Does Bakllava checkpoints uses openai/clip-vit-large-patch14 or openai/clip-vit-large-patch14-336 as the vision tower?
In finetune_lora.sh, it uses --vision_tower openai/clip-vit-large-patch14.
In finetune.sh, it uses --vision_tower openai/clip-vit-large-patch14-336.