英文

lmqg/mbart-large-cc25-dequad-qag 的模型卡片

这个模型是在 lmqg/qag_dequad (数据集名称: default)上,通过 lmqg facebook/mbart-large-cc25 进行了细调,用于问答对生成任务。

概述

用法

from lmqg import TransformersQG

# initialize model
model = TransformersQG(language="de", model="lmqg/mbart-large-cc25-dequad-qag")

# model prediction
question_answer_pairs = model.generate_qa("das erste weltweit errichtete Hermann Brehmer 1855 im niederschlesischen ''Görbersdorf'' (heute Sokołowsko, Polen).")
  • 使用 transformers
from transformers import pipeline

pipe = pipeline("text2text-generation", "lmqg/mbart-large-cc25-dequad-qag")
output = pipe("Empfangs- und Sendeantenne sollen in ihrer Polarisation übereinstimmen, andernfalls wird die Signalübertragung stark gedämpft. ")

评估

Score Type Dataset
QAAlignedF1Score (BERTScore) 69.25 default 12313321
QAAlignedF1Score (MoverScore) 50.71 default 12313321
QAAlignedPrecision (BERTScore) 70.69 default 12313321
QAAlignedPrecision (MoverScore) 51.81 default 12313321
QAAlignedRecall (BERTScore) 68.05 default 12313321
QAAlignedRecall (MoverScore) 49.78 default 12313321

训练超参数

在微调过程中使用了以下超参数:

  • 数据集路径:lmqg/qag_dequad
  • 数据集名称:default
  • 输入类型:['paragraph']
  • 输出类型:['questions_answers']
  • 前缀类型:None
  • 模型:facebook/mbart-large-cc25
  • 最大长度:512
  • 输出最大长度:256
  • 迭代轮数:6
  • 批次大小:2
  • 学习率:0.0001
  • fp16:False
  • 随机种子:1
  • 梯度累积步数:32
  • 标签平滑:0.15

完整配置见 fine-tuning config file

引用

@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",
}