Tokenizer padding issue

[vishaal27]

Hi, thanks for your great work!

I am trying to run quality scoring on image-caption pairs with your provided script. When I run it, I get this error:

ValueError: You are attempting to perform batched generation with padding_side='right' this may lead to unexpected behaviour for Flash Attention version of Qwen2. Make sure to  call `tokenizer.padding_side  = 'left'` before tokenizing the input. 

I guess that is because of this line:

UniFilter/data_scoring/data_quality_classifier_caption_scoring.py

Line 140 in 27d0893

model.config.tokenizer_padding_side = tokenizer.padding_side = "right"

I was able to fix it by fixing the padding to the left using:

model.config.tokenizer_padding_side = tokenizer.padding_side = "left"

I just wanted to confirm that this is indeed the fix, and that there is a bug in the original script?

Thanks!