ResearchTasksMulti-Head Attention Operator DevelopmentMulti-Head Attention Operator DevelopmentFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedMHA (hd=64, sl=1024) on A100 GPULLM-TL175.6TFLOPS2Feb 26, 2026