Hugging Face's logo

deepset
/
gelectra-base-germanquad-distilled

Question Answering Transformers PyTorch Safetensors
German electra exbert Inference Endpoints
Model card Files Community
1
gelectra-base-germanquad-distilled / README.md
MichelBartelsDeepset's picture
MichelBartelsDeepset
Update README.md
6a8efb1
preview code
|
raw history blame contribute delete
No virus
3.24 kB
metadata
language: de
datasets:
  - deepset/germanquad
license: mit
thumbnail: >-
  https://thumb.tildacdn.com/tild3433-3637-4830-a533-353833613061/-/resize/720x/-/format/webp/germanquad.jpg
tags:
  - exbert

bert_image

Overview

Language model: gelectra-base-germanquad-distilled
Language: German
Training data: GermanQuAD train set (~ 12MB)
Eval data: GermanQuAD test set (~ 5MB)
Infrastructure: 1x V100 GPU
Published: Apr 21st, 2021

Details

  • We trained a German question answering model with a gelectra-base model as its basis.
  • The dataset is GermanQuAD, a new, German language dataset, which we hand-annotated and published online.
  • The training dataset is one-way annotated and contains 11518 questions and 11518 answers, while the test dataset is three-way annotated so that there are 2204 questions and with 2204·3−76 = 6536answers, because we removed 76 wrong answers.
  • In addition to the annotations in GermanQuAD, haystack's distillation feature was used for training. deepset/gelectra-large-germanquad was used as the teacher model.

See https://deepset.ai/germanquad for more details and dataset download in SQuAD format.

Hyperparameters 

batch_size = 24
n_epochs = 6
max_seq_len = 384
learning_rate = 3e-5
lr_schedule = LinearWarmup
embeds_dropout_prob = 0.1
temperature = 2
distillation_loss_weight = 0.75

Performance

We evaluated the extractive question answering performance on our GermanQuAD test set. Model types and training data are included in the model name. For finetuning XLM-Roberta, we use the English SQuAD v2.0 dataset. The GELECTRA models are warm started on the German translation of SQuAD v1.1 and finetuned on \\germanquad. The human baseline was computed for the 3-way test set by taking one answer as prediction and the other two as ground truth.

"exact": 62.4773139745916
"f1": 80.9488017070188

performancetable

Authors

  • Timo Möller: timo.moeller [at] deepset.ai
  • Julian Risch: julian.risch [at] deepset.ai
  • Malte Pietsch: malte.pietsch [at] deepset.ai
  • Michel Bartels: michel.bartels [at] deepset.ai

About us

deepset logo We bring NLP to the industry via open source!
Our focus: Industry specific language models & large scale QA systems.

Some of our work:

Get in touch: Twitter | LinkedIn | Slack | GitHub Discussions | Website

By the way: we're hiring!