Model Card of lmqg/mt5-base-esquad-qg-ae

This model is fine-tuned version of google/mt5-base for question generation and answer extraction jointly on the lmqg/qg_esquad (dataset_name: default) via lmqg.

Overview

• Language model: google/mt5-base
• Language: es
• Training data: lmqg/qg_esquad (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="es", model="lmqg/mt5-base-esquad-qg-ae")

# model prediction
question_answer_pairs = model.generate_qa("a noviembre , que es también la estación lluviosa.")

• With transformers

from transformers import pipeline

pipe = pipeline("text2text-generation", "lmqg/mt5-base-esquad-qg-ae")

# answer extraction
answer = pipe("generate question: del <hl> Ministerio de Desarrollo Urbano <hl> , Gobierno de la India.")

# question generation
question = pipe("extract answers: <hl> En la diáspora somalí, múltiples eventos islámicos de recaudación de fondos se llevan a cabo cada año en ciudades como Birmingham, Londres, Toronto y Minneapolis, donde los académicos y profesionales somalíes dan conferencias y responden preguntas de la audiencia. <hl> El propósito de estos eventos es recaudar dinero para nuevas escuelas o universidades en Somalia, para ayudar a los somalíes que han sufrido como consecuencia de inundaciones y / o sequías, o para reunir fondos para la creación de nuevas mezquitas como.")

Evaluation

• Metric (Question Generation): raw metric file

	Score	Type	Dataset
BERTScore	83.97	default	lmqg/qg_esquad
Bleu_1	25.88	default	lmqg/qg_esquad
Bleu_2	17.67	default	lmqg/qg_esquad
Bleu_3	12.84	default	lmqg/qg_esquad
Bleu_4	9.62	default	lmqg/qg_esquad
METEOR	23.11	default	lmqg/qg_esquad
MoverScore	59.15	default	lmqg/qg_esquad
ROUGE_L	24.82	default	lmqg/qg_esquad

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

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	79.67	default	lmqg/qg_esquad
QAAlignedF1Score (MoverScore)	54.82	default	lmqg/qg_esquad
QAAlignedPrecision (BERTScore)	77.14	default	lmqg/qg_esquad
QAAlignedPrecision (MoverScore)	53.27	default	lmqg/qg_esquad
QAAlignedRecall (BERTScore)	82.44	default	lmqg/qg_esquad
QAAlignedRecall (MoverScore)	56.56	default	lmqg/qg_esquad

• Metric (Answer Extraction): raw metric file

	Score	Type	Dataset
AnswerExactMatch	57.98	default	lmqg/qg_esquad
AnswerF1Score	75.33	default	lmqg/qg_esquad
BERTScore	90.04	default	lmqg/qg_esquad
Bleu_1	37.35	default	lmqg/qg_esquad
Bleu_2	32.53	default	lmqg/qg_esquad
Bleu_3	28.86	default	lmqg/qg_esquad
Bleu_4	25.75	default	lmqg/qg_esquad
METEOR	43.74	default	lmqg/qg_esquad
MoverScore	80.94	default	lmqg/qg_esquad
ROUGE_L	49.61	default	lmqg/qg_esquad

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path: lmqg/qg_esquad
• dataset_name: default
• input_types: ['paragraph_answer', 'paragraph_sentence']
• output_types: ['question', 'answer']
• prefix_types: ['qg', 'ae']
• model: google/mt5-base
• max_length: 512
• max_length_output: 32
• epoch: 7
• batch: 32
• lr: 0.001
• fp16: False
• random_seed: 1
• gradient_accumulation_steps: 2
• 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",
}