Commits · 2f0bf5946c45a07e23842737a8742d672d022942 · OpenDAS / apex

07 Mar, 2019 1 commit
- make norm compatible with pytorch version <= 1.0.1 · 99495376
  Deyu Fu authored Mar 06, 2019
  
  99495376
06 Mar, 2019 1 commit
- change to use now great torch.norm · 94667417
  Deyu Fu authored Mar 06, 2019
  
  94667417
01 Mar, 2019 1 commit
- Adding distributed tests and support for FusedAdam · d8b5d1be
  Michael Carilli authored Feb 28, 2019
  
  d8b5d1be
24 Feb, 2019 1 commit
- Stashing work · d137b800
  Michael Carilli authored Feb 24, 2019
  
  d137b800
11 Feb, 2019 1 commit
- Stashing work · fad78c16
  Michael Carilli authored Feb 10, 2019
  
  fad78c16
10 Feb, 2019 1 commit
- Lazily trigger fused_norm for FusedAdam error on Windows · 69907b80
  Michael Carilli authored Feb 10, 2019
  
  69907b80
11 Jan, 2019 1 commit

apex.optimizers.FP16_Optimizer: add state_dict() and load_state_dict() · 475fca23

Li Dong authored Jan 11, 2019

Add state_dict() and load_state_dict() for apex.optimizers.FP16_Optimizer to store and recover optimizer states. These two methods are important to recover from a stored training checkpoint.

475fca23

11 Dec, 2018 2 commits
- better print · f06feced
  Deyu Fu authored Dec 10, 2018
  
  f06feced
- remove temporarily added lr warmup · c8ca4bf4
  Deyu Fu authored Dec 10, 2018
```
This way optimizer will be more general, and warm up should be handled by user
```
  c8ca4bf4
07 Dec, 2018 1 commit
- address comments · 4d9dcb57
  Deyu Fu authored Dec 07, 2018
  
  4d9dcb57
06 Dec, 2018 1 commit
- WIP: improve fused adam · be42aad5
  Deyu Fu authored Dec 05, 2018
  
  be42aad5