Linked Presentation: Fine-tuning giant neural networks on commodity hardware with automatic pipeline model parallelism
