How are Table 3 “Response Length” and “think ratio” computed？

[lmsdss]

Hi，thanks for the great work and for releasing the code!

Could you please clarify how Response Length and think ratio in Table 3 are computed in the code?

[sming256]

Thanks for your interest in our work!

In Table 3, Response Length is computed as the average number of output tokens produced by VideoAuto-R1 across the evaluated benchmarks (VideoMME, MVBench, LongVideoBench, MMVU, VideoMMMU, and MVP).

This metric is not included in the currently released code. You can save the model outputs during evaluation and then compute the token counts offline. Similarly, the think ratio can be obtained by measuring the fraction of outputs that contain the tag.

Hope this clarifies your concern.