Interpolated Policy Distillation: A Controllable Continuum Between Off-Policy and On-Policy Distillation
Interpolated Policy Distillation is introduced, which defines the next-token distribution at every decoding step as an explicit linear interpolation between the student and teacher distributions, demonstrating that token-level policy interpolation better balances trajectory quality and student learnability.
You-Xu Shi, Yi-Fan Sun, Da-Cheng Yin et al.
· 0 citations