Are you planning to compare your method with GSPO?

[Saoyu99]

I noticed that GSPO, which came out around the same time as your work, also makes adjustments to importance sampling ratio.
Do you have any plans to do a comparison with GSPO?

[callsys]

Hello, our method differs from GSPO primarily in the clipping strategy. We adopt token-level clipping (as in DeepSeek-Math), whereas GSPO employs sequence-level clipping (as in DeepSeek-R1). In our study, we compared these two approaches and found that while their overall performance is comparable, token-level clipping provides greater stability under the same clipping threshold.