Unverified50% confidenceSolutionExact time
Iris adopts an SFT-RL climbing strategy, alternating between supervised fine-tuning and reinforcement learning, where high-quality behavioral trajectories discovered during the RL phase are fed back to the SFT phase as new supervision signals.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring the RL phase, Iris deploys the reward judge and observation summarizer directly inside the training cluster to improve training efficiency.74% similarUnverified后训练通常涵盖监督微调（SFT）、强化学习对齐（RLHF/RLAIF）以及领域适配微调，视觉推理模型还涉及视觉指令微调69% similarUnverifiedModels trained with reinforcement learning (RL) outperform models trained solely with SFT on Agent tasks.64% similarUnverifiedRLHF训练中会加入KL散度惩罚项，防止模型偏离SFT阶段学到的基础行为太远63% similarVerified从预训练检查点到可商用产品需经历监督式指令微调（SFT）和基于人类反馈的强化学习（RLHF）两个后训练阶段63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/915227API
curl https://kongchang.com/api/v1/knowledge/claims/915227MCP
get_claim(id=915227)