About Validation Data

[AlephZr]

Hello, I have a quick question regarding the "Process Validation Data" step. Is the data used here directly sourced from the LLaVA project's Evaluation dataset, or does it require additional preprocessing?

The scripts mention JSON files for Validation Data as follows:

["gqa"]="${data_path}/gqa/conversations.json|${data_path}/gqa/images"

However, I couldn't find a conversations.json file in either the LLaVA project or the original GQA dataset. Was any preprocessing performed to generate this file?

Thank you very much for your help!