Unverified50% confidenceFactExact time
The paper finds that the ranking of downstream tasks is not consistent with the ranking of training loss, and schemes with lower training loss do not necessarily perform better in downstream evaluations.
1
Sources
50%
Confidence
Long-term
Relevance
10/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring the post-training phase, an excessively low loss may correspond to overfitting, where the model memorizes the training data and loses generalization ability.79% similarUnverified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75% 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74% similarUnverifiedRetraining involves a trade-off: some extremely high scores showed a slight decline on cleaner validation distributions, and the team considered it worthwhile to sacrifice perfect benchmark scores in exchange for improved real-world performance.72% similarUnverifiedIn semantic segmentation tasks, deep supervision is typically implemented as a weighted sum of the main loss and auxiliary losses, with the auxiliary loss weights often gradually decayed in the later stages of training.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/971844API
curl https://kongchang.com/api/v1/knowledge/claims/971844MCP
get_claim(id=971844)