hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
fe57c369c2ffaca76892b62ea727ccbba3d6cc48 | Ensure that TensorBoard's multiplexer is always loaded, even if there's lots of data.
In current code, if loading data will take a while, requests to the multiplexer will error while data is loading, because some of the EventAccumulators aren't loaded.
Change: 113769073 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,530,639,000 |
d6f732205870e87b495e1c01d7b79c6512339021 | Change repository names to contain _ instead of -
Change: 113773458 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,533,409,000 |
017c7b71bfb6187289bea3fcb582cf4dd987586b | Use select to turn android_tensorflow_lib into empty library on non-Android builds.
Change: 113774390 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,533,901,000 |
dd3e7502b08c6ad324a673337a0d3df5bc9cebac | TensorFlow: fix framework_function_test to only depend on
the additional libraries it needs.
Change: 113774883 | Vijay Vasudevan | vrv@google.com | 1,454,534,207,000 |
f8741f131d30050205d84034fe13450ed9c36cf2 | TensorFlow: fix to directory_watcher_test on mac. Fixes #964.
Change: 113779032 | Vijay Vasudevan | vrv@google.com | 1,454,536,477,000 |
bca664cf253e2da345e005b0d90cf4ffa8b3c588 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,454,537,790,000 |
ee649764d5635d1acfaa74a892542e03018e554f | Register ZerosLikeOp for all number types and bool
Fixes #952 | Martin Wicke | wicke@google.com | 1,454,543,081,000 |
0620d2ca5e9e19ae6171aadab1474705d4629eb4 | Merge pull request #981 from tensorflow/fix-952. | Martin Wicke | wicke@google.com | 1,454,543,112,000 |
72b6a6e974691deed1fb1c1de72192d493b64911 | Avoid unecessary conversion of the graph back to a graphdef
Change: 113781984 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,538,056,000 |
545dee2e9897e424b641f092eec9ffd4a277f9d1 | Put device-to-device GPU memory copies on a different stream.
Change: 113784244 | Xiaoqiang Zheng | zhengxq@google.com | 1,454,539,374,000 |
937ed32c6ecf4be1177e8bcbce2596aa28345613 | Add missing headers to android_tensorflow_lib declaration.
Change: 113786181 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,540,466,000 |
aba15cf9e9ba102c15b83b832afee9a9b107bc36 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,454,543,429,000 |
4f3bcc28bc3f742063f129fc0af09948f3872033 | Adding build info print-out for Jenkins Description Setter | Shanqing Cai | cais@google.com | 1,454,547,182,000 |
b09f5d87ceda88db719c6c47aeea69459deebe31 | Merge pull request #979 from caisq/jenkins-desc-setter-1. | Martin Wicke | wicke@google.com | 1,454,547,243,000 |
4a6d3bddbd3b5697374fe31574900c655de09904 | Python PIP test-on-install
Addressing jendap's comments
Adding --upgrade option to pip install
test_on_install: files not installed by required by the unit tests
Removing --upgrade flag for "pip install"
Renaming test_on_install.sh to pip.sh
Getting rid of absolute path /tensorflow
and making sure that the whl file and testing working directory
are in the workspace.
Putting output of individual tests in log files
Correcting a blacklisted test
Adding progress string
Echo log directory
Improving error message when container type is incorrect
Mac-friendly absolute path function
Verbose pip install
Adding pip upgrade option
More intelligent local python path finding
Using longer test log file path in case test file names have duplicates
Fixing bug of nonexistent directory for log files
Removing ci_test_on_install.sh and other changes
Moving the whl and test folder under one common parent folder: pip_test
Adding pip_test to .gitignore
Printing detailed folder if the .whl file can't be found in it
Changing succeeded to PASSED and similar wording changes
Adding new line at the end of .gitignore | Shanqing Cai | cais@google.com | 1,454,126,541,000 |
813c541f71dc5a0de5514f31d63bd1002b0dd243 | Merge pull request #939 from caisq/dev1a. | Vijay Vasudevan | vrv@google.com | 1,454,561,371,000 |
1ec8cedd50cb708290b2c895b51844dce22cfef5 | Revert "Python PIP test-on-install"
This reverts commit 4a6d3bddbd3b5697374fe31574900c655de09904. | Vijay Vasudevan | vrv@google.com | 1,454,562,225,000 |
1fdfd1f57b08dd283da2c830801470acd7dd2525 | - include <algorithm> for std::min.
- replace memmem with std::search, which is more portable. | Zero Chen | chengang31@gmail.com | 1,453,685,600,000 |
9a822330bbabca748409e1523dda7813ecba2304 | Merge pull request #887 from cg31/upstream. | Vijay Vasudevan | vrv@google.com | 1,454,562,440,000 |
8b91a166ea94f1df8c1b83b1b2aecf00a612e40e | Added missing PY2AND3 build annotation
Without this line pip package cannot be built in python3 | Chen Yu | chenyyuio@gmail.com | 1,454,481,178,000 |
36aca649e3a9e21215105d1f2f2fa5763516f2e0 | Merge pull request #972 from cwhy/patch-1. | Vijay Vasudevan | vrv@google.com | 1,454,569,544,000 |
aed59cedf8af99e1f5b6a4a3ad02a78cbcbe3ccb | Silenced some bogus compilation warning generated by nvcc | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,548,552,000 |
993b243484fbec4b966c7d6cbd58e6c58686672b | Merge pull request #983 from benoitsteiner/master. | Vijay Vasudevan | vrv@google.com | 1,454,569,591,000 |
477dc94f262e11d0f4651e5fb94eb8cb3dd6e74e | Upgraded to the latest version of Eigen that brings misc performance and
stability improvements
Change: 113791782 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,543,395,000 |
ed56c079c0ab5833b8ca7deb048089ff9a2acc87 | Silenced an annoying compilation warning
Change: 113815470 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,564,842,000 |
95f92f1f540ecc2b7e00b2903e991de36970b643 | Improve collection handling for new variables
Let Variable constructor (and forwarding functions) accept any iterable for the collections argument.
Add a add_to_collections function that adds a value to all the given collections, but makes sure the list of collections has no duplicates. Use this function in Variable constructor. This has the side effects that if you specify a list with duplicates for the collections argument, the variable will now only be added once to each collection, before it was added with the multiplicity of the collection in the collections list.
Change: 113821818 | Martin Wicke | wicke@google.com | 1,454,573,549,000 |
e0810f48fdbcd9a2f18a42485383091dd0622a4e | Fix assertions in `Graph.name_scope()` docs.
Change: 113846943 | Derek Murray | mrry@google.com | 1,454,599,488,000 |
fa181dd2db44504fa5d5f64782b3640170fc95d2 | Merge commit for internal changes | Manjunath Kudlur | keveman@gmail.com | 1,454,599,713,000 |
32180c011ce7ab053010f868ac371a55d41bd98f | Fix typos in strings and comments of core/kernels. | Dongjoon Hyun | dongjoon@apache.org | 1,454,610,921,000 |
2327abc98900740fadcdf4c50d75e9585488043c | Merge pull request #992 from dongjoon-hyun/fix_typos_in_kernel. | Martin Wicke | wicke@google.com | 1,454,617,576,000 |
31c58d41257211fc8485a36ebfd6a307a0615904 | Revert "Revert "Python PIP test-on-install""
(Roll forward of the original change)
This reverts commit 1ec8cedd50cb708290b2c895b51844dce22cfef5. | Vijay Vasudevan | vrv@google.com | 1,454,620,222,000 |
7fbee806b504c18be69ec656486896feed8a47dd | Shuffling letters before assigning them to training and validation datasets
Without this change we are getting poor results since training and validation
sets are not well chosen. | Lukas Krecan | lukas@krecan.net | 1,454,572,930,000 |
bf9325d452fafda689fe8ee24e2461d95305ae5d | Merge pull request #984 from lukas-krecan/master. | Vincent Vanhoucke | vanhoucke@google.com | 1,454,621,576,000 |
8cda66e19c0bdc6d8fc373fdcefcc644ab5ecfa3 | Make python_check relocatable
Uses $location in the genrule and a path that's relative to the
script to find expected files so that this works both as a local
rule and if tensorflow is used as a remote repository. | Kristina Chodorow | kchodorow@google.com | 1,454,620,855,000 |
3511bf32beb55ab3f2faf5db68ebf18e8f20c824 | Merge pull request #996 from kchodorow/master. | Martin Wicke | wicke@google.com | 1,454,627,302,000 |
ff01ec49e47053c7f1ea9d9809bedd4d1fe9ee91 | Remove the batch_select_op_test I accidentally introduced. This test class was
moved inside of cwise_ops_test; and this code is outdated.
Change: 113857436 | Eugene Brevdo | ebrevdo@gmail.com | 1,454,606,579,000 |
73327e4f07c0f57c8b4348795c7200a5eeef6ae8 | Fix a typo in the `Graph.name_scope()` docstring.
Change: 113857852 | Derek Murray | mrry@google.com | 1,454,606,847,000 |
cfe676dbad39dc254c7eb998e6029084133c1484 | Deleted 2 unnecessary declarations
Change: 113857958 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,606,912,000 |
791f7a3a490720fe42ceee4f6519eaf77ff2f1ca | tensorflow: relax the dimensionality check for tf.squeeze.
This CL relaxes the shape inference constraints on tf.squeeze.
Previously, trying to squeeze a dimension whose build-time shape
was unknown would result in the squeezed shape being completely
unknown. This causes errors on some otherwise working programs that
rely on TensorFlow's build-time shape inference.
With this change, it's assumed that, if there is an explicit squeeze_dims
argument to tf.squeeze() and the dims in question are not known at shape
inference time, the input dims will be assumed to be 1. If they are not,
an error will be thrown at runtime just as before.
Change: 113863596 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,610,127,000 |
4c6cb2392e1b933f876979975a1274cdb9480e97 | Make gpu_event_mgr_test less flakey.
The various subtests are supposed to ensure that tensors queued for
unref after some GPU stream event completes actually do get deleted.
However, the way in which this is normally accomplished is
asynchronous in a couple of ways. First, completed GPU stream events
are ordinarily detected by an independent polling thread. In many of
the tests, this is bypassed by synchronously calling into the polling
function to harvest completed events. However, event completion
itself is asynchronous and dependent on the CUDA driver. It is
suspected that the test failures have been due to delayed event
completion, such that the test criterion is tested prematurely.
Unfortunately simply synching on the CUDA device does not fix the
problem. But in these tests there's never any actual GPU work queued
on the streams, they only contain events that are eligible for
immeidate completion. So the fix applied here is to put a while loop
around the synchronous poll call that clears the queue, achieving the
expected state.
Change: 113877300 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,617,743,000 |
cfbd9f9b0eb27dcce8ca1778650fdcc57dbff508 | Merge nodes do not currently forward shapes correctly. Set statically the
known parts of the output shape are set after _dynamic_rnn.
Change: 113880947 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,620,105,000 |
1545866e734684bd05cb2de63498aad0380eb202 | Make tracking_allocator pass on the allocation attributes.
Change: 113886273 | Xiaoqiang Zheng | zhengxq@google.com | 1,454,623,322,000 |
b064d6199d2553c9c6ac6013c9f62b9af47f4524 | Fixes ZerosLike on GPU
Change: 113889504 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,625,122,000 |
6607d30417454bec519c6d75c91936d4ab4e9c1c | Improving seq2seq code style and documentation.
Change: 113892235 | Lukasz Kaiser | lukaszkaiser@gmail.com | 1,454,626,713,000 |
9c7f71244e7e91e83b7ef79ef0841cfe08d2154d | Updated documentation for adding an op.
- Added functions to get the directory where the headers and the library are
installed for use with -I and -L options to the compiler.
Change: 113894868 | Manjunath Kudlur | keveman@gmail.com | 1,454,628,224,000 |
b6a00d90764b07e9748fef2011ea1afdddefaa87 | Move away from "kernels" rule so that we can eventually
make the kernels/ directory its own package.
Change: 113895747 | Josh Levenberg | josh11b@tensorflow.org | 1,454,628,767,000 |
cd0b118025bac1ac4e33b3b6b580235102781cdc | Python3 test: getting easy ones out of the way | Shanqing Cai | cais@google.com | 1,454,619,891,000 |
e56438b28fbc8b2328a2f386c49a7e88adb3e07b | Merge pull request #997 from caisq/fix-python3-2. | Martin Wicke | wicke@google.com | 1,454,631,456,000 |
ea4268ed34989dfc863b7903f6131f8ac0aa4eff | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,454,633,697,000 |
eb0775fdae3445dd9d9a4775527f7e7f720e528f | Initial code for resnet example | terrytangyuan | terrytangyuan@gmail.com | 1,454,638,606,000 |
85309efad5f96782c85cc94a75f673f3b81a2103 | * Refactored using skflow helper models | terrytangyuan | terrytangyuan@gmail.com | 1,454,644,251,000 |
31521d500ca42dc17f0fea9d7328723d88d1e5e2 | * Refactored batch_normalize | terrytangyuan | terrytangyuan@gmail.com | 1,454,645,258,000 |
7bc52144b99ff2109f0cc9888a1f98eac409804a | + batch_norm_ops and option for batch_norm in conv_ops | terrytangyuan | terrytangyuan@gmail.com | 1,454,647,246,000 |
969b739def3b217d92c5ac9a84fc97f07d2d22b3 | * Fixed imports and pylint | terrytangyuan | terrytangyuan@gmail.com | 1,454,649,371,000 |
7b5cc04ef6d5cf6fe317be5234ef080fe348e1a9 | * Replaced with skflowow.ops.conv2d | terrytangyuan | terrytangyuan@gmail.com | 1,454,653,740,000 |
11a61b5fa479b77fd2cf3e0df05f0f82ee17b4ba | + Update ops during training | terrytangyuan | terrytangyuan@gmail.com | 1,454,653,889,000 |
a7961a6e5741f976c168bdcc0e0d4d4ce708813e | + resnet to README and ignored coverage for batch_norm | terrytangyuan | terrytangyuan@gmail.com | 1,454,654,835,000 |
e9488f9a89762da1bca00ad133028dfc969a7097 | + resnet conv_upscale layer | terrytangyuan | terrytangyuan@gmail.com | 1,454,657,950,000 |
29873795ceac5386c2673c59b3689ef2dc6bcb08 | CategoricalVocabulary returns len of mapping vocabulary including unknown token | Illia Polosukhin | ilblackdragon@gmail.com | 1,454,658,078,000 |
88e1be2ee13cc6709530f784cea8423182e00d15 | Merge branch 'master' of github.com:google/skflow | Illia Polosukhin | ilblackdragon@gmail.com | 1,454,658,087,000 |
60765a4b210fc54ec32a7729f8532a60c5217544 | Move the bulk of the TensorBoard server setup logic into its own file.
This will make it easier to write TensorBoard integration tests that test the
entire server by spinning up a server instance, pointing it at a log directory,
and making HTTP requests.
Change: 113899708 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,631,128,000 |
90a6144cef4e11caa739b25610b18f8ee0ee0c22 | Do not set shape on Variable if validate_shape is False, even if shape information is fully defined.
This is more consistent with the documented intent of validate_shape.
Change: 113908140 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,637,520,000 |
63b7a9807390e980e620d11d0bb2d179685d24fa | Use sparse_softmax_cross_entropy_with_logits in seq2seq in sequence_losses.
This removes the need for passing the number of classes, and is therefore
a breaking change to the API. If it breaks you: remove the parameter.
Change: 113911519 | Lukasz Kaiser | lukaszkaiser@gmail.com | 1,454,640,689,000 |
0c6c4b93dedf6a2c654e012a4ffe1df642834419 | Allows int32 on host memory work when a function passes int32 in/out.
Change: 113922125 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,655,143,000 |
16e91ab965d54ede8ceda946cd1db125a9901ec5 | Switch from newly-deprecated tensorflow/core:kernels to :all_kernels.
This will allow me to delete the kernels rule and then create a
kernels package.
Change: 113950706 | Josh Levenberg | josh11b@tensorflow.org | 1,454,689,281,000 |
434b79ec286c13bc0fb8b65ff824cf7ea3ee0c3e | Remove vocab_size argument for model_with_buckets.
Change: 113952458 | Manjunath Kudlur | keveman@gmail.com | 1,454,690,720,000 |
ce9c76f7dbff6226fa9ed0ea3338678028c641b4 | Merge commit for internal changes | Manjunath Kudlur | keveman@gmail.com | 1,454,690,963,000 |
aae0b8d33a6c2450784a2a9c6f57c8b0a4344a42 | Script for Jenkins parameterized matrix build
The script takes a few environment variables as input and invoke
ci_build.sh in the same directory in the proper way. It takes
information about the build type (cpu | gpu | android), Python version,
whether "-c opt" is used and whether PIP install-test is to be
performed. It automatically examines the parameters to determine the
correct build script to call and the argument flags to use with it. It
also examines whether Docker is available on the system to determine
whether the build will be done inside Docker or not. If GPU Docker build
is detectd, it'll map the CUDA devices and libraries automatically.
In addition, it looks at some optional environment variables to set
additional command flags for Docker and the build command. | Shanqing Cai | cais@google.com | 1,454,622,435,000 |
2eb674e76c579e835aaad687b0c62598a1312b57 | Merge pull request #998 from caisq/param-build-2. | Martin Wicke | wicke@google.com | 1,454,694,224,000 |
bf935c19ffccb3b84d807bbda9c983c1c156af4a | Fixing more tests in Python3
This CL fixes a number of (but not all) remaining Python3 test errors,
including:
//tensorflow/python:cwise_ops_test
//tensorflow/python:gradients_test
//tensorflow/python:topk_op_test
//tensorflow/python:parsing_ops_test
//tensorflow/python:learn_test
It also fixes a subset of the failing tests in:
//tensorflow/python:function_test | Shanqing Cai | cais@google.com | 1,454,619,891,000 |
4d8b1c5befb62d73487aa83eff0654ece342516a | Remove call to set seed that is then reset when used | Dan Becker | dansbecker@gmail.com | 1,454,715,103,000 |
7d634fc35616e27a22fc3c985ea8f8dc56a8b196 | Merge pull request #1006 from caisq/fix-python3-2. | Martin Wicke | wicke@google.com | 1,454,717,212,000 |
ca0108596018cda27742c9982419db34dead7e41 | Merge pull request #91 from dansbecker/remove-redundant-seed
Remove call to set seed. Seed is set again in line that would use it | Yuan (Terry) Tang | terrytangyuan@users.noreply.github.com | 1,454,720,292,000 |
3f06730be4c21d158b9714841098048bd48d89c6 | Enables `tf.image.resize_images()` to accept tensors for height, width.
Fixes #1001.
Change: 113956896 | Derek Murray | mrry@google.com | 1,454,694,066,000 |
94b0207d565ce304efb405e9af9e8ce1539e41d1 | Replay of PR #969 for eigen3
Change: 113963788 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,697,844,000 |
e8db9c8d18c725a91dbbb5c3abed2e47a5ca0b12 | Unconditionally use TopKV2 from Python now that 3 weeks have passed
We're past the forward compatibility window, so Python can freely use the more
flexible TopKV2 C++ op. The difference is that TopKV2 takes the k argument as
an input instead of an attr.
Change: 113965011 | Geoffrey Irving | geoffreyi@google.com | 1,454,698,451,000 |
2917e5bb00a70a66405f4b1996e48ffad1d3ae9f | Delete deprecated core:kernels target now that references to it have been
migrated to the new rules.
Change: 113980298 | Josh Levenberg | josh11b@tensorflow.org | 1,454,707,897,000 |
cd60028ff9c3882cd51641fcadea0259414e9460 | Internal changes to repo syncing process.
Change: 113982115 | Vijay Vasudevan | vrv@google.com | 1,454,709,025,000 |
155cb0ff84e9ece3bee0f06eb3bf5f92d5ca15ac | TensorFlow: make gpu_allocator_retry_test tagged with 'nomac',
since it is flaky and unneeded on the mac platform.
Change: 113983903 | Vijay Vasudevan | vrv@google.com | 1,454,710,078,000 |
3a30eb965698cdd885fcd68c8cd84e8ad59f154c | Deleted unused code
Change: 113987653 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,712,449,000 |
4bc28e2d3baa4b1a605a18f5d466b9bd2a6fdbd5 | Amend CROSSTOOL for newer Bazel
Change: 113989341 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,713,464,000 |
14efce84a4d1fceec3106e73100c20686c47e086 | Avoid casting the size_t returned by std::vector::size() into a signed 32bit integer
when iterating over the content of the vector.
Change: 113992100 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,454,715,345,000 |
408bbffa32eeef431d2e2c9357aba0117d957b9c | Make some of the reshape() error messages more self-explanatory.
Change: 113992253 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,715,448,000 |
5b37664fb3d66a2fc115fedb76a30bbbad1c5320 | Switch the CIFAR model to use the new tf.nn.sparse_softmax_cross_entropy_with_logits
op.
Change: 113992858 | David G. Andersen | dga@google.com | 1,454,715,899,000 |
8f80ef184034ef536efa4aee1dd15cfaf84a5c34 | Add support for allocators that require tensors with zero
elements to be backed by Tensor buffers. This is needed for
experimental device support.
Change: 113993978 | A. Unique TensorFlower | nobody@tensorflow.org | 1,454,716,709,000 |
136ac3c58ea2493ea6cfc90d3f186b8bc1929873 | Convert convolutional.py example to use new sparse_softmax_cross_entropy_with_logits op
Change: 113997793 | David G. Andersen | dga@google.com | 1,454,719,392,000 |
4754990b30f3096dbe34a19ea645e2db4e065ad8 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,454,720,882,000 |
5c95b45909fe8dc8fbde6ceb852d4d150d4d21c1 | * explicit bias for resnet | terrytangyuan | terrytangyuan@gmail.com | 1,454,722,417,000 |
5ce929404998d3c666f177f85e658a8c39d9715a | + Separate batch_norm for train and test | terrytangyuan | terrytangyuan@gmail.com | 1,454,771,715,000 |
241698b6ba6cd9b13d606a9e4603baa4f33891f2 | Move CPU/GPU memory copies into their own streams.
Change: 114000504 | Xiaoqiang Zheng | zhengxq@google.com | 1,454,721,471,000 |
3fa9676bbc41826689e9b0e11a45e3fbdceae258 | Consolidate the device function and device string handling in `tf.device()`.
The effect of this CL is to treat `with tf.device(device_name):` as
supplying a device function that *merges* `device_name` into the
device of ops created in that scope. (Merging is defined by
`tensorflow.python.framework.device.merge_device()`: essentially, for
each field defined in `device_name`, the merge function sets an op's
device to that if it has not already been set.) This makes it possible
to compose device blocks that set different parts of a device, and use
device strings in composition with device functions.
A secondary effect of this CL is that it causes `with
tf.device(None):` to interoperate properly with device functions. As
with other `tf.Graph` contexts, entering a `with tf.device(None):` now
has the effect of ignoring all currently set device functions in the
outer context.
This CL makes some breaking changes to corner cases in the
`tf.device()`, `tf.Graph`, `tf.Operation`, and `tf.Tensor` APIs:
* Within a `with tf.device(device_string):` scope, the given device
string will now be *merged* into the device for ops created in that
scope. See the implementation of
`tensorflow.python.framework.device.merge_device()` for details.
Previously, device strings were maintained in a single "default
device" field, rather than a stack, so device strings from outer
contexts would be completely ignored. To obtain the previous
behavior, use `with tf.device(None), tf.device(device_string):`
instead.
* Within a `with tf.Graph.device(None):` scope, no device functions
from the outer context will be executed.
Previously, the `None` applied only to the device string, and all
device functions would be applied unconditionally.
* The `tf.Graph.get_default_device()` method is removed, because it no
longer has a well-defined meaning. To create a no-op device scope,
you can simply use `with tf.device(""):`.
* The `tf.Operation.device` and `tf.Tensor.device` properties now
return an empty string when no device has been set for an op. This
makes it easier to write code like `with tf.device(op.device):`,
which is robust to `op` having or not having a device (in which case
the scope should be a no-op).
Change: 114003979 | Derek Murray | mrry@google.com | 1,454,725,147,000 |
e008ecf12607f691952a9bfe8971f406b30bc56e | Merge commit for internal changes | Manjunath Kudlur | keveman@gmail.com | 1,454,777,346,000 |
4b5929e2af3da46bc78f0c94a4857b4129c7a8a3 | + batch_norm_ops more params and available for non-convnet | terrytangyuan | terrytangyuan@gmail.com | 1,454,779,640,000 |
45b8212c666dcd0f08fde9d69dd60fb1255cdf20 | Minor comment spelling fix. | e3 | eerwitt@gmail.com | 1,454,791,109,000 |
5ab0602a6fb003029d40dc6b03de9b660f73330a | Merge pull request #1008 from eerwitt/minor-spelling-update
Minor comment spelling fix. | Vijay Vasudevan | vrv@google.com | 1,454,791,610,000 |
e235074045ee1bb2babb490e7e61716a8528ab8f | fix minor filename typo
fix minor filename typo | Moussa Taifi | moutai10@gmail.com | 1,454,803,257,000 |
5619a2ccc0fb517d9bbbcd16b7fba080efeb39dc | TensorFlow: fix word2vec timeouts on GPU by pinning model on
CPU. Right now word2vec is not well supported on GPU (missing
scatter sub, etc). We can remove this limitation later when
ScatterSub and other ops are all available on GPU. | Vijay Vasudevan | vrv@google.com | 1,454,782,761,000 |
7f80eb149b11ee343adb3f21f5c42f8ec131849e | Merge pull request #1007 from vrv/fix_word2vec_timeout. | Martin Wicke | wicke@google.com | 1,454,805,323,000 |
49fd28274b7a0c1e48bfb507cb73ead10d717e50 | Strip out external/XXXX/ prefix from GetPath in cc_op_gen
GetPath was not stripping external/XXX/ which make C++ compilation
of the generated header fail. All external repositories are added
to the include path in C++ compilation by Bazel. | Damien Martin-Guillerez | dmarting@google.com | 1,454,701,064,000 |
449f9dfc37c609bdaea7a4c647b93b827f7c63b6 | Merge pull request #1004 from damienmg/external_get_path. | Martin Wicke | wicke@google.com | 1,454,805,568,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.