OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2025-12-06 12:20:11 +01:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity
107,660 Commits 50 Branches 144 Tags 1.3 GiB
fc7529576a
Commit Graph

107660 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Mark Daoust
fc7529576a Move TFRecord examples to the exported class. 
PiperOrigin-RevId: 365149622
Change-Id: Ib7dfba397b5c2a9757cfd14d8a2537f38dadf35d
 2021-03-25 16:36:10 -07:00
Wenhao Jia
054d641e30 Minor build fix. 
PiperOrigin-RevId: 365148648
Change-Id: Ib0bd821dfaa7943d04e6da397bfd9a4942129536
 2021-03-25 16:30:48 -07:00
Sachin Joglekar
b8e4404f60 Simplify FP16 delegation: accept FP16 DEQUANT nodes only if the entire graph will be delegated, else leave the FP16 DEQUANTs (which are no-ops after first inference) on CPU. Also modify unit tests to reflect this. 
PiperOrigin-RevId: 365143125
Change-Id: Id663b84e581db06d0e904a4d59e52a54e8824057
 2021-03-25 16:07:39 -07:00
Allen Lavoie
2ed50f444f Prefer namedtuple properties to wrapper properties in tf.Module tracking logic 
Avoids trying to simplify the logic in the process, since apparently we still need the workarounds.

PiperOrigin-RevId: 365142662
Change-Id: I63f57b0480e67c8c26cd1b196fffc1eca8b0e9d8
 2021-03-25 16:03:40 -07:00
TensorFlower Gardener
5d6cc7bf97 Merge pull request #48083 from netfs:fb 
PiperOrigin-RevId: 365142282
Change-Id: Ib8413ac7fa7f7f31ab2c8267e65cd74f2c0dde6b
 2021-03-25 15:58:59 -07:00
Francois Chollet
bb1087d4f1 Remove dependency of Keras on moving_averages.assign_moving_average. 
PiperOrigin-RevId: 365140127
Change-Id: Id7acefcecca36f27fa4e0355ad9ae36d13c80cbe
 2021-03-25 15:48:31 -07:00
A. Unique TensorFlower
1630c8771c Fix structured tensor.from_pyval for objects like [{},{}] and [[{}],[{},{}]]. 
PiperOrigin-RevId: 365139003
Change-Id: Iafd3dd187f3a4361097171af0b0a7a41602cfe03
 2021-03-25 15:42:21 -07:00
Rick Chao
77c3c34d7e PSv2/cfit: Add a note in Model.fit that batch-level calls are currently unsupported with ParameterServerStrategy. 
PiperOrigin-RevId: 365136309
Change-Id: I030b16ae6d62a9b4cd8535d9206fb1a513c27956
 2021-03-25 15:24:50 -07:00
Reed Wanderman-Milne
2527a7ada3 Fix and re-enable segment determinism tests. 
The tests previously passed in Eager mode but not graph mode. Now they pass with both.

PiperOrigin-RevId: 365132406
Change-Id: I083e39afd8eb3ed4efc04846a5b067bc9a339bec
 2021-03-25 15:06:40 -07:00
Penporn Koanantakool
795969c326 Add PluggableDevice, oneDNN, and ACL changes to the release note. 
PiperOrigin-RevId: 365131281
Change-Id: I3ad4b4d9d1718254b379b0aeb737d549badf0951
 2021-03-25 15:02:34 -07:00
Prakalp Srivastava
9cc83b99a1 Update input-output alias information of the HLO module compiled using fallback. 
Uses the "tf.aliasing_output" arg attributes on the MLIR TensorFlow module to get the output to input aliasing.

This also makes the `mutable_proto()` function of `XlaComputation` public. This is necessary to modify the HLO module generated from the old bridge with aliasing information.

PiperOrigin-RevId: 365129546
Change-Id: Ic8da05c1f33a31b1325ba1855467877a28d9fa04
 2021-03-25 14:56:02 -07:00
Abhijit Karmarkar
9be40198b2 fix broken bazel macros, to fix TF serving build. 2021-03-25 14:50:30 -07:00
Raman Sarokin
6ce39236b6 Added partial support of logical operations. 
Supported cases:
Logical op -> bool tensor -> Cast -> float tensor.
bool tensor must be consumed only by Cast.

PiperOrigin-RevId: 365124686
Change-Id: I96dcdc86b16b9924ed2e9cd36277a13be6905dc9
 2021-03-25 14:37:37 -07:00
Jared Duke
b29119add7 Avoid using ruy by default for non-optimized paths 
PiperOrigin-RevId: 365124251
Change-Id: Ia799b63f632fd9890716f4167b70ff5287832c3e
 2021-03-25 14:31:11 -07:00
TensorFlower Gardener
75bc5ca8da Merge pull request #48061 from tfeher:trt_erf 
PiperOrigin-RevId: 365123391
Change-Id: Iaff5ddba27128a27513623f40358dfb9f58823d3
 2021-03-25 14:24:55 -07:00
Penporn Koanantakool
9159209df2 Fix "ld: library not found for -lrt" issue in macOS builds. 
PiperOrigin-RevId: 365121503
Change-Id: Ie4b89d58382e64b000fd7f27ffbac57a57e9f7cb
 2021-03-25 14:17:51 -07:00
Juhyun Lee
c67f680218 Move some custom ops back to model builder. 
PiperOrigin-RevId: 365116594
Change-Id: Ieea3ad01cc6518c5ccd1cb971ef941864fc9db8b
 2021-03-25 13:55:23 -07:00
Rick Chao
6fa290d1e1 PSv2/cfit: Documentation update in parameter_server_strategy_v2 for support with Model.fit. 
PiperOrigin-RevId: 365114639
Change-Id: I2a62cd8ddea555817319555373c7852d07615337
 2021-03-25 13:45:14 -07:00
Xiao Yu
2a0da9a9f3 Use ImmediateExecutionContext in TFE_NewCustomDeviceTensorHandle. 
PiperOrigin-RevId: 365113817
Change-Id: Ie9f4e48eb9c9786b5d3ce6d3b33a71e11269c973
 2021-03-25 13:40:53 -07:00
George Karpenkov
130f0e3165 [DistStrat] Rollback of rollback of not using merge-call inside an XLA context 
Additionally always initializes the collective ops in MirroredStrategy.

The failing test was fixed to explicitly require soft placement

PiperOrigin-RevId: 365108870
Change-Id: Iac1a3b1dd6a311ed55fdf8c85ddc6d4c374587ab
 2021-03-25 13:21:32 -07:00
Peter Hawkins
362d8c5197 [JAX] Refactor NaN/Inf checking in jitted functions. 
Avoid performing NaN/Inf checking in the common path for calling a jit-ted function. Instead, add a global/thread-local `posthook` function that, if, set, the C++ jit code calls with the inputs (function, args, kwargs, outputs). Use the posthook feature to implement NaN checking.

Add a `_cache_miss` attribute to the C++ JIT function objects to allow the NaN checking code to extract and call the cache miss function.

PiperOrigin-RevId: 365108787
Change-Id: I6e4158a07c9a2f25bb8dc71ef3056e6336f98aa9
 2021-03-25 13:16:19 -07:00
Francois Chollet
f80dfac82c Remove dependency of Keras on moving_averages.assign_moving_average. 
PiperOrigin-RevId: 365105680
Change-Id: I44cf2d91f8b0bfe565f66521fa4d599b001d1269
 2021-03-25 13:02:14 -07:00
Jiri Simsa
6b525ec5a6 [tf.data] Minor test cleanup. 
PiperOrigin-RevId: 365101396
Change-Id: I31d1c4230d50ec4ddbbdf6ce0538f1cb19b3eb51
 2021-03-25 12:46:26 -07:00
Thomas O'Malley
d592b6e634 Support Callbacks with ParameterServerStrategy in Model.fit 
- Error out if batch-level Callbacks are passed.
- Add support for fetching RemoteValues in to_numpy_or_float util.
- Ensure TensorBoard epoch-level logging works.
- Change default verbosity to epoch-level for PSStrategy

PiperOrigin-RevId: 365100910
Change-Id: Ia0b6eb87e184e75e0cae1ae57f3e35fb08b9b365
 2021-03-25 12:41:00 -07:00
Francois Chollet
66f5830642 Remove Keras dependency on private TF API no_automatic_dependency_tracking_scope. 
PiperOrigin-RevId: 365097996
Change-Id: I7ee1c7583d9790935d237e33f0fd8f669f6d5739
 2021-03-25 12:29:32 -07:00
Geeta Chavan
0ce4dc010a Prefer namedtuple properties to wrapper properties in tf.Module tracking logic 
PiperOrigin-RevId: 365091942
Change-Id: I95a93067cf1b7295e3899fe3a9394a94bb13c572
 2021-03-25 12:02:48 -07:00
TensorFlower Gardener
21aac84302 Merge pull request #46167 from Intel-tensorflow:quintinwang/default_device_ops 
PiperOrigin-RevId: 365089950
Change-Id: I6a83950dfc5e869e784e5eeb21d9aca51e64c411
 2021-03-25 11:52:22 -07:00
Tamas Bela Feher
5cb0941f7c TF-TRT Add Erf unary op converter 2021-03-25 19:46:41 +01:00
Penporn Koanantakool
aaccd281d3 Add a missing link option for clock_gettime. 
PiperOrigin-RevId: 365083195
Change-Id: I9291171feeb4ba5e4c5bcd22ffd61ed4f7ed4615
 2021-03-25 11:26:45 -07:00
A. Unique TensorFlower
3c462a926f Internal change 
PiperOrigin-RevId: 365082726
Change-Id: Ic5bfdbccedbe63e554ca6b9407f591a3913f8798
 2021-03-25 11:20:38 -07:00
Sanjoy Das
a46ba37466 Make sparse_to_dense_op_py_test a cuda_py_test 
This is to ensure that the GPU kernels added in PR #47234 are tested.

PiperOrigin-RevId: 365079858
Change-Id: Ibcc511cf3d3fc162c28efb6fa10d9bd794dd58a7
 2021-03-25 11:10:20 -07:00
TensorFlower Gardener
9d289452e6 Merge pull request #47988 from tfeher:trt_einsum_dynamic 
PiperOrigin-RevId: 365076348
Change-Id: I5fe50c0618caf33444156bc79114a817788f13db
 2021-03-25 10:56:52 -07:00
TensorFlower Gardener
19a2594380 Merge pull request #47969 from Intel-tensorflow:yang/graph-log 
PiperOrigin-RevId: 365072077
Change-Id: I2ee01cc4318a1c958e2852fe8663e87b76672a4f
 2021-03-25 10:49:32 -07:00
Katherine Wu
e51fbcabc5 Fix call function tracing when parent layer does not have training argument while having child layers that do. 
PiperOrigin-RevId: 365072075
Change-Id: Ib115573580aba6361383523f9add76028152e947
 2021-03-25 10:42:27 -07:00
Scott Zhu
f06bfe0d8d Remove the usage of tf private API ops._as_graph_element. 
PiperOrigin-RevId: 365064750
Change-Id: I02a6f074e064f0aafec3f0b4befc1b5847be24a3
 2021-03-25 10:09:51 -07:00
A. Unique TensorFlower
a08001f32a Internal change 
PiperOrigin-RevId: 365061180
Change-Id: I38205ff7e1183865a344cb9dce8b69de0d090a50
 2021-03-25 09:59:47 -07:00
A. Unique TensorFlower
6e170748b1 Do not log error on non-Android platforms when libneuralnetworks.so is not found. 
PiperOrigin-RevId: 365060905
Change-Id: I75913ad8969633b72ffe889fd2958f247a054625
 2021-03-25 09:54:01 -07:00
A. Unique TensorFlower
505f622908 Remove no_oss, no_pip, and no_windows tags for unit tests that passed without them. 
PiperOrigin-RevId: 365052542
Change-Id: I6d5d97efce507319e9a11dc6b501926c6481541d
 2021-03-25 09:18:38 -07:00
TensorFlower Gardener
9e26f3b56a Merge pull request #47775 from alenik01:aarch64_build_patch 
PiperOrigin-RevId: 365049267
Change-Id: I839a9fb319ccc347b50defeaabb1c3f1ec3cfb5c
 2021-03-25 09:03:00 -07:00
TensorFlower Gardener
324779bc63 Merge pull request #48042 from pengzhendong:master 
PiperOrigin-RevId: 365043051
Change-Id: Ia0c2ecd54e32d7829d9f8b8ede5292b83fbde3df
 2021-03-25 08:42:32 -07:00
A. Unique TensorFlower
7f2db2bbe1 Fix a bug in replay computation 
PiperOrigin-RevId: 365042972
Change-Id: I399ff8323695cb50df5ca1d58e8f8c35835f43b2
 2021-03-25 08:35:50 -07:00
TensorFlower Gardener
aaff3bafb4 Merge pull request #48000 from AdityaKane2001:master 
PiperOrigin-RevId: 365038838
Change-Id: I24d941387807f459efade8f61959f7ef24052cbf
 2021-03-25 08:17:34 -07:00
Adrian Kuegel
9a20283a53 kernel-gen-opt: set preloadDialectsInContext to false. 
This requires specifying dependent dialects in the ShapeSimplification pass.

PiperOrigin-RevId: 365023585
Change-Id: Ib33b73e68cbb941b637a6882ae84c0e49f88a6c1
 2021-03-25 06:38:19 -07:00
Geoffrey Martin-Noble
07dc7d984a Add folder for HLO NotOp 
PiperOrigin-RevId: 364989658
Change-Id: Ic699b4cd9366037e6bb534d55a2288c1283ee744
 2021-03-25 02:14:19 -07:00
A. Unique TensorFlower
f5b7f1f08c compat: Update forward compatibility horizon to 2021-03-25 
PiperOrigin-RevId: 364989100
Change-Id: I2237de43af83ad947e0a95728fbcd233faf7258e
 2021-03-25 02:09:40 -07:00
A. Unique TensorFlower
cc99437c60 Update GraphDef version to 716. 
PiperOrigin-RevId: 364988974
Change-Id: I628f5c8a3a992c6b4ae4ab0736e29893ef40e14f
 2021-03-25 02:05:37 -07:00
Chao Mei
48eb150c8e Fix the comment about where the weak declaration of AcquireXNNPACKDelegate could be found, and combined code sections w.r.t TFLITE_BUILD_WITH_XNNPACK_DELEGATE macro. 
PiperOrigin-RevId: 364983825
Change-Id: I58400bd0023b5e09a84f68872a6f9199d4edd0bb
 2021-03-25 01:16:25 -07:00
Sanjoy Das
b6de500fd0 The kernel does not build with ROCm so disable it in ROCm for now 
Doing this to make the ROCm build green.

PiperOrigin-RevId: 364982542
Change-Id: Idf5f94fbcbebf0b6d1da6d468905b86385fb2392
 2021-03-25 01:06:58 -07:00
Francois Chollet
38fca54c85 Scrub Python 2 stragglers from the Keras codebase. 
PiperOrigin-RevId: 364971112
Change-Id: I8326b3132480e6d0a7c14b88ebbfcd0e126115ca
 2021-03-24 23:19:55 -07:00
A. Unique TensorFlower
373883e4cc Add logging when individual TF2 behaviors are enabled/disabled 
PiperOrigin-RevId: 364965268
Change-Id: Ieefcbac42d8749e1e0cf6939ea3db38d0a769c4d
 2021-03-24 22:27:05 -07:00