---
license: cc-by-4.0
language: 
- de
task_categories:
- question-answering
tags:
- question-answering
- pytorch
- qa
- de
---

# SVALabs - InfoXLM German QA

In this repository we present our german question answering model.

The trained model is based on [InfoXLM-large](https://huggingface.co/microsoft/infoxlm-large) and was finetuned using the [SQuAD](https://huggingface.co/datasets/squad) dataset and the [GermanQuAD](https://huggingface.co/datasets/deepset/germanquad) dataset.

### Model Details

| | Description or Link  |
|---|---|
|**Base model**   | [```InfoXLM-large```](https://huggingface.co/microsoft/infoxlm-large) |
|**Finetuning task**| Question Answering  |
|**Source datasets**| [```SQuAD```](https://huggingface.co/datasets/squad); [```GermanQuAD```](https://huggingface.co/datasets/deepset/germanquad)|


### Performance

The model was tested on 1692 samples of the GermanQuAD test dataset (the other samples were used for validation)

F1-Score: 89.05

EM: 74.76

### Contact
- Nicole Wochatz, nicole.wochatz@sva.de
- Stefan Kammer, stefan.kammer@sva.de