture to the model card and removing the second question in the example json for GermanQuAD (the json is pretty hard to read)

But Im fine with closing this issue as well.

[julian-risch]

Good points, thanks. I added the picture and removed the second question.