Dataset Loader didn't work properly on Kaggle
wdprsto opened this issue · comments
Good afternoon,
This morning I was trying to run Donut on Kaggle. The structure of the dataset is similar with the one defined on the documentation. However, when I am trying train the model, an error occurred, saying that the "ground truth" didn't exist. While checking on the sample, it shows that the load_dataset recognize the folder name as label and ignore the metadata.jsonl file inside the folder.
I can read the jsonl file via command, tho.
I prepare the Donut with this code:
!git clone https://github.com/clovaai/donut.git
!cd donut && pip install .
Thank you for your help
Solved by installing datasets ver 2.4
ref