There is a problem with the dataset downloaded from Hugging Face.

[whu125]

Thank you for your excellent work！

I downloaded the dataset from huggingface, and when I loaded test-00000-of-00001.parquet, I found some abnormal data. Among them, 7 entries have duplicate option0 and option1 values, and the id also has an extra prefix. Did I do something wrong?

video_id @healthfood-7352910847997496619
id @healthfood-7352910847997496619_0

[teowu]

This is strange. I will check on that. Meanwhile, please use the lvb_val.json and lvb_test_wo_gt.json for evaluation first.

[whu125]

it seems that lvb_test_wo_gt.json has the same problem.