Eugene KhvedchenyainTowards Data ScienceEfficient PyTorch — Supercharging Training PipelineWhy reporting only the Top-1 accuracy of your model is often not enough.Aug 19, 2020Aug 19, 2020
Eugene KhvedchenyainTowards Data ScienceEfficient PyTorch — Part 1What is an efficient training pipeline? Is it the one, that produces a model with the best accuracy? Or the one that runs the fastest? Or…Jun 11, 20201Jun 11, 20201
Eugene KhvedchenyaWhy you should start using In-Place Activated BatchNormIn-Place Activated BatchNorm (InPlace-ABN) is memory efficient replacement for BatchNorm + Activation step. BN + Relu + Conv2d is an…Jun 21, 20182Jun 21, 20182