Unverified50% confidenceBenchmarkExact time
Penalizing automaton-rejected tokens during training can improve development-set Smatch++ by 3.6 to 11.0 points after 10% of training is completed, but this advantage gradually diminishes as training progresses
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedDuring the post-training phase, an excessively low loss may correspond to overfitting, where the model memorizes the training data and loses generalization ability.72% similarUnverified小版本号跃迁通常对应后训练（Post-Training）阶段的优化，而非预训练的重新训练72% similarUnverified指令微调不需要从头训练，通常仅需基础训练计算量的1%以下即可显著改变模型行为模式71% similarUnverified分配给预训练的算力越多，下游任务的pass@1奖励就越高，预训练损失越低模型在后续RL中获得的收益越大70% similarUnverified解决灾难性遗忘的方案是在训练时混入10%-20%的原始预训练数据70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002041API
curl https://kongchang.com/api/v1/knowledge/claims/1002041MCP
get_claim(id=1002041)