Unverified50% confidenceFactExact time
In the joint pre-training stage, one budget is randomly sampled per micro-batch and only its next-token prediction loss is optimized, with no cross-budget consistency objective added.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe experiments used three training random seeds paired with a fixed test set split, along with a limited hyperparameter tuning budget.74% similarUnverifiedBetween the two runs the team adjusted the generation budget (the first run occasionally hit the token limit before answering), so part of the fluctuation may come from the run framework rather than the model's judgment itself.72% similarUnverified采用域随机化（在训练早期就让ping在整个范围内随机采样）而非单向退火，可以让每个版本模型从一开始就见过各种延迟条件，避免后期崩溃风险71% similarUnverified在稠密检索中，批内负样本训练要求批次包含有信息量的负样本，随机采样会导致训练损失快速收敛但泛化性能差70% similarUnverified时间序列预测不能简单随机划分训练集和测试集，必须按时间顺序切分以避免数据泄露69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098758API
curl https://kongchang.com/api/v1/knowledge/claims/1098758MCP
get_claim(id=1098758)