Fetching the paper…

Maximizing Parallelism in Distributed Training for Huge Neural Networks · Around