Unverified50% confidenceFactExact time
Qwen2.5后训练首次在在线阶段使用GRPO，GRPO是DeepSeek团队提出的PPO轻量化变体，去掉独立价值网络，用组内平均奖励作为基线
1
Sources
50%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
UnverifiedGRPO is a reinforcement learning algorithm proposed by the DeepSeek team that eliminates the Critic Model from PPO, computing intra-group relative rewards as a baseline by generating a set of responses to the same question.79% similarVerifiedGRPO (Group Relative Policy Optimization), proposed by DeepSeek, requires no separate training of a value network and has become the mainstream algorithm choice for current agentic RL training.75% similarVerifiedGRPO由DeepSeek团队系统化提出并在训练DeepSeek-R1系列模型时得到大规模验证74% similarUnverifiedGRPO和GSPO均是PPO算法的改进变体，DeepSeek-R1凭借GRPO在推理能力上实现大幅跃升74% similarUnverifiedDeepSeek extensively uses the GRPO algorithm in place of traditional RLHF during the training phase to reduce training costs.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/585439API
curl https://kongchang.com/api/v1/knowledge/claims/585439MCP
get_claim(id=585439)