https://hackershare.dev/en/bookmarks/757799
Faster distributed GPU training with Reduction Server on Vertex AI