[KongchangAI]
Unverified50% confidenceFactExact time

在PPO和GRPO算法中,重要性采样校正权重正比于新旧策略概率之比exp(logprob_new - logprob_old),由于指数函数放大,0.001的logprob偏差会产生可观的权重误差

1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen

Sources

Related Entities

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/972290
API
curl https://kongchang.com/api/v1/knowledge/claims/972290
MCP
get_claim(id=972290)