Training dataset in single image llava onevision model

[ZhaoyangLi-nju]

I would like to know whether training single image version of llava-ov needs to read the training data in the order in single_image.yaml, or whether it is random.