- 27 Mar, 2020 2 commits
-
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 303356961
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 303225001
-
- 26 Mar, 2020 1 commit
-
-
Hongkun Yu authored
PiperOrigin-RevId: 303050809
-
- 25 Mar, 2020 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 302977474
-
Will Cromar authored
PiperOrigin-RevId: 302921283
-
- 24 Mar, 2020 2 commits
-
-
Ran Chen authored
For some strategies we don't do all reduce, so all_reduce_sum_gradients can be misleading. The parameter is also changed to experimental because of issues with CentralStorageStrategy. PiperOrigin-RevId: 302734837
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 302552996
-
- 19 Mar, 2020 3 commits
-
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 301915584
-
Sergey Mironov authored
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 301741984
-
- 17 Mar, 2020 4 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 301441181
-
ayushmankumar7 authored
-
Hongkun Yu authored
PiperOrigin-RevId: 301308623
-
Ken Franko authored
PiperOrigin-RevId: 301277545
-
- 12 Mar, 2020 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 300481917
-
Hongkun Yu authored
PiperOrigin-RevId: 300477605
-
- 11 Mar, 2020 3 commits
-
-
Will Cromar authored
PiperOrigin-RevId: 300433601
-
Sergey Mironov authored
-
Sergey Mironov authored
-
- 10 Mar, 2020 2 commits
-
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 300203487
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 300009545
-
- 09 Mar, 2020 1 commit
-
-
Hongkun Yu authored
PiperOrigin-RevId: 299901483
-
- 07 Mar, 2020 1 commit
-
-
Hongkun Yu authored
PiperOrigin-RevId: 299594839
-
- 06 Mar, 2020 1 commit
-
-
Zongwei Zhou authored
In BERT SQuAD, disable explicit allreduce for now to keep the original clip_by_global_norm math. With explicit allreduce, the gradients before allreduce are scaled so even if we move clip_by_global_norm before allreduce (as in TF1 and pre-TF 2.2) it will operate on scaled gradients, the math will be changed. So with explicit allreduce, it is better to move clip_by_global_norm to after allreduce. PiperOrigin-RevId: 299278082
-
- 05 Mar, 2020 6 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 299169021
-
Hongkun Yu authored
PiperOrigin-RevId: 299160422
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 299149155
-
Jing Li authored
PiperOrigin-RevId: 299126470
-
Zongwei Zhou authored
PiperOrigin-RevId: 299007295
-
Chen Chen authored
PiperOrigin-RevId: 298986381
-
- 04 Mar, 2020 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 298905114
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 298817372
-
- 03 Mar, 2020 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 298692558
-
George Karpenkov authored
Removed with plans to re-add later once the feature stabilizes more. PiperOrigin-RevId: 298486867
-
- 02 Mar, 2020 1 commit
-
-
Will Cromar authored
PiperOrigin-RevId: 298466825
-
- 01 Mar, 2020 2 commits
- 26 Feb, 2020 3 commits
-
-
George Karpenkov authored
PiperOrigin-RevId: 297412889
-
Hongkun Yu authored
PiperOrigin-RevId: 297383836
-
George Karpenkov authored
PiperOrigin-RevId: 297366405
-