We provide an experimental version of the vBPE dictionary for the research community.
Based on this, you can skip the vBPE training of Being-VL-0.5 and directly proceed with subsequent model development.
Please note that this is a visual BPE dictionary pre-trained on a small image dataset, intended only for rapid development and verification. For better performance, we still recommend that you train your own visual BPE dictionary based on your datasets.