Self-Compressing Neural Networks is dynamic quantization-aware training that puts the size of the model in the loss Paper: https://arxiv.org/pdf/2301.13142 Code: https://github.com/geohot/ai-notebooks/blob/master/mnist_self_compression.ipynb