@misc{indiciae3b7bcfd64f6c, title = {Learning What to Distill: Bilevel Top-K Token Selection for Self-Distillation in Large Language Models}, author = {Heng Liang and Xinwen Zhang and Hongchang Gao}, year = {2026}, url = {https://searcharxiv.com/abs/2610.07247}, note = {Source identifier: 2610.07247} }