TY  - RPRT
TI  - Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning
AU  - Yixuan Even Xu
AU  - Yash Savani
AU  - Fei Fang
AU  - J. Zico Kolter
PY  - 2026
UR  - https://arxiv.org/abs/2504.13818
ID  - 2504.13818
ER  -