Text Generation on ShareGPT
2.01Speedup vs ARHiSpec
Evaluation Results
| Method | Links | |
|---|---|---|
| HiSpecModel=Llama3-8B2025.10 | 2.01 | |
| HiSpecModel=Llama2-7B2025.10 | 1.7 | |
| HiSpecModel=Llama2-70B2025.10 | 1.64 | |
| HiSpecModel=Llama2-13B2025.10 | 1.62 | |
| HiSpecModel=CodeLlama-34B2025.10 | 1.55 | |
| LookAheadModel=Llama2-7B2025.10 | 1.37 | |
| LayerSkipModel=Llama2-7B2025.10 | 1.31 | |
| LookAheadModel=CodeLlama-34B2025.10 | 1.31 | |
| LayerSkipModel=Llama2-70B2025.10 | 1.3 | |
| LookAheadModel=Llama2-13B2025.10 | 1.26 | |
| LayerSkipModel=CodeLlama-34B2025.10 | 1.23 | |
| LayerSkipModel=Llama2-13B2025.10 | 1.21 | |
| LookAheadModel=Llama3-8B2025.10 | 1.2 | |
| AdaDecodeModel=CodeLlama-34B2025.10 | 1.19 | |
| LayerSkipModel=Llama3-8B2025.10 | 1.14 | |
| SWIFTModel=Llama2-70B2025.10 | 1.1 | |
| SWIFTModel=CodeLlama-34B2025.10 | 1.09 | |
| SWIFTModel=Llama2-13B2025.10 | 1.02 | |
| AdaDecodeModel=Llama3-8B2025.10 | 1 |