ResearchTasksSingle-stream decodeFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedSmolLM2-135MvLLM cudagraph8.65Latency (ms/token)6Jun 9, 2026622.9 MB 4-layer modelCUDA-graph eager1.08Latency (ms/token)4Jun 9, 2026