Language Modeling on WikiText-2, PTB3, and WikiText-103
4.645Perplexity (WikiText-2)SASQ
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SASQModel=LLaMA2-13B2025.12 | 4.645 | 13.536 | 5.052 | 7.744 | |
| BaselineModel=LLaMA2-13B, Precision=FP162025.12 | 4.884 | 36.621 | 7.099 | 16.201 | |
| QuaRotModel=LLaMA2-13B2025.12 | 4.9 | 36.75 | 7.113 | 16.254 | |
| SMQ DynamicModel=LLaMA2-13B2025.12 | 4.927 | 37.738 | 7.204 | 16.623 | |
| AWQModel=LLaMA2-13B2025.12 | 4.97 | — | — | — | |
| SASQModel=LLaMA1-7B2025.12 | 5.182 | 9.073 | 5.191 | 6.482 | |
| SASQModel=LLaMA2-7B2025.12 | 5.214 | 12.581 | 5.492 | 7.762 | |
| BaselineModel=LLaMA2-7B, Precision=FP162025.12 | 5.473 | 27.78 | 6.783 | 13.345 | |
| QuaRotModel=LLaMA2-7B2025.12 | 5.5 | 24.998 | 6.768 | 12.421 | |
| SMQ DynamicModel=LLaMA2-7B2025.12 | 5.51 | 27.994 | 6.828 | 13.444 | |
| AWQModel=LLaMA2-7B2025.12 | 5.6 | — | — | — | |
| BaselineModel=LLaMA1-7B, Precision=FP162025.12 | 5.683 | 8.576 | 5.762 | 6.674 | |
| SpinQuantModel=LLaMA2-7B2025.12 | 5.7 | — | — | — | |
| SMQ DynamicModel=LLaMA1-7B2025.12 | 5.716 | 8.62 | 5.796 | 6.711 | |
| AWQModel=LLaMA1-7B2025.12 | 5.78 | — | — | — | |
| SASQModel=LLaMA3-8B2025.12 | 6.124 | 10.553 | 6.135 | 7.604 | |
| BaselineModel=LLaMA3-8B, Precision=FP162025.12 | 6.14 | 9.992 | 6.259 | 7.464 | |
| SMQ DynamicModel=LLaMA3-8B2025.12 | 6.243 | 10.128 | 6.365 | 7.579 | |
| SpinQuantModel=LLaMA3-8B2025.12 | 6.5 | — | — | — | |
| SASQModel=Qwen2.5-7B2025.12 | 6.526 | 12.957 | 6.439 | 8.641 | |
| BaselineModel=Qwen2.5-7B, Precision=FP162025.12 | 6.85 | 11.286 | 6.878 | 8.338 | |
| SMQ DynamicModel=Qwen2.5-7B2025.12 | 6.994 | 11.503 | 7.013 | 8.503 | |
| LLM-QATModel=LLaMA3-8B2025.12 | 7.7 | — | — | — | |
| SASQModel=Qwen2.5-1.5B2025.12 | 8.465 | 16.92 | 8.227 | 11.204 | |
| BaselineModel=Qwen2.5-1.5B, Precision=FP162025.12 | 9.27 | 15.429 | 9.149 | 11.283 | |
| SMQ DynamicModel=Qwen2.5-1.5B2025.12 | 9.425 | 15.693 | 9.283 | 11.467 | |
| LLM-QATModel=LLaMA1-7B2025.12 | 10.3 | — | — | — | |
| LLM-QATModel=LLaMA2-7B2025.12 | 11.4 | — | — | — | |
| LLM-QATModel=LLaMA2-13B2025.12 | 14.5 | — | — | — | |
| SMQ StaticModel=LLaMA2-7B2025.12 | 448.524 | 863.796 | 387.995 | 566.772 |