- 25 3月, 2016 18 次提交
-
-
由 A. Unique TensorFlower 提交于
Change: 118080678
-
由 Vincent Vanhoucke 提交于
Change: 118080472
-
由 RJ Ryan 提交于
Change: 118078260
-
由 Xiaoqiang Zheng 提交于
Change: 118075047
-
由 Vijay Vasudevan 提交于
Also change DirectSessionTest to use the public Session constructor when possible, to discourage use of DirectSession as anything but an opaque object that implements the Session API. The one exception is the experimental API, which requires a known down_Cast. Change: 118072249
-
由 Craig Citro 提交于
This fixes situations like Vincent hit, where a stale base image would lead to new packages based on old base packages. Change: 118071412
-
由 Vijay Vasudevan 提交于
3 (but enough to trigger grpc limits). Reduced runtime in non-opt mode from 30s to 8s. Change: 118059583
-
由 Geoffrey Irving 提交于
This allows individual tests to depend on a subset of kernels. Also harden immutable_constant_op_test to not segfault if the unexpected happens. Change: 118059523
-
由 Eugene Brevdo 提交于
Change: 118058606
-
由 A. Unique TensorFlower 提交于
Change: 118058044
-
由 Olivia Nordquist 提交于
Change: 118049024
-
由 Vijay Vasudevan 提交于
Change: 118047026
-
由 A. Unique TensorFlower 提交于
1. Add a BufferBase base class for Buffer, and define the non-type-dependent functions there. 2. Pull common LOG(ERROR) in tensor.cc out from a templated function. Change: 118046104
-
由 A. Unique TensorFlower 提交于
- Add a RunOptions proto in RunStepRequest. - Still need to pass it back to the grpc session. Change: 118045453
-
由 Derek Murray 提交于
Missing device errors are handled at a higher level in the distributed runtime. Logging a warning when a task is not found in a particular channel cache is unhelpful, because the runtime may search multiple caches (e.g. one per job) and misses are to be expected. Change: 118038406
-
由 A. Unique TensorFlower 提交于
for mobile apps): (1) Change many interfaces in node_def_builder.h, node_def_util.h, op_kernel.h, node_builder.h, mirror_pad_mode.h, padding.h to use 'StringPiece', rather than 'const string&'. The interfaces that were changed tend to be heavily used in the registration of ops and kernels, and often caused extra string construction code to be emitted in the macro expansion of each op or kernel registration. (2) Move some repetitive CHECK operations into non-inlined routines in tensor.cc, rather than having them in inlined or templated routines in tensor.h (new Tensor::CheckDataType, Tensor::CheckTypeAndIsAligned, and Tensor::CheckIsAlignedAndSingleElement routines) (3) Factored out internal template<size_t NDIMS> Tensor::FillDimsAndValidateCompatibleShape routine, to be shared across more specialized templated routines (typically specialized on both DataType and NDIMS). (4) Added new non-inlined TensorShape::CheckDimsMatch(int NDIMS) routine in tensor_shape.cc, that can be called from various TensorShape routines templated on NDIMS. (5) Don't inline single-argument StrCat, since it involves a string creation, etc. (6) Remove inline keyword from template <typename... AV> StrCat version that handles 5 or more arguments. Reduces text size for third_party/tensorflow/core/libandroid_tensorflow_lib.so built in Google build environment by 1.43%, as measured by: % blaze build -c opt --config=android_arm \ third_party/tensorflow/core:android_tensorflow_lib % size blaze-bin/third_party/tensorflow/core/libandroid_tensorflow_lib.so Change: 118036659
-
由 A. Unique TensorFlower 提交于
Simplify loss ops, only return scalar in public API (see discussion at go/tf_contrib_ops). Add softmax loss. Change: 118034691
-
由 Dan Smilkov 提交于
Replace graph_def with graph when passed to the SummaryWriter and update tutorials to reflect the new API. Change: 118033430
-
- 24 3月, 2016 22 次提交
-
-
由 David G. Andersen 提交于
Change: 118000502
-
由 Vijay Vasudevan 提交于
Change: 117999025
-
由 A. Unique TensorFlower 提交于
Change: 117998775
-
由 A. Unique TensorFlower 提交于
inlined into the call site regardless of the graph optimizer option setting. Change: 117994922
-
由 A. Unique TensorFlower 提交于
sess = tf.Session() print(sess.run(norm)) print(sess.run(norm)) sess = tf.Session() print(sess.run(norm)) print(sess.run(norm)) Change: 117994542
-
由 A. Unique TensorFlower 提交于
Use this method with the `with` keyword to specify that ops created within the scope of a block should be added to this graph. Change: 117993733
-
由 A. Unique TensorFlower 提交于
Change: 117993641
-
由 Dan Smilkov 提交于
Adding ability for users to store `RunOutputs` information to the events file and serve it via the TensorBoard back-end. The `RunOutputs` information contains execution statatistics, such as compute time and memory usage for each node in the subgraph executed by a particular `session.run()`. A follow-up change will add the front-end support to overlay this data onto the graph. Change: 117981334
-
由 Dan Mané 提交于
Change: 117978853
-
由 Zongheng Yang 提交于
Change: 117977121
-
由 Dan Mané 提交于
Change: 117976198
-
由 Yuan Yu 提交于
In backprop, we may create zero tensors for the gradient of cond. Before this CL, the zero tensors are created unconditionally. This CL changes it to create the tensors only when they are needed. Change: 117970158
-
由 A. Unique TensorFlower 提交于
Change: 117967381
-
由 A. Unique TensorFlower 提交于
Change: 117961829
-
由 A. Unique TensorFlower 提交于
and cuda library. Change: 117960844
-
由 Andrew Harp 提交于
Change: 117958944
-
由 A. Unique TensorFlower 提交于
Change: 117956472
-
由 A. Unique TensorFlower 提交于
Change: 117953780
-
由 David G. Andersen 提交于
storing it internally as an int64 (improve future compatibility with sampling from huge tensors. Note that this CL isn't enough to permit huge tensors, because we also clip their size at other points in the process, but it's a step forward.) Change: 117953355
-
由 Sherry Moore 提交于
DebugString() HasTensor() GetVariableToShapeMap() Example use can be found in PywrapTensorSliceReaderTest in third_party/tensorflow/python/training/saver_test.py For example, To use (in Python): try: reader = tf.train.NewCheckpointReader(your_checkpoint_file) print(reader.DebugString()) except pywrap_tensorflow.StatusNotOK as e: print(str(e)) Change: 117951901
-
由 A. Unique TensorFlower 提交于
Change: 117950844
-
由 Jianmin Chen 提交于
Change: 117950091
-