List your company

Speculative Decoding Efficiency Metrics on Average Performance

6.73SpeedupTAPS
1.24922.67214.0955.5179May 30, 2026Jun 3, 2026Jun 7, 2026Jun 12, 2026Jun 16, 2026Jun 20, 2026Jun 25, 2026

Evaluation results

MethodLinks
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.056.737.92
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.056.147.91
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.055.917.89
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.055.727.83
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.055.138.89
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.054.916
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.054.866.03
TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.054.285.75
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.054.245.69
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.054.035.48
TAPSHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.053.875.74
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.053.58.9
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.053.488.88
DFlashHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.052.994.19
DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.052.974.2
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.052.658.95
HyperDFlashTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.062.533.36
DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.052.396.48
HyperDFlashTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.062.252.97
MTPTemperature=0, Drafted steps per verification round=3, Decoding mode=Think-high mode2026.062.112.76
MTPTemperature=1, Drafted steps per verification round=3, Decoding mode=Think-high mode2026.061.962.56
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.051.872.62
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.051.813.05
EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.051.762.96
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.051.722.95
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.051.682.83
DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.051.616.47
MTPTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.061.62.89
Vanilla DFlashTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.061.572.01
Vanilla DFlashTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.061.491.91
EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.051.482.09
MTPTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.061.462.65