Unverified50% confidenceFactExact time
GRPO随DeepSeek系列模型的成功而广受关注，尤其在提升数学推理和代码生成能力方面表现突出
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/29/2026
First Seen
Valid until: 11/27/2026
Sources
Related Entities
Related Claims
VerifiedDeepSeek-R1通过在代码和数学两个可验证域上大规模运行GRPO算法取得成功79% similarUnverifiedDeepSeek团队报告，仅用GRPO在数学和代码数据上进行强化学习而不经过监督微调，模型便自发出现反思、回溯验证等链式推理行为（aha moment）76% 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72% similarUnverifiedDeepSeek explored the GRPO algorithm as an attempt to improve upon the original RLHF framework.71% 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70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/817571API
curl https://kongchang.com/api/v1/knowledge/claims/817571MCP
get_claim(id=817571)