Model Card of lmqg/mt5-small-koquad-qg

This model is fine-tuned version of google/mt5-small for question generation task on the lmqg/qg_koquad (dataset_name: default) via lmqg.

Overview

• Language model: google/mt5-small
• Language: ko
• Training data: lmqg/qg_koquad (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="ko", model="lmqg/mt5-small-koquad-qg")

# model prediction
questions = model.generate_q(list_context="1990년 영화 《 남부군 》에서 단역으로 영화배우 첫 데뷔에 이어 같은 해 KBS 드라마 《지구인》에서 단역으로 출연하였고 이듬해 MBC 《여명의 눈동자》를 통해 단역으로 출연하였다.", list_answer="남부군")

• With transformers

from transformers import pipeline

pipe = pipeline("text2text-generation", "lmqg/mt5-small-koquad-qg")
output = pipe("1990년 영화 《 <hl> 남부군 <hl> 》에서 단역으로 영화배우 첫 데뷔에 이어 같은 해 KBS 드라마 《지구인》에서 단역으로 출연하였고 이듬해 MBC 《여명의 눈동자》를 통해 단역으로 출연하였다.")

Evaluation

• Metric (Question Generation): raw metric file

	Score	Type	Dataset
BERTScore	82.89	default	lmqg/qg_koquad
Bleu_1	25.31	default	lmqg/qg_koquad
Bleu_2	18.59	default	lmqg/qg_koquad
Bleu_3	13.98	default	lmqg/qg_koquad
Bleu_4	10.57	default	lmqg/qg_koquad
METEOR	27.52	default	lmqg/qg_koquad
MoverScore	82.49	default	lmqg/qg_koquad
ROUGE_L	25.64	default	lmqg/qg_koquad

• Metric (Question & Answer Generation, Reference Answer): Each question is generated from the gold answer. raw metric file

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	87.52	default	lmqg/qg_koquad
QAAlignedF1Score (MoverScore)	85.15	default	lmqg/qg_koquad
QAAlignedPrecision (BERTScore)	87.57	default	lmqg/qg_koquad
QAAlignedPrecision (MoverScore)	85.23	default	lmqg/qg_koquad
QAAlignedRecall (BERTScore)	87.49	default	lmqg/qg_koquad
QAAlignedRecall (MoverScore)	85.09	default	lmqg/qg_koquad

• Metric (Question & Answer Generation, Pipeline Approach): Each question is generated on the answer generated by lmqg/mt5-small-koquad-ae. raw metric file

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	80.52	default	lmqg/qg_koquad
QAAlignedF1Score (MoverScore)	82.95	default	lmqg/qg_koquad
QAAlignedPrecision (BERTScore)	77.56	default	lmqg/qg_koquad
QAAlignedPrecision (MoverScore)	79.39	default	lmqg/qg_koquad
QAAlignedRecall (BERTScore)	83.8	default	lmqg/qg_koquad
QAAlignedRecall (MoverScore)	87.02	default	lmqg/qg_koquad

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path: lmqg/qg_koquad
• dataset_name: default
• input_types: ['paragraph_answer']
• output_types: ['question']
• prefix_types: None
• model: google/mt5-small
• max_length: 512
• max_length_output: 32
• epoch: 7
• batch: 16
• lr: 0.001
• fp16: False
• random_seed: 1
• gradient_accumulation_steps: 4
• 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",
}