ResearchBenchmarksSpeculative Decoding on Spec-Bench Llama2-7B v1.0 (test)Follow2.73MT ScoreUniVer2.45962.52982.62.6702May 6, 2026Evaluation ResultsMethodMethodLinksMT ScoreTranslation ScoreSummarization ScoreQA ScoreMath ScoreRAG ScoreAvg. Speedup (τ)Avg. TPSUniVerCategory=OT-basedCategory=OT-based2026.052.732.592.312.842.662.772.6562.5TraversalCategory=RRSw-basedCategory=RRSw-based2026.052.552.492.222.652.582.692.5359.6GreedyCategory=OT-basedCategory=OT-based2026.052.482.522.272.682.532.712.5359.1RRSwCategory=RRSw-basedCategory=RRSw-based2026.052.472.412.152.62.452.522.4357.9