DeepSpeed
Support fp32 grad clipping and fix max_grad_norm confusion
#232

Merged

Commits

updates to support fp32 grad clipping and disable max_grad_norm support

jeffra committed 5 years ago
revert unused function

jeffra committed 5 years ago
clipping default 0

jeffra committed 5 years ago
update tests

jeffra committed 5 years ago
update tests and fix bug

jeffra committed 5 years ago
Merge branch 'master' into jeffra/max_grad_update

jeffra committed 5 years ago
rename inside fp32 test

jeffra committed 5 years ago
Merge branch 'jeffra/max_grad_update' of github.com:microsoft/DeepSpeed into jeffra/max_grad_update

jeffra committed 5 years ago