- 27 Aug, 2021 1 commit
-
-
Hongkun Yu authored
PiperOrigin-RevId: 393366505
-
- 06 Jun, 2021 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 377801393
-
Hongkun Yu authored
PiperOrigin-RevId: 377801393
-
- 03 May, 2021 2 commits
-
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 371640518
-
A. Unique TensorFlower authored
PiperOrigin-RevId: 371640518
-
- 29 Jan, 2021 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 354439331
-
Hongkun Yu authored
PiperOrigin-RevId: 354439331
-
- 21 Dec, 2020 1 commit
-
-
Samuel Marks authored
-
- 30 Oct, 2020 2 commits
-
-
Pankaj Kanwar authored
PiperOrigin-RevId: 339927293
-
Pankaj Kanwar authored
PiperOrigin-RevId: 339927293
-
- 09 Oct, 2020 2 commits
- 29 Aug, 2020 2 commits
-
-
Zongwei Zhou authored
PiperOrigin-RevId: 329042049
-
Zongwei Zhou authored
PiperOrigin-RevId: 329042049
-
- 12 Aug, 2020 2 commits
-
-
Hongkun Yu authored
PiperOrigin-RevId: 326286926
-
Hongkun Yu authored
PiperOrigin-RevId: 326286926
-
- 24 Mar, 2020 1 commit
-
-
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
-
- 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 1 commit
-
-
Zongwei Zhou authored
PiperOrigin-RevId: 299007295
-