Unverified50% confidenceDecision RuleExact time
In GRPO baseline replacement experiments, the group size must be fixed; otherwise the variance of the mean baseline changes with sample size and confounds the results.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGRPO's supervision granularity is sequence-level, with the entire response sharing a single advantage value, so it cannot indicate which step in the reasoning chain went wrong.71% similarUnverifiedGRPO generates a group of candidate outputs for the same input and updates the model using intra-group relative reward signals rather than absolute reward signals, reducing reliance on an independent value network (critic).65% similarUnverifiedGRPO's original paper and mainstream reproductions all use the within-group mean as the baseline, without exploring alternatives in depth.65% similarUnverified验证集和测试集不应做任何数据增强，一旦施加变换会引入系统性偏差，破坏对模型泛化能力的无偏估计63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1109172API
curl https://kongchang.com/api/v1/knowledge/claims/1109172MCP
get_claim(id=1109172)