Distillation of Distillation Is Not Distillation

HanchungLee · x · 2026-07-16

The core argument here is: "Distillation of distillation is not distillation."

While concise, the main point is that after multiple rounds of compression or retraining, the model's behavior is no longer equivalent to the original distillation process.

Original post →

More from Research

Research channel →