Quantization Performance Summary on Aggregated Benchmarks HellaSwag, MMLU, Arc-C, MATH-500
1.014Average ScoreSmoothQuant
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SmoothQuantBit=W8A8, Format=HiF2026.02 | 1.014 | -1.4 | |
| SmoothQuantBit=W8A8, Format=MXFP2026.02 | 1.001 | -0.1 | |
| BF16Format=BF162026.02 | 1 | — | |
| SVDQuantBit=W8A8, Format=INT2026.02 | 0.998 | 0.2 | |
| RTNBit=W8A8, Format=INT2026.02 | 0.995 | 0.5 | |
| SmoothQuantBit=W8A8, Format=INT2026.02 | 0.995 | 0.5 | |
| SVDQuantBit=W8A8, Format=MXFP2026.02 | 0.993 | 0.7 | |
| SVDQuantBit=W8A8, Format=HiF2026.02 | 0.989 | 1.1 | |
| RTNBit=W8A8, Format=MXFP2026.02 | 0.986 | 1.4 | |
| RTNBit=W8A8, Format=HiF2026.02 | 0.978 | 2.2 | |
| SmoothQuantBit=W4A4, Format=HiF2026.02 | 0.973 | 2.7 | |
| SVDQuantBit=W4A4, Format=HiF2026.02 | 0.973 | 2.7 | |
| RTNBit=W4A4, Format=HiF2026.02 | 0.97 | 3 | |
| SVDQuantBit=W4A4, Format=NVFP2026.02 | 0.962 | 3.8 | |
| SmoothQuantBit=W4A4, Format=NVFP2026.02 | 0.952 | 4.8 | |
| RTNBit=W4A4, Format=NVFP2026.02 | 0.943 | 5.7 | |
| SmoothQuantBit=W4A4, Format=MXFP2026.02 | 0.938 | 6.2 | |
| SVDQuantBit=W4A4, Format=MXFP2026.02 | 0.93 | 7 | |
| RTNBit=W4A4, Format=MXFP2026.02 | 0.914 | 8.6 | |
| SVDQuantBit=W4A4, Format=INT2026.02 | 0.51 | 49 | |
| SmoothQuantBit=W4A4, Format=INT2026.02 | 0.393 | 60.7 | |
| RTNBit=W4A4, Format=INT2026.02 | 0.326 | 67.4 |