Unverified50% confidenceFactExact time
针对特定工具做专项强化训练的副作用是对非训练分布工具的泛化能力下降，学术上称为分布外泛化失败
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Related Claims
Unverified训练-部署鸿沟指模型因训练与部署环境的数据分布偏移、软件依赖版本差异或硬件指令集不同而导致的性能下降76% similarUnverifiedBehavioral cloning suffers from the distribution shift problem: during training the policy only encounters the state distribution covered by expert trajectories, and during deployment the accumulation of small deviations causes the actual state distribution visited to gradually diverge from the training distribution.76% similarUnverifiedrank值过高、学习率设置不当或训练步数过多是LoRA训练过拟合的常见诱因71% similarUnverifiedA 2017 study by Keskar et al. showed that large-batch training tends to converge to sharp minima corresponding to worse generalization, while small-batch training tends toward flat minima.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/126883API
curl https://kongchang.com/api/v1/knowledge/claims/126883MCP
get_claim(id=126883)