Model Card of lmqg/bart-large-squad-qg

This model is fine-tuned version of facebook/bart-large for question generation task on the lmqg/qg_squad (dataset_name: default) via lmqg.

Overview

• Language model: facebook/bart-large
• Language: en
• Training data: lmqg/qg_squad (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="en", model="lmqg/bart-large-squad-qg")

# model prediction
questions = model.generate_q(list_context="William Turner was an English painter who specialised in watercolour landscapes", list_answer="William Turner")

• With transformers

from transformers import pipeline

pipe = pipeline("text2text-generation", "lmqg/bart-large-squad-qg")
output = pipe("<hl> Beyonce <hl> further expanded her acting career, starring as blues singer Etta James in the 2008 musical biopic, Cadillac Records.")

Evaluation

• Metric (Question Generation): raw metric file

	Score	Type	Dataset
BERTScore	91	default	lmqg/qg_squad
Bleu_1	58.79	default	lmqg/qg_squad
Bleu_2	42.79	default	lmqg/qg_squad
Bleu_3	33.11	default	lmqg/qg_squad
Bleu_4	26.17	default	lmqg/qg_squad
METEOR	27.07	default	lmqg/qg_squad
MoverScore	64.99	default	lmqg/qg_squad
ROUGE_L	53.85	default	lmqg/qg_squad

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

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	95.54	default	lmqg/qg_squad
QAAlignedF1Score (MoverScore)	70.82	default	lmqg/qg_squad
QAAlignedPrecision (BERTScore)	95.59	default	lmqg/qg_squad
QAAlignedPrecision (MoverScore)	71.13	default	lmqg/qg_squad
QAAlignedRecall (BERTScore)	95.49	default	lmqg/qg_squad
QAAlignedRecall (MoverScore)	70.54	default	lmqg/qg_squad

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

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	93.23	default	lmqg/qg_squad
QAAlignedF1Score (MoverScore)	64.76	default	lmqg/qg_squad
QAAlignedPrecision (BERTScore)	93.13	default	lmqg/qg_squad
QAAlignedPrecision (MoverScore)	64.98	default	lmqg/qg_squad
QAAlignedRecall (BERTScore)	93.35	default	lmqg/qg_squad
QAAlignedRecall (MoverScore)	64.63	default	lmqg/qg_squad

• Metrics (Question Generation, Out-of-Domain)

Dataset	Type	BERTScore	Bleu_4	METEOR	MoverScore	ROUGE_L	Link
lmqg/qg_squadshifts	amazon	90.93	6.53	22.3	60.87	25.03	link
lmqg/qg_squadshifts	new_wiki	93.23	11.12	27.32	66.23	29.68	link
lmqg/qg_squadshifts	nyt	92.49	8.12	25.25	64.06	25.29	link
lmqg/qg_squadshifts	reddit	90.95	5.95	21.5	60.59	22.37	link
lmqg/qg_subjqa	books	88.07	0.63	11.58	55.56	12.37	link
lmqg/qg_subjqa	electronics	87.83	0.87	15.35	56.35	16.02	link
lmqg/qg_subjqa	grocery	87.79	0.53	15.13	57.02	12.34	link
lmqg/qg_subjqa	movies	87.49	0.0	11.86	55.29	12.51	link
lmqg/qg_subjqa	restaurants	87.98	0.0	12.42	55.43	13.08	link
lmqg/qg_subjqa	tripadvisor	88.91	0.0	13.72	56.05	14.03	link

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path: lmqg/qg_squad
• dataset_name: default
• input_types: ['paragraph_answer']
• output_types: ['question']
• prefix_types: None
• model: facebook/bart-large
• max_length: 512
• max_length_output: 32
• epoch: 4
• batch: 32
• lr: 5e-05
• 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",
}