ResearchDatasetsMoE modelsMoE modelsFollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsKernel Throughput EvaluationMoE Models (OLMoE, Qwen3, DSv3, Mixtral) beta=0.567Latency12Expert Pruning EfficiencyMoE Models0.22Calibration Time (h)6Loss curve fitting across model sizesMoE models (various sizes)0.341ASMT MAPE3