Verified65% confidenceFactExact time
At the training level, reasoning-enhanced models typically leverage reinforcement learning algorithms such as GRPO (Group Relative Policy Optimization), using the correctness of the final answer as the reward signal
3
Sources
65%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
VerifiedGRPO (Group Relative Policy Optimization) is an improved variant of PPO that requires no additional value network training; instead, it calculates the reward signal by sampling a group of responses to the same question and using intra-group relative performance as a baseline.78% similarUnverifiedWhen GRPO is used for long-form RAG training, the answer-level scalar advantage value is evenly distributed to each sentence in the answer, causing copied sentences and genuinely grounded sentences to receive the same reinforcement signal.76% similarUnverifiedGRPO is a reinforcement learning method that requires no value network; it updates the policy by sampling a group of responses to the same prompt and using the within-group relative advantage as the reward signal, and it performs well on math and reasoning tasks, but its reward signal itself carries no geometric information.74% 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73% similarVerifiedThe MTP (Multi-Token Prediction) auxiliary training objective allows the model to be used with speculative decoding during inference to improve throughput72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/583801API
curl https://kongchang.com/api/v1/knowledge/claims/583801MCP
get_claim(id=583801)