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

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

Overview

• Language model: google/mt5-base
• Language: it
• Training data: lmqg/qg_itquad (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="it", model="lmqg/mt5-base-itquad-qg-ae")

# model prediction
question_answer_pairs = model.generate_qa("Dopo il 1971 , l' OPEC ha tardato ad adeguare i prezzi per riflettere tale deprezzamento.")

• With transformers

from transformers import pipeline

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

# answer extraction
answer = pipe("generate question: <hl> Dopo il 1971 <hl> , l' OPEC ha tardato ad adeguare i prezzi per riflettere tale deprezzamento.")

# question generation
question = pipe("extract answers: <hl> Il 6 ottobre 1973 , la Siria e l' Egitto, con il sostegno di altre nazioni arabe, lanciarono un attacco a sorpresa su Israele, su Yom Kippur. <hl> Questo rinnovo delle ostilità nel conflitto arabo-israeliano ha liberato la pressione economica sottostante sui prezzi del petrolio. All' epoca, l' Iran era il secondo esportatore mondiale di petrolio e un vicino alleato degli Stati Uniti. Settimane più tardi, lo scià d' Iran ha detto in un' intervista: Naturalmente[il prezzo del petrolio] sta andando a salire Certamente! E come! Avete[Paesi occidentali] aumentato il prezzo del grano che ci vendete del 300 per cento, e lo stesso per zucchero e cemento.")

Evaluation

• Metric (Question Generation): raw metric file

	Score	Type	Dataset
BERTScore	81.15	default	lmqg/qg_itquad
Bleu_1	23.3	default	lmqg/qg_itquad
Bleu_2	15.39	default	lmqg/qg_itquad
Bleu_3	10.74	default	lmqg/qg_itquad
Bleu_4	7.72	default	lmqg/qg_itquad
METEOR	18.56	default	lmqg/qg_itquad
MoverScore	57.15	default	lmqg/qg_itquad
ROUGE_L	22.81	default	lmqg/qg_itquad

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

	Score	Type	Dataset
QAAlignedF1Score (BERTScore)	81.98	default	lmqg/qg_itquad
QAAlignedF1Score (MoverScore)	56.35	default	lmqg/qg_itquad
QAAlignedPrecision (BERTScore)	81.19	default	lmqg/qg_itquad
QAAlignedPrecision (MoverScore)	56	default	lmqg/qg_itquad
QAAlignedRecall (BERTScore)	82.83	default	lmqg/qg_itquad
QAAlignedRecall (MoverScore)	56.75	default	lmqg/qg_itquad

• Metric (Answer Extraction): raw metric file

	Score	Type	Dataset
AnswerExactMatch	60.7	default	lmqg/qg_itquad
AnswerF1Score	74.04	default	lmqg/qg_itquad
BERTScore	91.12	default	lmqg/qg_itquad
Bleu_1	40.14	default	lmqg/qg_itquad
Bleu_2	34.56	default	lmqg/qg_itquad
Bleu_3	30.56	default	lmqg/qg_itquad
Bleu_4	26.87	default	lmqg/qg_itquad
METEOR	43.51	default	lmqg/qg_itquad
MoverScore	82.62	default	lmqg/qg_itquad
ROUGE_L	45.82	default	lmqg/qg_itquad

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path: lmqg/qg_itquad
• 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: 13
• batch: 32
• lr: 0.0005
• 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",
}