On the type of visual long-thought data

[jhdonghj]

Thanks for your excellent work!

In the paper there are two types of visual long-thought data, one is generated by QVQ and another is generated by self distillation.

I'm curious which type of visual long-thought data was released as "RUC-AIBOX/Virgo-Visual-Long-Thought-Dataset". Could you clarify it?

[Richar-Du]

Thanks for your interest. All the data in "RUC-AIBOX/Virgo-Visual-Long-Thought-Dataset" are distilled from Virgo-72B instead of QVQ, because according to the results in Table 2, the data distilled from Virgo-72B is better than those distilled from QVQ.