Speculative Decoding on AIME 25
7.08SpeedupTAPS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.05 | 7.08 | — | 8.62 | |
| TAPSHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.05 | 6.64 | — | 8.32 | |
| DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.05 | 5.68 | — | 6.75 | |
| DFlashHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.05 | 5.62 | — | 6.74 | |
| TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.05 | 4.84 | — | 8.1 | |
| TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.05 | 4.82 | — | 8.19 | |
| DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.05 | 4.23 | — | 9.36 | |
| TAPSHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 3.98 | — | 5.21 | |
| DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.05 | 3.83 | — | 9.67 | |
| DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.05 | 3.73 | — | 4.92 | |
| DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.05 | 3.57 | — | 4.73 | |
| TAPSHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 3.5 | — | 5.16 | |
| LogitSpec2025.07 | 3.33 | 3.76 | — | |
| DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.05 | 2.96 | — | 9.22 | |
| DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.05 | 2.91 | — | 9.55 | |
| DFlashHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 2.65 | — | 3.68 | |
| DFlashHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 2.56 | — | 3.65 | |
| DDTreeHardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 2.25 | — | 5.97 | |
| EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 2.02 | — | 2.81 | |
| EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-4B2026.05 | 1.79 | — | 3.05 | |
| EAGLE-3Hardware=NVIDIA A40-48GB GPU, Target Model=Qwen3-8B2026.05 | 1.79 | — | 3 | |
| EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-4B2026.05 | 1.64 | — | 2.79 | |
| EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 1.64 | — | 2.34 | |
| EAGLE-3Hardware=NVIDIA A800-SXM4-80GB GPU, Target Model=Qwen3-8B2026.05 | 1.63 | — | 2.74 | |
| DDTreeHardware=NVIDIA A40-48GB GPU, Target Model=LLaMA-3.1-8B-Instruct2026.05 | 1.47 | — | 5.91 | |
| Vanilla2025.07 | 1 | 1 | — |