Large Language Model Evaluation on Qwen-32B
80.81MMLUFP16
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FP16Format=-, Quantization=FP162025.09 | 80.81 | 92.04 | 83.97 | 76.56 | 83.35 | — | |
| RTNFormat=NVFP, Quantization=RTN2025.09 | 79.85 | 94.24 | 83.27 | 75.22 | 83.15 | 99.76 | |
| GPTQFormat=NVFP, Quantization=GPTQ2025.09 | 79.54 | 92.87 | 83.24 | 75.93 | 82.9 | 99.46 | |
| GPTQ+Had128Format=NVFP, Quantization=GPTQ+Had1282025.09 | 79.11 | 90.52 | 83.15 | 76.09 | 82.22 | 98.65 | |
| RTN+Had16Format=NVFP, Quantization=RTN+Had162025.09 | 78.9 | 89.23 | 82.6 | 76.48 | 81.8 | 98.15 | |
| GPTQ+Had128Format=MXFP, Quantization=GPTQ+Had1282025.09 | 78.9 | 90.9 | 82.29 | 75.22 | 81.83 | 98.18 | |
| GPTQ+Had16Format=NVFP, Quantization=GPTQ+Had162025.09 | 78.6 | 90.9 | 82.93 | 75.14 | 81.89 | 98.26 | |
| RTN+Had128Format=NVFP, Quantization=RTN+Had1282025.09 | 78.49 | 89.69 | 82.47 | 75.37 | 81.51 | 97.79 | |
| GPTQ+Had32Format=MXFP, Quantization=GPTQ+Had322025.09 | 78.46 | 82.41 | 82.72 | 75.06 | 79.66 | 95.58 | |
| RTN+Had128Format=MXFP, Quantization=RTN+Had1282025.09 | 78.36 | 88.1 | 81.66 | 75.3 | 80.86 | 97.01 | |
| RTN+Had32Format=MXFP, Quantization=RTN+Had322025.09 | 78.22 | 93.03 | 81.76 | 75.93 | 82.24 | 98.67 | |
| RTNFormat=MXFP, Quantization=RTN2025.09 | 77.07 | 72.33 | 81.52 | 75.22 | 76.54 | 91.83 | |
| GPTQFormat=MXFP, Quantization=GPTQ2025.09 | 77.01 | 88.55 | 81.79 | 74.9 | 80.56 | 96.66 |