Megatron-DeepSpeed

277185f8 - [microsoft/Megatron-DeepSpeed sync] Commits including 2021-08-09 (#58)

[microsoft/Megatron-DeepSpeed sync] Commits including 2021-08-09 (#58) * Use new zero.Init() API (#10) * query deepspeed global grad norm (#8) Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com> Co-authored-by: Shaden Smith <Shaden.Smith@microsoft.com>