[question] why is the LLaVA 1.5 performance different from the original ones?

[WeitaiKang]

Dear authors,

Thanks for your great job!

I am confused about why is the performance of LLaVA1.5 shown in your paper different from the ones published in the original paper https://arxiv.org/pdf/2310.03744 ?

For example, the VQAv2 of LLaVA1.5 is 78.5 in their table 3, but it's 76.54 in your paper? Could you explain where is your reference source? Or do I misunderstand something here?

Looking forward to hearing back from you!