Unverified80% confidenceFactTime unknown
Hermes模型使用DPO（直接偏好优化）等对齐技术进行训练
1
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedAlignment algorithms such as DPO (Direct Preference Optimization) reduce training costs while improving the consistency of model performance on open-ended instructions.77% similarUnverified训练SWE-bench智能体的先进方法包括拒绝采样（Rejection Sampling）、DPO（Direct Preference Optimization）以及过程奖励模型（Process Reward Model）70% similarUnverified后训练优化包括指令微调、RLHF或DPO策略调整、系统提示词工程改进及工具调用和多模态融合能力强化66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/28647API
curl https://kongchang.com/api/v1/knowledge/claims/28647MCP
get_claim(id=28647)