ResearchDatasetsClimbMixClimbMixFollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsIntrinsic Tokenization EvaluationClimbMix (held-out)99Vocabulary Utilization24Language ModelingClimbMix FineWeb-Edu 1M-token (val)30.12PPL20Language ModelingClimbMix pretraining corpus (val held-out)0.712BPB (val)14