[microsoft/Megatron-DeepSpeed sync] Commits including 2021-08-09

[stas00]

This PR catches on 2 commits:

git remote add other https://github.com/microsoft/Megatron-DeepSpeed
git fetch other
git cherry-pick 3123a2943173
# resolve conflict
git cherry-pick f0131ce0287f008d85101fd

New checkpoint is deepspeedai/Megatron-DeepSpeed@f0131ce028

[stas00]

updated tr1-13B as well:

git checkout tr1-13B
git cherry-pick 3c9d748
git push