Expired50% confidenceFactTime unknown
Temperature and Top-k are decoding strategies at the inference stage and do not affect the training process
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Demystifying Transformer: A Word-Continuation Function, Deconstructed
bilibiliAI算法工程师Fiona
Related Claims
UnverifiedWarmup策略在训练初期使用极小学习率,待参数进入合理范围后再提升,被大型语言模型训练广泛采用65% similarUnverifiedEven at high temperature settings, the model is still exploring the semantic space covered by the training distribution and will not produce concepts that have never appeared in any form in the training data.64% similarUnverifiedDuring the pre-training phase, LLMs optimize token prediction probability through maximum likelihood estimation, lacking an inherent mechanism to 'know what they don't know,' and there is no necessary correlation between confidence and accuracy.63% similarVerifiedFeature Stores address the issue of feature computation consistency between training time and inference time, known as training-serving skew.62% similarUnverified计算最优(Compute-Optimal)并不等于集群最优(Cluster-Optimal),真正决定训练效率的因素远不止浮点运算次数61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/54657API
curl https://kongchang.com/api/v1/knowledge/claims/54657MCP
get_claim(id=54657)