DeepSpeed
Fixes for training models with bf16 + freshly initialized optimizer via `load_module_only`
#4141
Merged
Merged
Commits
-
haileyschoelkopf committed 2 years ago
-
haileyschoelkopf committed 2 years ago
-
haileyschoelkopf committed 2 years ago
-
haileyschoelkopf committed 2 years ago
-
tjruwase committed 2 years ago
-
tjruwase committed 2 years ago
-
loadams committed 2 years ago
-
loadams committed 2 years ago
