ai4bharat/samanantar · Unable to download data for specific languages

sam_df = datasets.load_dataset('ai4bharat/samanantar',
                               language='or',
                               trust_remote_code=True
                               )['train'].to_pandas()

Ideally, this should download the data for Odia language only. When I execute it, it shows that it has to download 7GB of data.
Now, if I change the language to Assamese ('or), it shows that it has to download 7GB of data.
How is this possible given that the number of rows for Odia and Assamese have a 7-fold difference?
No matter what's the language, it's downloading 7GB of data (which I suppose is the entire dataset).

Any guidance is appreciated.