Mathematical Reasoning on GSM8K (100 examples)
5.47Speed-upEAGLE-3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EAGLE-3Model=Vicuna-13B, Attack Status=Clean2026.05 | 5.47 | 5.95 | — | — | |
| EAGLE-2Model=Vicuna-7B, Attack Status=Clean2026.05 | 4.99 | 4.68 | — | — | |
| EAGLE-2Model=Vicuna-13B, Attack Status=Clean2026.05 | 4.84 | 4.91 | — | — | |
| HydraModel=Vicuna-7B, Attack Status=Clean2026.05 | 4.37 | 3.84 | — | — | |
| HydraModel=Vicuna-13B, Attack Status=Clean2026.05 | 4.29 | 3.58 | — | — | |
| EAGLEModel=Vicuna-7B, Attack Status=Clean2026.05 | 4.11 | 3.62 | — | — | |
| EAGLEModel=Vicuna-13B, Attack Status=Clean2026.05 | 3.66 | 3.79 | — | — | |
| MedusaModel=Vicuna-7B, Attack Status=Clean2026.05 | 3.46 | 2.47 | — | — | |
| MedusaModel=Vicuna-13B, Attack Status=Clean2026.05 | 3.39 | 2.51 | — | — | |
| EAGLE-2Model=Vicuna-7B, Attack Status=Attacked (MISTLETOE)2026.05 | 2.71 | 2.84 | — | — | |
| EAGLE-2Model=Vicuna-13B, Attack Status=Attacked (MISTLETOE)2026.05 | 2.62 | 3.25 | — | — | |
| MedusaModel=Vicuna-7B, Attack Status=Attacked (MISTLETOE)2026.05 | 2.5 | 1.82 | — | — | |
| HydraModel=Vicuna-7B, Attack Status=Attacked (MISTLETOE)2026.05 | 2.24 | 3.07 | — | — | |
| EAGLEModel=Vicuna-7B, Attack Status=Attacked (MISTLETOE)2026.05 | 2.02 | 3.06 | — | — | |
| EAGLE-3Model=Vicuna-13B, Attack Status=Attacked (MISTLETOE)2026.05 | 1.83 | 2.79 | — | — | |
| HydraModel=Vicuna-13B, Attack Status=Attacked (MISTLETOE)2026.05 | 1.72 | 3.07 | — | — | |
| EAGLEModel=Vicuna-13B, Attack Status=Attacked (MISTLETOE)2026.05 | 1.69 | 2.96 | — | — | |
| MedusaModel=Vicuna-13B, Attack Status=Attacked (MISTLETOE)2026.05 | 1.46 | 2.32 | — | — |