TY  - RPRT
TI  - Why Does RL Generalize Better Than SFT? A Data-Centric Perspective on VLM Post-Training
AU  - Aojun Lu
AU  - Tao Feng
AU  - Hangjie Yuan
AU  - Wei Li
AU  - Yanan Sun
PY  - 2026
UR  - https://arxiv.org/abs/2602.10815
ID  - 2602.10815
ER  -