experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
88,838 Commits 49 Branches 133 Tags 692 MiB
snyk-fix-b912e6cb5f397e5762e290d21a5012b6
Commit Graph

3 Commits

Author SHA1 Message Date
Anjali Sridhar
0f3562ba77 Another round of refactoring of values.py to split utility functions that use distributed Variable types defined in values.py. 
PiperOrigin-RevId: 316147517
Change-Id: I72e17b02e8f41c9cee40f4ec7f56fec2f7d860a9
 2020-06-12 12:04:39 -07:00
Rick Chao
e580b59f5a Fix typo in mirrored_run.py. 
PiperOrigin-RevId: 307974425
Change-Id: I40cee1e4f9b6ba14984db507689f17e6bdeee71e
 2020-04-22 22:00:01 -07:00
Ran Chen
b16d24a342 Separate mirrored call_for_each_replica to its own file 
Both ParameterServer and Mirrored uses this, and itself is complicated enough.

This also fix a issue that you can't strategy.run(tf.function) under CentralStorageStrategy, by applying the same workaround we have in MirroredStrategy.

PiperOrigin-RevId: 304437119
Change-Id: I6a7a67b88e7a5b7217aa9ffe05882d0ef4097896
 2020-04-02 11:11:09 -07:00