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

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

Overview

• Language model: google/mt5-small
• 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/mt5-small-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/mt5-small-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	90.01	default	lmqg/qg_squad
Bleu_1	54.07	default	lmqg/qg_squad
Bleu_2	37.62	default	lmqg/qg_squad
Bleu_3	28.18	default	lmqg/qg_squad
Bleu_4	21.65	default	lmqg/qg_squad
METEOR	23.83	default	lmqg/qg_squad
MoverScore	62.75	default	lmqg/qg_squad
ROUGE_L	48.95	default	lmqg/qg_squad

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

Dataset	Type	BERTScore	Bleu_4	METEOR	MoverScore	ROUGE_L	Link
lmqg/qg_dequad	default	73.53	0.0	4.81	50.37	1.56	link
lmqg/qg_esquad	default	74.94	0.59	6.02	50.62	5.21	link
lmqg/qg_frquad	default	72.91	1.71	8.24	50.96	15.84	link
lmqg/qg_itquad	default	72.6	0.54	5.89	50.23	5.01	link
lmqg/qg_jaquad	default	66.08	0.0	0.51	46.53	6.08	link
lmqg/qg_koquad	default	66.34	0.0	0.73	45.86	0.06	link
lmqg/qg_ruquad	default	70.89	0.0	1.78	49.1	0.99	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: google/mt5-small
• max_length: 512
• max_length_output: 32
• epoch: 15
• batch: 64
• lr: 0.0005
• fp16: False
• random_seed: 1
• gradient_accumulation_steps: 1
• 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",
}