Commits

Branch selector

 master

User selector

guptapriya

Datepicker

All time

Commit History

Commits on Sep 27, 2019

• Mark tf.keras.utils.multi_gpu_model as deprecated.

  Show description for bdb461a
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  bdb461a

  Copy full SHA for bdb461a

Commits on Aug 12, 2019

• Automated rollback of commit cadb1283346e08ab780e9cc7338916ebf6ebb482. Revert #22231.

  Show description for 13e631e
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  13e631e

  Copy full SHA for 13e631e

Commits on Aug 1, 2019

• tf.saved_model: Re-create concrete functions at saving time - this ensures that if the cache key has changed, the function will be traced again. For example, when the model is run with distribution…

  Show description for 91d7124
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  91d7124

  Copy full SHA for 91d7124

Commits on Jul 30, 2019

• Re-enable tests with old optimizers + dist strat + keras.

  Show description for 5dd7dbb
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  5dd7dbb

  Copy full SHA for 5dd7dbb

Commits on Jul 29, 2019

• Handle distributed variables correctly in tf.saved_model saving. In particular, this allows saving models containing hub.KerasLayer and trained with distribution strategies.

  Show description for 6801a4b
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  6801a4b

  Copy full SHA for 6801a4b

Commits on Jul 26, 2019

• Disable some Keras + TPUStrategy tests because they are flaky.

  Show description for d0181e1
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  d0181e1

  Copy full SHA for d0181e1

Commits on Jul 23, 2019

• Add missing experimental_between_graph implementation in OneDeviceStrategy.

  Show description for ea461a4
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  ea461a4

  Copy full SHA for ea461a4

Commits on Jul 19, 2019

• Test LSTM v2 layer in eager mode in dist strat keras tests

  Show description for 2ecc2ff
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  2ecc2ff

  Copy full SHA for 2ecc2ff

Commits on Jul 16, 2019

• Update docs of MultiWorkerMirroredStrategy (and some related docs in base class)

  Show description for 31cc24f
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  31cc24f

  Copy full SHA for 31cc24f

Commits on Jul 4, 2019

• Update docs for tf.distribute.OneDeviceStrategy.

  Show description for f234584
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  f234584

  Copy full SHA for f234584

Commits on Jun 14, 2019

• Use simplified implementation of executing_eagerly_outside_functions

  Show description for e48c981
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  e48c981

  Copy full SHA for e48c981

• Simplified implementation of executing_eagerly_outside_functions. Use this in ops.device to detect this case. The previous change to ops.device affected legacy graph mode as well when inside defuns…

  Show description for 10c5d10
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  10c5d10

  Copy full SHA for 10c5d10

Commits on Jun 11, 2019

• Set device in eager context when inside function building graph as well, in addition to setting it on the graph.

  Show description for 0dfa20a
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  0dfa20a

  Copy full SHA for 0dfa20a

Commits on May 23, 2019

• Make `trainable` argument of variables default to `None` everywhere consistently, and update the documentation to be consistent as well.

  Show description for f047cef
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  f047cef

  Copy full SHA for f047cef

Commits on May 15, 2019

• More descriptive error message for missing aggregation type in mirrored variables.

  Show description for c10a7b0
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  c10a7b0

  Copy full SHA for c10a7b0

Commits on Apr 25, 2019

• Update error message when tf.functions cross merge_call boundary when using MirroredStrategy

  Show description for f54bc43
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  f54bc43

  Copy full SHA for f54bc43

Commits on Apr 18, 2019

• Remove irrelevant callbacks warning with distribute strategies.

  Show description for 5aa52f3
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  5aa52f3

  Copy full SHA for 5aa52f3

Commits on Apr 16, 2019

• Add migration instructions for tf.contrib.distribute module.

  Show description for e2bd334
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  e2bd334

  Copy full SHA for e2bd334

Commits on Apr 14, 2019

• Add make_numpy_dataset to tf.distribute.Strategy (V1 and V2).

  Show description for 256add6
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  256add6

  Copy full SHA for 256add6

• tf.distribute.Strategy.reduce: Remove default value for axis argument for Strategy V2, but keep for Strategy V1.

  Show description for 471993a
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  471993a

  Copy full SHA for 471993a

Commits on Feb 28, 2019

• MirroredStrategy: Throw an error if there is a merge call in the middle of a `fn` passed to call_for_each_replica which changes the graph being used while calling `fn`. This is because we switch fr…

  Show description for 3295fdb
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  3295fdb

  Copy full SHA for 3295fdb

Commits on Feb 17, 2019

• Parameter Server Distribution strategy: Do not try to remove from trainable collection if not present.

  Show description for ca228e7
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  ca228e7

  Copy full SHA for ca228e7

Commits on Feb 16, 2019

• Disable flaky dist strat input tests with callable input.

  Show description for 4a1d3ba
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  4a1d3ba

  Copy full SHA for 4a1d3ba

Commits on Feb 13, 2019

• Distribution Strategies: Unit test for regularizer loss scaling.

  Show description for cf48294
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  cf48294

  Copy full SHA for cf48294

Commits on Feb 12, 2019

• Distribution Strategy in Keras: Scale regularization losses in keras compile. Regular losses are already scaled. We need to scale regularization and other custom losses as well.

  Show description for 0839049
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  0839049

  Copy full SHA for 0839049

Commits on Feb 8, 2019

• Recurrent layer: Force runtime type string on CPUs, as otherwise it breaks when the containing code is explicitly placed on a GPU, for example when using distribution strategy.

  Show description for 0f5fd00
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  0f5fd00

  Copy full SHA for 0f5fd00

Commits on Feb 7, 2019

• Update one device strategy tests to use combinations. This correctly skips the GPU tests when none is available.

  Show description for 3457887
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  3457887

  Copy full SHA for 3457887

Commits on Feb 2, 2019

• MNIST eager dist strat example: Add an option to run the step in a tf.function. Also changed the example to run with TF 2.0 from compat.

  Show description for 9018715
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  9018715

  Copy full SHA for 9018715

Commits on Jan 27, 2019

• Build different models for training, test and predict in keras when used with distribution strategy.

  Show description for e4b4b2e
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  e4b4b2e

  Copy full SHA for e4b4b2e

Commits on Jan 26, 2019

• Remove no_pip tag from distribute strategy tests.

  Show description for e9e5028
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  e9e5028

  Copy full SHA for e9e5028

• Increase shards for one of the keras DS test so it doesn't time out.

  Show description for a29d085
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  a29d085

  Copy full SHA for a29d085

• Keras learning phase scope bug fix.

  Show description for 8234d5a
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  8234d5a

  Copy full SHA for 8234d5a

Commits on Jan 24, 2019

• MNIST example with eager and distribution strategy.

  Show description for c184af0
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  c184af0

  Copy full SHA for c184af0

Commits on Dec 19, 2018

• Keras + Distribution Strategy: Use a slightly lower tolerance for correctness test for default distribution strategy as the weights have more variance whenever the training is run on GPUs.

  Show description for 627fd02
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  627fd02

  Copy full SHA for 627fd02

• Disable keras correctness test for default distribution strategy.

  Show description for 0b2eafa
  

  guptapriya

  authored and

  tensorflower-gardener

  committed
  0b2eafa

  Copy full SHA for 0b2eafa

Pagination

Previous

Next