List your company

Multi-turn Dialogue on MT-Bench (Acceptance Length τ and Speedup Ratio)

5.78Acceptance Length (τ)PPOW
3.92884.40944.895.3706May 14, 2026

Evaluation results

MethodLinks
PPOWModel=Q3-32B, Temperature=0.02026.055.783.54
PPOWModel=Q3-8B, Temperature=0.02026.055.583.02
EAGLE-3Model=L31-8B, Temperature=0.02026.055.532.91
PPOWModel=L31-8B, Temperature=0.02026.055.472.72
PPOWModel=L33-70B, Temperature=0.02026.055.453.73
EAGLE-3Model=Q3-32B, Temperature=0.02026.055.253.47
GRIFFINModel=L31-8B, Temperature=0.02026.055.142.58
PPOWModel=L33-70B, Temperature=1.02026.055.143.52
EAGLE-3Model=L33-70B, Temperature=0.02026.055.123.63
GRIFFINModel=L33-70B, Temperature=0.02026.055.083.6
EAGLE-3Model=Q3-8B, Temperature=0.02026.054.952.64
PPOWModel=Q3-32B, Temperature=1.02026.054.952.81
EAGLE-3Model=L33-70B, Temperature=1.02026.054.823.4
PPOWModel=Q3-8B, Temperature=1.02026.054.722.82
GRIFFINModel=L33-70B, Temperature=1.02026.054.73.38
EAGLE-3Model=Q3-32B, Temperature=1.02026.054.512.6
EAGLE-3Model=L31-8B, Temperature=1.02026.054.172.35
PPOWModel=L31-8B, Temperature=1.02026.054.122.29
GRIFFINModel=L31-8B, Temperature=1.02026.054.032.24
EAGLE-3Model=Q3-8B, Temperature=1.02026.0542.41