Fetching the paper…

GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism · Around