Do you have a plan to support Chinese?

[asahi417]

Hi, we are eager to extend the languages we support. In fact, as long as there is a good QA dataset in the language, we can train QAG model on that language easily, but for Chinese, we didn't find a good QA dataset so far, so that's the current bottleneck.

[AntonOfTheWoods]

@yanshuaibupt , I am also very interested in doing this for Chinese, so if you would like to collaborate on this then get in touch!

[Antlerkeke]

Hi, we are eager to extend the languages we support. In fact, as long as there is a good QA dataset in the language, we can train QAG model on that language easily, but for Chinese, we didn't find a good QA dataset so far, so that's the current bottleneck.

May you can use this:https://huggingface.co/datasets/lijingxin/squad_zen

[asahi417]

Hi, we are eager to extend the languages we support. In fact, as long as there is a good QA dataset in the language, we can train QAG model on that language easily, but for Chinese, we didn't find a good QA dataset so far, so that's the current bottleneck.

May you can use this:https://huggingface.co/datasets/lijingxin/squad_zen

Thanks for sharing the resource! I'll have a look the dataset, and soon start training if it's in the same format as the other datasets!

[asahi417]

We are training QAG models on the Chinese SQuAD, and they should be ready by the end of this week!

[asahi417]

FYI, these are the datasets formatted into our QG and QAG dataset. https://huggingface.co/datasets/lmqg/qag_zhquad https://huggingface.co/datasets/lmqg/qg_zhquad

[asahi417]

Hi, I'll close this thread and open another one specific for the new language request, so please follow that one instead.

https://github.com/asahi417/lm-question-generation/issues/20

Meanwhile Chinese QAG is already available on https://autoqg.net/ and lmqg now! With lmqg, you can use it as below.

from lmqg import TransformersQG

model = TransformersQG(language="zh", model_ae="lmqg/mt5-small-zhquad-ae", model="lmqg/mt5-small-zhquad-qg")
context = "与转导或结合不同，转化依赖于大量的细菌基因产物，这些基因产物专门相互作用来完成这个复杂的过程，因此转化显然是细菌对DNA转移的适应。为了使细菌结合、吸收供体DNA并将其重组为自己的染色体，它必须首先进入一种称为能力的特殊生理状态（见自然能力）。在枯草芽孢杆菌中，大约40个基因是培养能力所必需的。枯草芽孢杆菌转化过程中转移的DNA长度可以在染色体的三分之一到整个染色体之间。转化在细菌物种中似乎很常见，到目前为止，已知至少有60种物种具有自然转化能力。自然界能力的发展通常与应激性环境条件有关，似乎是一种促进受体细胞DNA损伤修复的适应。"
model.generate_qa(context)
[('在染色体中发现的DNA长度是多少?', '枯草芽孢杆菌转化过程中转移的DNA长度可以在染色体的三分之一到整个染色体之间。')]

We're training a few more models that can potentially improve the quality, and will announce it once they are ready.