Unverified50% confidenceFactExact time
GRPO、PPO等算法在大语言模型后训练中被成功应用
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified推理增强模型训练层面通常借助GRPO（Group Relative Policy Optimization）等强化学习算法，以最终答案正确性作为奖励信号71% similarUnverifiedQwen3采用了类GRPO算法进行推理能力的强化训练67% similarUnverifiedJustGRPO的核心修复方案是在自回归顺序下训练GRPO，推理阶段保持并行解码67% similarUnverified推理增强方法通常借助RLHF或GRPO等算法，在MATH、HumanEval、AIME等基准上取得突破63% similarUnverifiedProximal Policy Optimization（PPO）是机器人强化学习的主力策略梯度算法工具62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/829054API
curl https://kongchang.com/api/v1/knowledge/claims/829054MCP
get_claim(id=829054)