← All tags

#distributed

1 page

GPipe: Easy Scaling with Micro-Batch Pipeline Parallelism

Training giant neural networks by pipelining micro-batches across devices