Question Answering on MMLU (STEM, Hums, Social, Others, Avg.)
53.8STEM AccuracyBaseline
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BaselineModel=LLaMA3-8B, Precision=FP16, Zero-shot=true2026.05 | 53.8 | 54.9 | 73.3 | 70.4 | 63.1 | |
| OmniQuantModel=LLaMA3-8B, Precision=W4A16, Zero-shot=true2026.05 | 49.4 | 49.1 | 66.6 | 64.4 | 57.4 | |
| OSAQ+GPTQModel=LLaMA3-8B, Precision=W4A16, Zero-shot=true2026.05 | 48 | 52.7 | 66.7 | 65.3 | 58.2 | |
| GPTQModel=LLaMA3-8B, Precision=W4A16, Zero-shot=true2026.05 | 47.3 | 52.3 | 66 | 64.9 | 57.6 | |
| BaselineModel=LLaMA2-7B, Precision=FP16, Zero-shot=true2026.05 | 34.4 | 39.8 | 47.3 | 47.1 | 42.2 | |
| OSAQ+GPTQModel=LLaMA2-7B, Precision=W4A16, Zero-shot=true2026.05 | 33.6 | 37.5 | 43.5 | 43.9 | 39.6 | |
| GPTQModel=LLaMA2-7B, Precision=W4A16, Zero-shot=true2026.05 | 32.7 | 36.9 | 42.6 | 42.6 | 38.7 | |
| OSAQ+GPTQModel=LLaMA2-7B, Precision=W3A16, Zero-shot=true2026.05 | 29.3 | 29.4 | 33.8 | 31.1 | 30.9 | |
| OmniQuantModel=LLaMA2-7B, Precision=W3A16, Zero-shot=true2026.05 | 29.1 | 31.1 | 30.6 | 30.4 | 30.3 | |
| OmniQuantModel=LLaMA2-7B, Precision=W4A16, Zero-shot=true2026.05 | 28.8 | 32.2 | 34.7 | 35.8 | 32.9 | |
| GPTQModel=LLaMA2-7B, Precision=W3A16, Zero-shot=true2026.05 | 28.2 | 27 | 32.1 | 29.9 | 29.3 | |
| OSAQ+GPTQModel=LLaMA3-8B, Precision=W3A16, Zero-shot=true2026.05 | 27 | 30.3 | 35.3 | 31.1 | 30.9 | |
| OmniQuantModel=LLaMA3-8B, Precision=W3A16, Zero-shot=true2026.05 | 26.3 | 27.8 | 29.5 | 29.9 | 28.4 | |
| GPTQModel=LLaMA3-8B, Precision=W3A16, Zero-shot=true2026.05 | 26.2 | 29.2 | 34.4 | 30 | 30 |