Long-Context Generation on Qwen3 Context length 40K
5.37Throughput Speedup (α)SpecPV
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SpecPVModel Size=14B, Partial KV Cache Budget=4K2025.12 | 5.37 | 3.71 | |
| SpecPVModel Size=14B, Partial KV Cache Budget=2K2025.12 | 5.37 | 3.66 | |
| SpecPVModel Size=14B, Partial KV Cache Budget=8K2025.12 | 5.06 | 3.79 | |
| SpecPVModel Size=4B, Partial KV Cache Budget=2K2025.12 | 4.84 | 3.2 | |
| SpecPVModel Size=8B, Partial KV Cache Budget=4K2025.12 | 4.83 | 3.37 | |
| SpecPVModel Size=4B, Partial KV Cache Budget=4K2025.12 | 4.73 | 3.17 | |
| SpecPVModel Size=8B, Partial KV Cache Budget=2K2025.12 | 4.61 | 3.13 | |
| SpecPVModel Size=4B, Partial KV Cache Budget=8K2025.12 | 4.5 | 3.14 | |
| SpecPVModel Size=8B, Partial KV Cache Budget=8K2025.12 | 4.46 | 3.25 | |
| EAGLE3-YARNModel Size=14B2025.12 | 3.12 | 3.81 | |
| EAGLE3-YARNModel Size=8B2025.12 | 2.78 | 3.35 | |
| EAGLE3-YARNModel Size=4B2025.12 | 2.54 | 3.05 |