-
Notifications
You must be signed in to change notification settings - Fork 59
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Dataset used for finetuning mt5 model #43
Comments
Yes, I only use CSpider (Chinese version of Spider) dataset to fine-tune mT5, which contains 7000 (+1659) training examples. However, honestly, I don't know how much data you need to prepare. Ultimate performance depends on the quality and quantity of the training data as well as the capabilities of the foundation model (e.g., we use T5 for English Spider and mT5 for Chinese CSpider). In my experiments, I found that the Chinese capability of mT5 is not that strong, which may be due to the existence of the "curse of multilinguality". Therefore, choosing a suitable and powerful Persian language model is also important. |
For a detailed description of the term "curse of multilinguality", please refer to https://aclanthology.org/2020.acl-main.747.pdf. |
Thank you for your answer and the heads up! |
Hi
First of all, thank you for your great work on this project. You've reached among best results on Spider benchmark and your clear and complete readme file allowed me to run your code very easily.
I want to see if I can finetune a text2natsql model on mt5 like you did on CSpider. I was wondering how much data I have to create as I want to create a dataset like CSpider but in Persian languge.
Was CSpider the only dataset used for finetuning mt5 backbone or other datasets were also used?
The text was updated successfully, but these errors were encountered: