TY  - RPRT
TI  - On-Policy Self-Training: Data-Free Self-Improvement in Large Language Models
AU  - Leon Sick
AU  - Lukas Hoyer
AU  - Yannick Strümpler
AU  - Christian Reisswig
AU  - Timo Ropinski
AU  - Federico Tombari
PY  - 2026
UR  - https://arxiv.org/abs/2610.32770
ID  - 2610.32770
ER  -