| Commit message (Collapse) | Author | Age |
|
|
|
| |
Change: 149498198
|
|
|
|
| |
Change: 149496080
|
|
|
|
| |
Change: 149495758
|
|
|
|
| |
Change: 149492005
|
|
|
|
| |
Change: 149490754
|
|
|
|
| |
Change: 149488853
|
|
|
|
|
| |
Remove unnecessary temp memory recording which belongs to OpKernelConstruction instead of OpKernelContext.
Change: 149488690
|
|
|
|
|
|
| |
LocalClient::Compile() and LocalExecutable::Run().
Change: 149482633
|
|
|
|
|
|
| |
reduce the clutter in the usage of EstimatorSpec. Forexample, We can check scaffold.saver instead of if scaffold and scaffold.saver...
Change: 149482422
|
|
|
|
| |
Change: 149468728
|
|
|
|
|
| |
res_t corresponds to fn1 and res_f corresponds to fn2.
Change: 149461652
|
|
|
|
| |
Change: 149454449
|
|
|
|
| |
Change: 149448989
|
|
|
|
| |
Change: 149446004
|
|
|
|
| |
Change: 149441560
|
|
|
|
| |
Change: 149431599
|
|
|
|
|
| |
This code path is unused; Tensorflow ended up settling on having XLA allocate result buffers using Tensorflow's allocator. Remove it to reduce the proliferation of ExecuteXYZ() methods.
Change: 149423775
|
|
|
|
| |
Change: 149423117
|
|
|
|
| |
Change: 149423095
|
|
|
|
|
|
|
|
|
| |
Adds missing MaskedMatmul kernel to the CMake build.
Adds trivial shape functions for tf.contrib.factorization ops.
Enable testing for tf.contrib.factorization in the CMake build.
Fixes #7935.
Change: 149391496
|
|
|
|
| |
Change: 149388619
|
|
|
|
|
|
| |
problems.
Change: 149376544
|
|
|
|
| |
Change: 149376492
|
|
|
|
| |
Change: 149375959
|
|
|
|
| |
Change: 149375859
|
|
|
|
| |
Change: 149374684
|
|
|
|
| |
Change: 149369506
|
|
|
|
|
|
|
| |
We need a system for getting arbitrary metadata and assets for TensorBoard plugins. Examples include configuration protobufs, vocabulary files, and sprite images. We need to be able to declare the plugin assets at graph construction time, and then serialize them to disk using the tf.summary.FileWriter. And we want to do so using the FileWriter.add_graph API, so that new features will Just Work for existing code that instantiates a FileWriter and adds the graph.
This CL adds a system meeting those requirements. Plugin authors will create an asset class that extends tf.summary.PluginAsset, and implement a serialize method that dumps all the required assets to disk. At the time of graph construction, tf.summary.get_plugin_asset(AssetClass) gives them the current instance for their plugin assets (constructing it if necessary). They may then configure it. When tf.summary.FileWriter.add_graph is called with the graph, it retrieves all PluginAssets and makes a subdirectory for each one. Then, it calls serialize_to_directory on each asset.
Change: 149367010
|
|
|
|
| |
Change: 149364578
|
|
|
|
| |
Change: 149363494
|
|
|
|
| |
Change: 149360110
|
|
|
|
|
|
| |
Android Studio builds on Windows. Resolves #6385
Change: 149358567
|
|
|
|
| |
Change: 149358501
|
|
|
|
|
|
| |
documentation, so GLOBAL_VARIABLES is used only when collections==None.
Change: 149357125
|
|
|
|
|
|
| |
MakeShapeFromPartialTensorShape(). Update existing users of MakeShapeFromShapeProto to use the new helper methods where possible.
Change: 149353816
|
|
|
|
| |
Change: 149349958
|
|
|
|
|
|
|
|
| |
* Changed URL of in several installation examples.
* Added the URL for Python 3.6.
* Clarified the fix for protocol-dependency issues (Github issue #7940).
* Fixed link to Docker documentation (Buganizer issue #35347827)
Change: 149340618
|
|
|
|
| |
Change: 149340560
|
|
|
|
|
|
| |
arguments.
Change: 149340183
|
|
|
|
| |
Change: 149337075
|
|
|
|
|
|
| |
have determined that buffer forwarding should be safe on Kepler and all newer GPUs as long as no reads via __ldg() follow writes to the same address in the Cuda kernel.
Change: 149334900
|
|
|
|
|
|
|
| |
Generalize common backward convolution dimension computation code to support > 2 dimensions.
Fix crash for zero-element Tensors passed to Conv3DBackpropFilter on CPU.
Change: 149329330
|
|
|
|
|
|
| |
SDK tools version to match bazel build and ensure it adds libtensorflow_inference.so to the APK as well. Resolves #8059
Change: 149313216
|
|
|
|
|
| |
For some reason they were hidden in a refactoring to bijector impl.
Change: 149309294
|
|
|
|
| |
Change: 149308271
|
|
|
|
|
|
| |
- Sprite visualizer in 2D mode: Make the point size be the thumbnail size, since in 2D mode we only allow semantic zooming.
- Support '~' (homedir) for the sprite_image_path in the projector_config.pbtxt
Change: 149303375
|
|
|
|
| |
Change: 149296141
|
|
|
|
|
|
| |
The kernel can read its (forwarded) input after it has started writing
to it, leading to non-deterministic and incorrect results.
Change: 149215447
|
|
|
|
| |
Change: 149211889
|
|
|
|
| |
Change: 149209556
|