README.md

metadata

license: cc-by-4.0
metrics:
  - bleu4
  - meteor
  - rouge-l
  - bertscore
  - moverscore
language: de
datasets:
  - lmqg/qag_dequad
pipeline_tag: text2text-generation
tags:
  - questions and answers generation
widget:
  - text: >-
      Empfangs- und Sendeantenne sollen in ihrer Polarisation übereinstimmen,
      andernfalls wird die Signalübertragung stark gedämpft. 
    example_title: Questions & Answers Generation Example 1
model-index:
  - name: lmqg/mbart-large-cc25-dequad-qag
    results:
      - task:
          name: Text2text Generation
          type: text2text-generation
        dataset:
          name: lmqg/qag_dequad
          type: default
          args: default
        metrics:
          - name: BLEU4 (Question & Answer Generation)
            type: bleu4_question_answer_generation
            value: 3.25
          - name: ROUGE-L (Question & Answer Generation)
            type: rouge_l_question_answer_generation
            value: 15.66
          - name: METEOR (Question & Answer Generation)
            type: meteor_question_answer_generation
            value: 15.25
          - name: BERTScore (Question & Answer Generation)
            type: bertscore_question_answer_generation
            value: 72.14
          - name: MoverScore (Question & Answer Generation)
            type: moverscore_question_answer_generation
            value: 52.36
          - name: QAAlignedF1Score-BERTScore (Question & Answer Generation)
            type: qa_aligned_f1_score_bertscore_question_answer_generation
            value: 69.25
          - name: QAAlignedRecall-BERTScore (Question & Answer Generation)
            type: qa_aligned_recall_bertscore_question_answer_generation
            value: 68.05
          - name: QAAlignedPrecision-BERTScore (Question & Answer Generation)
            type: qa_aligned_precision_bertscore_question_answer_generation
            value: 70.69
          - name: QAAlignedF1Score-MoverScore (Question & Answer Generation)
            type: qa_aligned_f1_score_moverscore_question_answer_generation
            value: 50.71
          - name: QAAlignedRecall-MoverScore (Question & Answer Generation)
            type: qa_aligned_recall_moverscore_question_answer_generation
            value: 49.78
          - name: QAAlignedPrecision-MoverScore (Question & Answer Generation)
            type: qa_aligned_precision_moverscore_question_answer_generation
            value: 51.81

Model Card of lmqg/mbart-large-cc25-dequad-qag

This model is fine-tuned version of facebook/mbart-large-cc25 for question & answer pair generation task on the lmqg/qag_dequad (dataset_name: default) via lmqg.

Overview

• Language model: facebook/mbart-large-cc25
• Language: de
• Training data: lmqg/qag_dequad (default)
• Online Demo: https://autoqg.net/
• Repository: https://github.com/asahi417/lm-question-generation
• Paper: https://arxiv.org/abs/2210.03992

Usage

• With lmqg

from lmqg import TransformersQG

# initialize model
model = TransformersQG(language="de", model="lmqg/mbart-large-cc25-dequad-qag")

# model prediction
question_answer_pairs = model.generate_qa("das erste weltweit errichtete Hermann Brehmer 1855 im niederschlesischen ''Görbersdorf'' (heute Sokołowsko, Polen).")

• With transformers

from transformers import pipeline

pipe = pipeline("text2text-generation", "lmqg/mbart-large-cc25-dequad-qag")
output = pipe("Empfangs- und Sendeantenne sollen in ihrer Polarisation übereinstimmen, andernfalls wird die Signalübertragung stark gedämpft. ")

Evaluation

• Metric (Question & Answer Generation): raw metric file

	Score	Type	Dataset
BERTScore	72.14	default	lmqg/qag_dequad
Bleu_1	19.75	default	lmqg/qag_dequad
Bleu_2	9.9	default	lmqg/qag_dequad
Bleu_3	5.42	default	lmqg/qag_dequad
Bleu_4	3.25	default	lmqg/qag_dequad
METEOR	15.25	default	lmqg/qag_dequad
MoverScore	52.36	default	lmqg/qag_dequad
QAAlignedF1Score (BERTScore)	69.25	default	lmqg/qag_dequad
QAAlignedF1Score (MoverScore)	50.71	default	lmqg/qag_dequad
QAAlignedPrecision (BERTScore)	70.69	default	lmqg/qag_dequad
QAAlignedPrecision (MoverScore)	51.81	default	lmqg/qag_dequad
QAAlignedRecall (BERTScore)	68.05	default	lmqg/qag_dequad
QAAlignedRecall (MoverScore)	49.78	default	lmqg/qag_dequad
ROUGE_L	15.66	default	lmqg/qag_dequad

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path: lmqg/qag_dequad
• dataset_name: default
• input_types: ['paragraph']
• output_types: ['questions_answers']
• prefix_types: None
• model: facebook/mbart-large-cc25
• max_length: 512
• max_length_output: 256
• epoch: 6
• batch: 2
• lr: 0.0001
• fp16: False
• random_seed: 1
• gradient_accumulation_steps: 32
• label_smoothing: 0.15

The full configuration can be found at fine-tuning config file.

Citation

@inproceedings{ushio-etal-2022-generative,
    title = "{G}enerative {L}anguage {M}odels for {P}aragraph-{L}evel {Q}uestion {G}eneration",
    author = "Ushio, Asahi  and
        Alva-Manchego, Fernando  and
        Camacho-Collados, Jose",
    booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, U.A.E.",
    publisher = "Association for Computational Linguistics",
}