More examples are provided at mdpi2022.
To train a full precision network:
python3 examples/mnist.py
# Test set: Average loss: 0.0234, Accuracy: 9922/10000 (99%)To train an 4-bit quantized and 75% pruned network:
python3 examples/mnist.py --pq
# Test set: Average loss: 0.0270, Accuracy: 9911/10000 (99%)