Speculative Decoding on Biology
7.18MATHedgeSpec
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HedgeSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 7.18 | 93.78 | |
| HedgeSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 6.67 | 82.22 | |
| EagleBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 5.95 | 71.2 | |
| HedgeSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 5.93 | 39.13 | |
| Vanilla EagleTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 5.74 | 74.41 | |
| HedgeSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 5.66 | 62.08 | |
| HedgeSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 5.56 | 60.8 | |
| HedgeSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 5.44 | 38.54 | |
| UCBSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 5.22 | 70.27 | |
| EXP3SpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 4.93 | 67.7 | |
| UCBSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 4.9 | 64.53 | |
| EXP3SpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 4.67 | 60.28 | |
| UCBSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 4.25 | 31.73 | |
| UCBSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.16 | 47.29 | |
| EXP3SpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 4.1 | 30.31 | |
| UCBSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 4.08 | 30.14 | |
| Vanilla EagleTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 4.07 | 47.13 | |
| UCBSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 4.07 | 45.8 | |
| EagleBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.06 | 46.22 | |
| EXP3SpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.03 | 46.5 | |
| EXP3SpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 3.96 | 28.04 | |
| EXP3SpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 3.88 | 43.07 | |
| EagleBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 2.62 | 19.3 | |
| Vanilla EagleTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 2.52 | 19.25 | |
| LLaMA-3.1-8B-ITTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 1 | 18.03 | |
| Qwen-3-8BTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 1 | 14.95 | |
| Qwen-3-32BTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 1 | 8.35 | |
| LLaMA-3.1-8B-ITBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 1 | 18.44 | |
| Qwen-3-8BBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 1 | 14.64 | |
| Qwen-3-32BBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 1 | 8.43 |