Speculative Decoding on SQL
8.06MATHedgeSpec
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HedgeSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 8.06 | 103.31 | |
| HedgeSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 7.89 | 95.81 | |
| HedgeSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 7.52 | 81.94 | |
| HedgeSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 7.16 | 45.3 | |
| HedgeSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 7.12 | 76.75 | |
| HedgeSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 6.58 | 46.24 | |
| EagleBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 5.99 | 80.41 | |
| Vanilla EagleTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 5.93 | 77.99 | |
| UCBSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 5.71 | 76.58 | |
| UCBSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 5.39 | 66.85 | |
| EXP3SpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 5.37 | 73.29 | |
| UCBSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 5.28 | 61.28 | |
| UCBSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 5.19 | 56.55 | |
| EXP3SpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 5.07 | 64.06 | |
| UCBSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 5.07 | 37.36 | |
| EXP3SpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.97 | 56.74 | |
| EXP3SpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 4.95 | 54.09 | |
| EXP3SpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 4.88 | 35.29 | |
| UCBSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 4.78 | 34.9 | |
| EXP3SpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 4.54 | 33.54 | |
| EagleBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 4.2 | 44.6 | |
| Vanilla EagleTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 4.18 | 48.06 | |
| Vanilla EagleTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 3.06 | 23.04 | |
| EagleBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 3 | 21.34 | |
| LLaMA-3.1-8B-ITTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10 | 1 | 18.25 | |
| Qwen-3-8BTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10 | 1 | 15.14 | |
| Qwen-3-32BTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.10 | 1 | 8.82 | |
| LLaMA-3.1-8B-ITBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10 | 1 | 17.86 | |
| Qwen-3-8BBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10 | 1 | 14.65 | |
| Qwen-3-32BBackbone=Qwen-3-32B, Inference Setting=Greedy2025.10 | 1 | 8.55 |