LLaVA-NeXT icon indicating copy to clipboard operation
LLaVA-NeXT copied to clipboard

Published 20 hours ago • verified publisher icon LLaVA-VL

Why plain prompt version in pretraining stage?

Open serwansj opened this issue 2 months ago • 0 comments

Hi,

is the prompt version in the pre-training stage of onevision (see https://github.com/LLaVA-VL/LLaVA-NeXT/blob/main/scripts/train/pretrain_clip.sh) set to plain on purpose? Should it not be qwen_2? If it is done on purpose could you explain why you decide not to use a prompt template in the pre-training stage?

serwansj avatar Sep 10 '25 00:09 serwansj