Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

General Speculative Decoding Performance on MT-Bench, HumanEval, and GSM8K Aggregate

6.52Average Acceptance Length (τ)

PPOW

1.7362.9784.225.462May 14, 2026May 16, 2026May 19, 2026May 21, 2026May 24, 2026May 26, 2026May 29, 2026
Updated 2d ago

Evaluation Results

MethodLinks
2026.05
6.523.39
2026.05
6.443.92
2026.05
6.43.46
2026.05
6.294.36
2026.05
6.163.17
2026.05
6.093.41
2026.05
64.21
2026.05
5.993.81
2026.05
5.974.07
2026.05
5.953.31
2026.05
5.944.15
2026.05
5.894.12
2026.05
5.694
2026.05
5.582.86
2026.05
5.523.9
2026.05
5.482.72
2026.05
5.362.93
2026.05
5.092.83
2026.05
5.092.67
2026.05
5.082.47
2026.05
5.062.81
2026.05
5.054.14
2026.05
4.93.33
2026.05
4.43.61
2026.05
4.232.88
2026.05
3.982.71
2026.05
3.863.17
2026.05
3.833.83
2026.05
3.642.99
2026.05
3.582.44
2026.05
3.512.39
2026.05
3.463.16
2026.05
3.443.44
2026.05
3.433.43
2026.05
3.42.79
2026.05
3.322.72
2026.05
3.283.28
2026.05
3.243.24
2026.05
3.213.21
2026.05
3.153.15
2026.05
3.133.13
2026.05
3.012.75
2026.05
2.942.69
2026.05
2.912.91
2026.05
2.882.63
2026.05
2.882.88
2026.05
2.882.63
2026.05
2.872.87
2026.05
2.861.95
2026.05
2.842.84
2026.05
2.832.83
2026.05
2.822.82
2026.05
2.822.82
2026.05
2.812.81
2026.05
2.752.26
2026.05
2.742.74
2026.05
2.722.72
2026.05
2.72.47
2026.05
2.651.8
2026.05
2.652.65
2026.05
2.642.64
2026.05
2.622.62
2026.05
2.562.56
2026.05
2.552.55
2026.05
2.532.53
2026.05
2.512.06
2026.05
2.492.49
2026.05
2.492.49
2026.05
2.482.48
2026.05
2.482.48
2026.05
2.462.46
2026.05
2.42.4
2026.05
2.382.38
2026.05
2.351.88
2026.05
2.332.33
2026.05
2.332.33
2026.05
2.332.13
2026.05
2.292.29
2026.05
2.282.28
2026.05
2.282.28
2026.05
2.242.24
2026.05
2.222.03
2026.05
2.21.47
2026.05
2.192.19
2026.05
2.172.17
2026.05
2.171.74
2026.05
2.131.89
2026.05
2.12.1
2026.05
2.11.4
2026.05
2.091.67
2026.05
2.082.08
2026.05
2.081.66
2026.05
2.082.08
2026.05
2.031.8
2026.05
2.021.35
2026.05
2.022.02
2026.05
1.981.76
2026.05
1.931.54
2026.05
1.931.29
2026.05
1.921.28
Showing 100 of 112 rows