Zero-shot Evaluation on Downstream tasks average
81Avg Zero-shot AccuracyFP16
Evaluation results
| Method | Links | |
|---|---|---|
| FP16Precision=FP16, Model Architecture=L3-70B2026.04 | 81 | |
| FP16Precision=FP16, Model Architecture=L3.1-8B-Inst2026.04 | 75.4 | |
| FP16Precision=FP16, Model Architecture=L2-13B2026.04 | 73.2 | |
| FP16Precision=FP16, Model Architecture=L2-7B2026.04 | 70.4 | |
| QuaRot+GPTAQ+OursPrecision=W2A16, Model Architecture=L3-70B2026.04 | 59 | |
| QuaRot+GPTAQPrecision=W2A16, Model Architecture=L3-70B2026.04 | 58.3 | |
| QuaRot+GPTAQ+OursPrecision=W2A16, Model Architecture=L2-13B2026.04 | 58.2 | |
| QuaRot+GPTAQ+OursPrecision=W2A16, Model Architecture=L3.1-8B-Inst2026.04 | 55.9 | |
| QuaRot+GPTAQPrecision=W2A16, Model Architecture=L2-13B2026.04 | 55.8 | |
| QuaRot+GPTAQPrecision=W2A16, Model Architecture=L3.1-8B-Inst2026.04 | 54.7 | |
| QuaRot+GPTAQ+OursPrecision=W2A16, Model Architecture=L2-7B2026.04 | 54 | |
| QuaRot+GPTAQPrecision=W2A16, Model Architecture=L2-7B2026.04 | 51.5 | |
| QuaRot+GPTQPrecision=W2A16, Model Architecture=L3.1-8B-Inst2026.04 | 50.7 | |
| QuaRot+GPTQPrecision=W2A16, Model Architecture=L2-13B2026.04 | 50.5 | |
| QuaRot+GPTQPrecision=W2A16, Model Architecture=L3-70B2026.04 | 45.2 | |
| QuaRot+GPTQPrecision=W2A16, Model Architecture=L2-7B2026.04 | 45 |