Avoid rewrapping modules with DDP and Float16Module on repeated trainer.fit calls#12141
Open
farhadrgh wants to merge 8 commits intoNVIDIA:mainfrom farhadrgh:farhadr/megatron-skip-rewrap
+21-8
Commits
Commits on Feb 7, 2025
Commits on Feb 11, 2025
- committed
- committed