Image Classification on MNIST (Quantization/Bit-width Performance)
98.66Accuracy (w32a32)GPTQ
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GPTQArchitecture=KAGN Simple2025.11 | 98.66 | 98.65 | 97.96 | 90.11 | 27.39 | 97.96 | 90.11 | 27.39 | |
| AdaRoundArchitecture=KAGN Simple2025.11 | 98.66 | 98.76 | 96.59 | 88.41 | 29.83 | 96.59 | 88.41 | 29.83 | |
| AWQArchitecture=KAGN Simple2025.11 | 98.66 | 98.7 | 97.35 | 57.65 | 13.36 | 97.35 | 57.65 | 13.36 | |
| BRECQArchitecture=KAGN Simple2025.11 | 98.66 | 98.73 | 96.11 | 76.07 | 23.33 | 96.22 | 81.7 | 12.8 | |
| GPTQArchitecture=KAN FCN2025.11 | 97.99 | 98 | 97.86 | 97.07 | 76.58 | 95.88 | 88.99 | 40.9 | |
| AdaRoundArchitecture=KAN FCN2025.11 | 97.99 | 97.97 | 97.78 | 96.36 | 42.87 | 95.75 | 88.88 | 32.35 | |
| AWQArchitecture=KAN FCN2025.11 | 97.99 | 97.98 | 97.8 | 95.53 | 33.65 | 95.55 | 88.73 | 21.02 | |
| BRECQArchitecture=KAN FCN2025.11 | 97.99 | 73.77 | 72.58 | 73.6 | 50.51 | 70.87 | 86.93 | 52.38 | |
| GPTQArchitecture=KAN ConvNet2025.11 | 95.69 | 95.69 | 95.65 | 95.63 | 88.18 | 10.1 | 10.1 | 10.1 | |
| AdaRoundArchitecture=KAN ConvNet2025.11 | 95.69 | 95.69 | 95.61 | 95.2 | 80.03 | 9.73 | 10.28 | 10.25 | |
| AWQArchitecture=KAN ConvNet2025.11 | 95.69 | 95.68 | 95.62 | 95.01 | 62.7 | 10.1 | 10.1 | 10.32 | |
| BRECQArchitecture=KAN ConvNet2025.11 | 95.69 | 36.37 | 37.79 | 30.31 | 35.07 | 75.6 | 93.52 | 64.97 |