Algorithm 2) Ring-reduce
— Akshay 🚀 (@akshay_pachaar) 17 août 2025
Phase #1) Share-reduce
First, gradients are divided into G segments on every GPU (G = total number of GPUs).
Check this 👇 pic.twitter.com/15v6GIivHo
Algorithm 2) Ring-reduce Phase #1) Share-reduce First, gradients are divided into G segments on every GPU (G = total number of GPUs). Check this
