benchmarks

Model Training Summary

Three different models were trained on MNIST and CIFAR-10 datasets using PyTorch on a device with mps. The results of the training are summarized below:

LeNet on MNIST

  • Epoch: 1
  • Batch size: 4,21
  • Loss: 0.0733
  • Train accuracy: 97.33%
  • Validation accuracy: 97.77%
  • Test accuracy: 97.39%
  • Time / epoch without evaluation: 6.19 sec
  • Total Training Time: 9.38 sec

MLP on MNIST

  • Epoch: 1
  • Batch size: 4,21
  • Loss: 0.3499
  • Train accuracy: 91.63%
  • Validation accuracy: 93.48%
  • Test accuracy: 92.15%
  • Time / epoch without evaluation: 3.25 sec
  • Total Training Time: 6.24 sec

VGG16 on CIFAR-10

  • Epoch: 1
  • Batch size: 1,406
  • Loss: 2.0943
  • Train accuracy: 31.80%
  • Validation accuracy: 32.26%
  • Test accuracy: 32.40%
  • Time / epoch without evaluation: 1,405.05 sec
  • Total Training Time: 2,340.54 sec