ResearchTasksPrefixLM AttentionPrefixLM AttentionFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedLlama3.1 405B (q=128, k=8) (1k)CuBridge122.22PrefixLM Attention TFLOPS (1k)4May 7, 2026Qwen2.5 72B (q=64, k=8) (1k)CuBridge103.61PrefixLM Attention Throughput (TFLOPS)4May 7, 2026Llama2-7b (q=32, k=32) (8k)CuBridge163.7TFLOPS (PrefixLM Attention)4May 7, 2026