SPG+IM outperforms SPG+entropy in every accuracy cell across four tasks and three generation lengths, with the largest gaps on planning tasks. SPG+entropy also underperforms unmodified SPG in every cell, a negative result showing that preferentially masking individually predictable tokens can be worse than uniform random masking.

SPG+IM outperforms SPG+entropy in every accuracy cell across four tasks and three generation lengths, with the largest gaps on planning tasks. SPG+entropy also underperforms unmodified SPG in every cell, a negative result showing that preferentially masking individually predictable tokens can be worse than uniform random masking. · CiteArk

测量结果

已评估 0/36 项

获得支持 0有争议或冲突 0尚未评估 36

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
model: SPG · dataset: GSM8K · generationLengthTokens: 128	78.5%	—	尚未评估
model: SPG · dataset: GSM8K · generationLengthTokens: 256	84.9%	—	尚未评估
model: SPG · dataset: GSM8K · generationLengthTokens: 512	84.2%	—	尚未评估
model: SPG · dataset: MATH-500 · generationLengthTokens: 128	34.6%	—	尚未评估
model: SPG · dataset: MATH-500 · generationLengthTokens: 256	39.6%	—	尚未评估
model: SPG · dataset: MATH-500 · generationLengthTokens: 512	42.6%	—	尚未评估
model: SPG · dataset: Countdown · generationLengthTokens: 128	68.8%	—	尚未评估
model: SPG · dataset: Countdown · generationLengthTokens: 256	70.7%	—	尚未评估
model: SPG · dataset: Countdown · generationLengthTokens: 512	70.3%	—	尚未评估
model: SPG · dataset: Sudoku · generationLengthTokens: 128	82.9%	—	尚未评估
model: SPG · dataset: Sudoku · generationLengthTokens: 256	94%	—	尚未评估
model: SPG · dataset: Sudoku · generationLengthTokens: 512	93.1%	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果