Inference Performance and Accuracy for Long-Context Summarization on GovReport
3.09SpeedupGraft
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GraftBackbone=LLaMA3.1-8B2026.05 | 3.09 | — | — | 5.95 | |
| EAGLE3-64KBackbone=LLaMA3.1-8B2026.05 | 2.84 | — | — | 5.81 | |
| GraftBackbone=Qwen3-14B2026.05 | 2.09 | — | — | 3.46 | |
| EAGLE3-64KBackbone=Qwen3-14B2026.05 | 1.83 | — | — | 3.36 | |
| GraftBackbone=Qwen3-8B2026.05 | 1.77 | — | — | 3.87 | |
| GraftBackbone=Qwen3-4B2026.05 | 1.68 | — | — | 3.23 | |
| TriforceBackbone=LLaMA3.1-8B2026.05 | 1.64 | — | — | 2.41 | |
| EAGLE3-64KBackbone=Qwen3-8B2026.05 | 1.59 | — | — | 3.79 | |
| TokenSwiftBackbone=LLaMA3.1-8B2026.05 | 1.58 | — | — | 1.74 | |
| KnapSpecBackbone=Llama3.1-70B, Layers (L)=802026.02 | 1.47 | 13 | 94.4 | — | |
| EAGLE3-64KBackbone=Qwen3-4B2026.05 | 1.44 | — | — | 3.12 | |
| SWIFTBackbone=Llama3.1-70B, Layers (L)=802026.02 | 1.33 | 8.62 | 91.6 | — | |
| KnapSpecBackbone=Llama3.1-8B, Layers (L)=322026.02 | 1.28 | 25 | 97 | — | |
| KnapSpecBackbone=Llama3.2-3B, Layers (L)=282026.02 | 1.24 | 36.47 | 95.2 | — | |
| CLaSpBackbone=Llama3.1-70B, Layers (L)=802026.02 | 1.22 | 6.82 | 93.4 | — | |
| CLaSpBackbone=Llama3.1-8B, Layers (L)=322026.02 | 1.11 | 14.15 | 91.7 | — | |
| KnapSpecBackbone=Llama3.2-1B, Layers (L)=162026.02 | 1.11 | 58.79 | 96.1 | — | |
| SWIFTBackbone=Llama3.2-3B, Layers (L)=282026.02 | 1.08 | 18.45 | 64 | — | |
| CLaSpBackbone=Llama3.2-3B, Layers (L)=282026.02 | 1.08 | 18.91 | 85.7 | — | |
| SWIFTBackbone=Llama3.1-8B, Layers (L)=322026.02 | 1.05 | 13.72 | 62.1 | — | |
| SWIFTBackbone=Llama3.2-1B, Layers (L)=162026.02 | 1.03 | 31.32 | 83.8 | — | |
| CLaSpBackbone=Llama3.2-1B, Layers (L)=162026.02 | 1.02 | 33.75 | 83.6 | — | |
| ARBackbone=Llama3.1-70B, Layers (L)=802026.02 | 1 | 6.75 | — | — | |
| ARBackbone=Llama3.1-8B, Layers (L)=322026.02 | 1 | 15.71 | — | — | |
| ARBackbone=Llama3.2-3B, Layers (L)=282026.02 | 1 | 22.58 | — | — | |
| ARBackbone=Llama3.2-1B, Layers (L)=162026.02 | 1 | 46.56 | — | — | |
| DELBackbone=Llama3.2-1B, Layers (L)=162026.02 | 0.89 | 49.31 | 55.6 | — | |
| DELBackbone=Llama3.1-70B, Layers (L)=802026.02 | 0.87 | 6.62 | 0.1 | — | |
| DELBackbone=Llama3.2-3B, Layers (L)=282026.02 | 0.84 | 23.72 | 36 | — | |
| DELBackbone=Llama3.1-8B, Layers (L)=322026.02 | 0.74 | 15.72 | 0.1 | — | |
| MagicDecBackbone=LLaMA3.1-8B2026.05 | 0.68 | — | — | 2.05 |