OCR00020: Get the data split Eric used.

[ta4tsering]

Descriptions:
To compare between the different OCR architecture, we need to have a single data split of training, eval and test sets. if we have that we can easily compare between models we trained with a single test split. To do so we need to get all the distribution of the OCR training data that Eric used to train the models that he trained.

Completion Criteria:
Get the splits for

• Lhasa Kanjur
• Derge Tenjur
• Lithang Kanjur
• Google Books
• Norbuketaka
• Betsug
• Durtsa

Subtasks:

• Download the models trained from hugging face
• parse the data distribution if available on the hugging face.
• create data split json for OCR

[ta4tsering]

I have got the data distribution for the GoogleBooks, Norbuketaka, and Betsug. for others the data distribution is not available on the models repo in the hugging face. so I have messaged @eric86y to share those distribution with us and he said he will upload them today.