总览 / 概念 / GRPO
↗ 源 Markdown
概念
grpo rl optimization
创建2026-04-15
更新2026-04-15
阅读量级1 分钟
概念导读

GRPO (Group Relative Policy Optimization) is a reinforcement learning training method that optimizes policy by comparing relative rewards within groups of samples, rather than relying on absolute reward values. GRPO is widely adopted in modern LLM post-training and is referenced across multiple wiki pages including vold, on-policy-self-distillation, and casc...

  1. Related
  2. References

GRPO (Group Relative Policy Optimization)

GRPO (Group Relative Policy Optimization) is a reinforcement learning training method that optimizes policy by comparing relative rewards within groups of samples, rather than relying on absolute reward values. GRPO is widely adopted in modern LLM post-training and is referenced across multiple wiki pages including vold, on-policy-self-distillation, and cascade-rl. It is closely related to grpo-rl-training as a specific training methodology.

  • vold — VOLD uses GRPO in its vision-language on-policy distillation framework
  • on-policy-self-distillation — Self-distillation methods can incorporate GRPO-style optimization
  • cascade-rl — Cascade RL may leverage GRPO for multi-domain post-training
  • grpo-rl-training — Detailed GRPO-based RL training methodology

References

  • Shao et al. (2024). "DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models."