GRPO can suffer from Lazy Likelihood Displacement, where correct responses gain little or lose likelihood, and selectively downweighting penalties on harmful tokens (NTHR) mitigates this and boosts math reasoning performance.
The local elasticity of neural networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the Effect of Negative Gradient in Group Relative Deep Reinforcement Optimization
GRPO can suffer from Lazy Likelihood Displacement, where correct responses gain little or lose likelihood, and selectively downweighting penalties on harmful tokens (NTHR) mitigates this and boosts math reasoning performance.