How to train model for other language?

[wushixian]

The package is really good, but I want to use it for other language, and I tried to modify the model path and use other T5 model, but the model cannot be loaded? May I know how to train a model from scratch? Thanks a lot.

[Shivanandroy]

@wushixian: KeyPhraseTransformer, under the hood, uses a T5 model (but trained in a specific way). If you want to extract keywords from other languages, I would suggest the following:

• First, train a mT5 model on the dataset of your language. You can use SimpleT5 (https://github.com/Shivanandroy/simpleT5) to train it in just 3 lines of code
• Format of your dataset should be - (in target text, keywords must be separated with pipe " | ": source_text = "Wushixian is a NLP expert." target_text: "Wushixian | NLP expert"
• Then, you can use KeyPhraseTransformer to use this model as below:

  from keyphrasetransformer import KeyPhraseTransformer
  kp = KeyPhraseTransformer(model_type="mt5", model_name="<your trained mT5 model path>")

[wushixian]

Thanks a lot, I'll try to train a model.

[z-arabi]

Hello,

I tried to fine tune this model on persian language. I have a question. What will happen when we call this function? It didn't take me long time to execute. How exactly the fine tuning happens in this model?

kp = KeyPhraseTransformer(model_type="mt5", model_name="folder")