Speculative Decoding on Aggregate of Sixteen Datasets
133Throughput (tokens/s)Hybrid Verified
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Hybrid VerifiedTarget Model=Qwen3-8B2026.05 | 133 | 189.7 | 4.27 | 9.7 | 90.3 | |
| EAGLE3Target Model=Qwen3-8B2026.05 | 109.5 | 240.2 | 3.16 | 0 | 100 | |
| Frequency HybridTarget Model=Qwen3-8B2026.05 | 105 | 188.4 | 4.31 | 8 | 92 | |
| SuffixTarget Model=Qwen3-8B2026.05 | 42.1 | 397.8 | 1.51 | 97.6 | 0 | |
| GreedyTarget Model=Qwen3-8B2026.05 | 39.9 | 998.7 | — | — | — |