DeepSpeedZeroOptimizer: refactor bit16 flattening to support more accelerators#4833

Merged

tjruwase merged 11 commits intomicrosoft:masterfrom nelyahu:zeroOptParamsFlatenning

Jan 11, 2024

+33-17

Commits on Dec 12, 2023

ZeroOptimizer: avoid storage sharing when flatenning params
nelyahu
committed

Commits on Dec 18, 2023

Commits on Dec 19, 2023

Use meta tensor instead of reconstructing CPU tensors
nelyahu
committed

Commits on Jan 2, 2024

Merge branch 'master' into zeroOptParamsFlatenning
tjruwase
authored

Commits on Jan 3, 2024

Commits on Jan 5, 2024

Merge branch 'master' into zeroOptParamsFlatenning
mrwyattii
authored

Commits on Jan 10, 2024