A fast communication-overlapping library for tensor/expert parallelism on GPUs. - View it on GitHub