ResearchDatasetsSequence lengthFollowBenchmarksTask NameDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyTask NameDataset NameSOTA ResultTrendResultsHardware Sequence Processing PerformanceSequence length n=64 at 10 GBaud0Latency (s)6KV-Cache Footprint AnalysisSequence length T=2048 BF16 (inference)67KV Cache Size (MB)4