提交 · 5528b56647d39908ab0c9d2e97ff939d18f9b30f · wux_labs / Tensorflow

23 2月, 2019 38 次提交
- A
  Ignore get_configs that aren't JSON-serializable when saving checkpoints · 5528b566
  由 Allen Lavoie 提交于 2月 22, 2019
```
This is extra "nice to have" metadata, and otherwise it looks like a checkpointing error. Not worth bothering people about.

PiperOrigin-RevId: 235247436
```
  5528b566
- J
  [XLA] Add tests for edge cases of sqrt(x) and pow(x, 0.5). · fbe459fd
  由 Justin Lebar 提交于 2月 22, 2019
```
These are subtly different:

 - sqrt(-inf) = nan, but
   pow(-inf, 0.5) = +inf

Add exhaustive tests for sqrt, pow(0.5), and move the exhaustive fp16 test for
log into the new file.  Doing this requires a nontrivial rewrite of the
exhaustive test; hopefully this is the last time.

While we're here, we enable a few more exhaustive tests.

Also remove the special-case handling in GPU elemental IR emitter for
pow(x, 0.5) now that we have a sqrt HLO.

PiperOrigin-RevId: 235245101
```
  fbe459fd
- R
  Splitting out the memory usage tests into their own test file as that was... · 61aced48
  由 Rohan Jain 提交于 2月 22, 2019
```
Splitting out the memory usage tests into their own test file as that was making the test suite for MultiDeviceIterator too big and timing out.

PiperOrigin-RevId: 235242183
```
  61aced48
- P
  Automate clear losses at the end of a forward pass for custom training loops · eb741ced
  由 Pavithra Vijay 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235241120
```
  eb741ced
- A
  Clarify behavior of fit/predict/evaluate when inputs are Datasets and step args are None. · 0b3d833e
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235238247
```
  0b3d833e
- R
  Remove early_stopping.py from estimator/contrib in favor of... · 56915ed7
  由 Rick Chao 提交于 2月 22, 2019
```
Remove early_stopping.py from estimator/contrib in favor of estimator/python/estimator/early_stopping.py. And the test.

PiperOrigin-RevId: 235236119
```
  56915ed7
- T
  Merge pull request #24245 from wangsiyu:fix_init_checkpoint_ds · b81bc639
  由 TensorFlower Gardener 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235235216
```
  b81bc639
- T
  Merge pull request #25727 from ppwwyyxx:patch-15 · 4aa557a5
  由 TensorFlower Gardener 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235233898
```
  4aa557a5
- S
  [TF:XLA] Bump open source llvm revision to r354665 · 198cb0e2
  由 Sanjoy Das 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235233320
```
  198cb0e2
- W
  Fix callbacks_test path separator on Windows. · a0c1fec6
  由 William Chargin 提交于 2月 22, 2019
```
In cl/234179031, we improved this test to use `os.path.join` rather than
a hard-coded `+ "/tb"` in a few places, but apparently missed some.

Tested:
After 1036s/tb/something_else/, tests still pass on my local gLinux
desktop. Before this change, tests would fail after that substitution,
albeit not for the same reason as the traceback in b/125762969: we get a
normal `assertItemsEqual` failure rather than an ENOENT. Running Kokoro
Windows presubmits per go/tf-testing passes.

RELNOTES: n/a
PiperOrigin-RevId: 235232824
```
  a0c1fec6
- A
  Automated rollback of commit 4bf9ea42 · dab94104
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235232101
```
  dab94104
- P
  Add module aliases for v2 initializers: tf.initializers = tf.keras.initializers · 8d0a53de
  由 Pavithra Vijay 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235229813
```
  8d0a53de
- F
  Patch Simple/Union/Slurm/TPUClusterResolver to use the new num_accelerators function prototype. · fa52a202
  由 Frank Chen 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235229164
```
  fa52a202
- A
  Internal change · 590d8b56
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235228999
```
  590d8b56
- D
  [tf.data] Use `get_legacy_output_classes()` in `ConcatenateDataset` and `WindowDataset`. · 254b798b
  由 Derek Murray 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235228690
```
  254b798b
- F
  Fix progbar formatting. · f513d6ab
  由 Francois Chollet 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235226794
```
  f513d6ab
- A
  Work around missing support for complex multiplication in Cuda on Windows. · 56756d99
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235226607
```
  56756d99
- A
  Block ~KernelAndDeviceOp() until the execution of deferred ops finishes. · 8349a613
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235226116
```
  8349a613
- P
  - Disable `run_eagerly` and distribution strategy if there are symbolic... · b5b52ec3
  由 Pavithra Vijay 提交于 2月 22, 2019
```
- Disable `run_eagerly` and distribution strategy if there are symbolic tensors added to the model using `add_metric` or `add_loss`.
- Added error message for when subclassed model/Sequential model without input_shape is used with DS.
- Add error message when `run_eagerly` is enabled with DS.

PiperOrigin-RevId: 235225004
```
  b5b52ec3
- J
  Update java release version (1.13.0rc2) · cc5cb4fb
  由 James Ring 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235221008
```
  cc5cb4fb
- G
  Remove dependency on variable ops kernels in xla · 83887352
  由 Gunhan Gulsoy 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235219008
```
  83887352
- T
  Merge pull request #25967 from vladmos:incompatible · e51a68c6
  由 TensorFlower Gardener 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235218660
```
  e51a68c6
- D
  Fix tf.dtypes.as_dtypes() error message when invalid argument is a multi-element tuple. · 78e5608e
  由 Derek Murray 提交于 2月 22, 2019
```
Previously, the tuple would be unpacked in string formatting, and if it had more than one element we would get an argument unpacking error, instead of the intended TypeError.

PiperOrigin-RevId: 235217943
```
  78e5608e
- D
  [tf.data] Remove internal uses of Dataset.output_{shapes,types,classes} in preparation for 2.0. · f230b566
  由 Derek Murray 提交于 2月 22, 2019
```
Internal users of these properties should now use dataset_ops.get_legacy_output_{shapes,types,classes}() instead. In a future CL we will remove the properties from the DatasetV2.

PiperOrigin-RevId: 235217553
```
  f230b566
- A
  Tiny simplification of TensorProto compression code. · a9dab429
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235215099
```
  a9dab429
- T
  Merge pull request #24979 from benbarsdell:safe-fp16-layer-norm · 74347cdb
  由 TensorFlower Gardener 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235214785
```
  74347cdb
- A
  Use Arena for SequencExample allocation in example_parsing_ops.cc · 25996457
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
Provides better memory locality which speeds up most ops, and makes destruction trivial.

PiperOrigin-RevId: 235213734
```
  25996457
- A
  Allow tf.Module definitions with tf.function(input_signature=)-decorated methods · 809c74f0
  由 Allen Lavoie 提交于 2月 22, 2019
```
An unfortunate combination of the name_scope method wrapping not forwarding argspecs and input_signatures inspecting functions for keyword arguments.

PiperOrigin-RevId: 235211265
```
  809c74f0
- A
  Compress all constant nodes in the graph after the constant folding pass in Grappler. · c6a2c72f
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
Notice that most of the constants created by the constant folding pass are already in compressed format. This just extends this practice to all constants in the graph that have at least 64 elements and where the achievable compression ratio is grater or equal to 2.

PiperOrigin-RevId: 235211073
```
  c6a2c72f
- B
  Restrict tpu_embedding hosts to just worker jobs. Pass ClusterDef to TpuEmbedding. · 848e56c9
  由 Bruce Fontaine 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235210894
```
  848e56c9
- A
  [xla] plumb ReplicaId into local Python client · 85538ad4
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235210508
```
  85538ad4
- A
  Internal build visibility change. · e17c55f2
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235207900
```
  e17c55f2
- Z
  Improve Ftrl doc. · ab6dee70
  由 Zhenyu Tan 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235207665
```
  ab6dee70
- R
  Renames to prepare for the TF2.0 API. HashTable -> StaticHashTable and... · 32187f06
  由 Rohan Jain 提交于 2月 22, 2019
```
Renames to prepare for the TF2.0 API. HashTable -> StaticHashTable and MutableDenseHashTable -> DenseHashTable. Also insert -> insert_or_assign and remove -> erase for DenseHashTable.

PiperOrigin-RevId: 235206233
```
  32187f06
- G
  Reduce spammy logs. This help to make debugging easier when VLOG(1) is enabled. · df139a8a
  由 Guangda Lai 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235201217
```
  df139a8a
- D
  Avoid mixing other arguments with **kwargs in generic function proxies. · 0924cc63
  由 Dan Moldovan 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235197577
```
  0924cc63
- A
  Internal Change · 6f2eaa3b
  由 A. Unique TensorFlower 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235197341
```
  6f2eaa3b
- D
  Work around the limitation in Dataset which prevents for/Dataset loops with no state. · 565952cc
  由 Dan Moldovan 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235192783
```
  565952cc
22 2月, 2019 2 次提交

B
Be more consistent in handling bound methods in AutoGraph fallback pathway · a2bb5db1
由 Brian Lee 提交于 2月 22, 2019
```
PiperOrigin-RevId: 235184105
```
a2bb5db1

Improve documentation and error message for tf.split · 9d508106

由 James Keeling 提交于 2月 22, 2019

This function previously stated that it could accept a rank-0 Tensor as its `num_or_size_splits` argument. This is in fact not correct: TF graphs must have a static number of output tensors for any operation, so the number of splits has to be known statically or we cannot validate the correctness of the graph.

This method will now raise a ValueError if called with a rank-0 Tensor. I also make the documentation more explicit. Strangely, the test was testing for the incorrect behaviour, so I also fixed that.

PiperOrigin-RevId: 235157900

9d508106