Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
Yue et al., 2025, Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (arXiv preprint arXiv:2504.13837)
Subscriber deck — free during early access
1 cards in “4.4. Effects of RL Training” · whole deck
4.4. Effects of RL Training
- What happens to pass@1 and pass@256 as RLVR training continues?Training-set pass@1 kept rising (from 26.1 to 42.5) while pass@256 kept falling. More rollouts per prompt (32 instead of 8) helped pass@k slightly, but the base model still won at large \(k\).
- Yue et al., 2025, Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (arXiv preprint arXiv:2504.13837), 4.4. Effects of RL Training
Source
- Yang Yue, Zhiqi Chen, Rui Lu, Andrew Zhao, Zhaokai Wang, Yang Yue, Shiji Song and Gao Huang. 2025. Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?. In arXiv preprint arXiv:2504.13837. https://arxiv.org/abs/2504.13837