Speculative Decoding on MedQA
6.47Match Rate (MAT)HedgeSpec
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HedgeSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 6.47 | 77.26 | |
| HedgeSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 6.08 | 65.82 | |
| HedgeSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 6.01 | 72.65 | |
| HedgeSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 5.92 | 38.6 | |
| HedgeSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 5.84 | 63.67 | |
| HedgeSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 5.46 | 39.29 | |
| EagleBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 4.96 | 66.48 | |
| Vanilla EagleTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 4.65 | 58.82 | |
| UCBSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 4.46 | 61.14 | |
| EXP3SpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 4.25 | 58.13 | |
| UCBSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.25 | 48.96 | |
| UCBSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 4.19 | 30.99 | |
| UCBSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 4.12 | 41.97 | |
| UCBSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 4.11 | 53.79 | |
| EXP3SpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 4.11 | 52.23 | |
| EXP3SpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.07 | 45.02 | |
| UCBSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 4.06 | 29.09 | |
| EXP3SpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 3.96 | 28.76 | |
| EXP3SpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 3.95 | 40.96 | |
| EXP3SpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 3.92 | 27.6 | |
| EagleBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 3.84 | 44.13 | |
| Vanilla EagleTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 3.76 | 42.89 | |
| EagleBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 2.57 | 18.33 | |
| Vanilla EagleTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 2.55 | 18.68 | |
| LLaMA-3.1-8B-ITTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 1 | 18.81 | |
| Qwen-3-8BTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 1 | 14.7 | |
| Qwen-3-32BTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 1 | 8.38 | |
| LLaMA-3.1-8B-ITBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 1 | 18.57 | |
| Qwen-3-8BBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 1 | 14.59 | |
| Qwen-3-32BBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 1 | 8.2 |