ResearchTasksContinual TrainingContinual TrainingFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updated100B English tokens Weak distribution shift (near-tie)Const+Linear1.991Observed Loss5May 21, 2026100B Chinese tokens Strong distribution shiftLinear1.818Observed Loss3May 21, 2026