ResearchTasksGeneral speculative decoding performanceGeneral speculative decoding performanceFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedMean (MT-Bench, HumanEval, GSM8K)PPOW6.52Average Acceptance Length (τ)112Jun 1, 2026