[Question] About training data

[Nonoor]

Question

Hello, thank you for open sourcing such interesting work!
In the paper, it is mentioned that "We fine-tune the whole model using the exact visual instruction data from LLaVA-1.5 and LLaVA-NeXT, respectively." Does this implies that the m3 model is indeed fine-tuned using the LLaVA-NeXT data? However, it appears that the LLaVA-NeXT dataset is not openly available for public use. Can you open source the data you use for finetuning?