Dear author, Thank you for the great work on LLaVA-3D May I ask how how much GPU memory is required to train the LLaVA-3D model using finetune.sh? Appreciate your help!