Question About the val Data

[xinwei666]

Hello, thank you for the excellent open-source project. After reviewing your code and the data on Hugging Face, I'm a bit confused. I noticed that the val_multiturn_llava_infoseek.json file contains only 2,500 samples (which are from the training set), while the paper mentions that you used 73k samples for validation. I would like to ask whether I misunderstood something or if I didn't use your data correctly.

[federico1-creator]

Hi @xinwei666 , thank you for your interest in our project.

You can access the evaluation data for Infoseek and Encyclopedic at the following link:
https://github.com/aimagelab/ReflectiVA?tab=readme-ov-file#data-infoseek

Best,
ReflectiVa Team