Summarization on CNN/DM (Speedup vs AR)
2.08Speedup (vs AR)HiSpec
Evaluation Results
| Method | Links | |
|---|---|---|
| HiSpecModel=Llama3-8B2025.10 | 2.08 | |
| HiSpecModel=Llama2-7B2025.10 | 1.95 | |
| LayerSkipModel=Llama2-7B2025.10 | 1.93 | |
| HiSpecModel=Llama2-70B2025.10 | 1.93 | |
| LayerSkipModel=Llama3-8B2025.10 | 1.7 | |
| HiSpecModel=CodeLlama-34B2025.10 | 1.59 | |
| HiSpecModel=Llama2-13B2025.10 | 1.57 | |
| LookAheadModel=Llama2-13B2025.10 | 1.55 | |
| AdaDecodeModel=CodeLlama-34B2025.10 | 1.53 | |
| LookAheadModel=CodeLlama-34B2025.10 | 1.5 | |
| LayerSkipModel=Llama2-70B2025.10 | 1.47 | |
| LookAheadModel=Llama3-8B2025.10 | 1.44 | |
| LayerSkipModel=Llama2-13B2025.10 | 1.43 | |
| LayerSkipModel=CodeLlama-34B2025.10 | 1.42 | |
| LookAheadModel=Llama2-7B2025.10 | 1.34 | |
| SWIFTModel=Llama2-70B2025.10 | 1.17 | |
| SWIFTModel=CodeLlama-34B2025.10 | 1.13 | |
| AdaDecodeModel=Llama3-8B2025.10 | 1.12 | |
| SWIFTModel=Llama2-13B2025.10 | 1.03 |