ResearchBenchmarksMulti-turn dialogue evaluation on ruMT-Bench 1 (test)Multi-turn dialogue evaluation on ruMT-Bench 1 (test)Follow1.79SpeedupT-pro 2.01.6861.7131.741.767Dec 11, 2025Evaluation resultsMethodMethodLinksSpeedupAcceptance LengthT-pro 2.0speculative_decoding=E...speculative_decoding=EAGLE, mode=Standard, temperature=0.82025.121.793.31T-pro 2.0speculative_decoding=E...speculative_decoding=EAGLE, mode=Reasoning, temperature=0.82025.121.693.1