Image Classification on CIFAR-10 (Accuracy, Norm. Energy, Runtime)
93.71AccuracyTRAM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TRAMModel=ResNet50, Quantization=w8a8, Lambda (λ)=12026.05 | 93.71 | 74.97 | 0.69 | |
| FP32Model=ResNet502026.05 | 93.65 | — | — | |
| TRAMModel=ResNet50, Quantization=w8a8, Lambda (λ)=102026.05 | 93.65 | 55.64 | 0.7 | |
| AccMulModel=ResNet502026.05 | 93.56 | 100 | — | |
| TRAMModel=ResNet50, Quantization=w8a8, Lambda (λ)=10002026.05 | 92.97 | 35.81 | 0.69 | |
| TRAMModel=ResNet50, Quantization=w8a8, Lambda (λ)=1002026.05 | 92.79 | 35.9 | 0.69 | |
| MARLIN-1Model=ResNet50, Quantization=w8a82026.05 | 92.14 | 80.67 | 111.1 | |
| MARLIN-2Model=ResNet50, Quantization=w8a82026.05 | 91.7 | 76.67 | 111.1 | |
| ALWANN-1Model=ResNet50, Quantization=w8a82026.05 | 89.08 | 78.47 | — | |
| ALWANN-2Model=ResNet50, Quantization=w8a82026.05 | 88.58 | 70.02 | — |