Structured Task on BFCL
3.13SpeedupHybrid Verified Decoding
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=SuffixDecoding2026.05 | 3.13 | |
| Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=SuffixDecoding2026.05 | 2.47 | |
| Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=Greedy Decoding2026.05 | 2.31 | |
| Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=Greedy Decoding2026.05 | 2.19 | |
| Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=SuffixDecoding2026.05 | 1.32 | |
| Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=Frequency-Based Hybrid Decoding2026.05 | 1.27 | |
| Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=Frequency-Based Hybrid Decoding2026.05 | 1.13 | |
| Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=Greedy Decoding2026.05 | 1.02 | |
| Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=Frequency-Based Hybrid Decoding2026.05 | 1.01 | |
| Hybrid Verified DecodingTarget Model=Qwen3-8B, Reference Baseline=EAGLE32026.05 | 0.98 | |
| Hybrid Verified DecodingTarget Model=Qwen3-4B, Reference Baseline=EAGLE32026.05 | 0.98 | |
| Hybrid Verified DecodingTarget Model=Llama3.1-8B, Reference Baseline=EAGLE32026.05 | 0.9 |