- 17 3月, 2016 9 次提交
-
-
由 Vijay Vasudevan 提交于
Change: 117304114
-
由 Yutaka Leon 提交于
Add tf.contrib.lookup.string_to_index and tf.contrib.lookup.index_to_string to map strings to IDs and viceversa. Change: 117303981
-
由 Yutaka Leon 提交于
- Add a generic immutable hash table op that is initialized once and used to map key tensors to values. - Add a table initializer given the keys and values as tensors. Example use case: keys = tf.constant([0, 1], tf.int64) values = tf.constant(["hello", "world"]) default_value = "UNK" table = tf.contrib.lookup.HashTable(tf.contrib.lookup.KeyValueTensorInitializer(keys, values), default_value) input_tensor = tf.constant([0, 2], tf.int64) out = table.lookup(input_tensor). tf.initialze_all_tables().run() # or table.init.run() print out.eval() # Returns ["hello", "UNK"] Change: 117301920
-
由 Martin Wicke 提交于
Change: 117301677
-
由 Derek Murray 提交于
This test is flaky. We need a more robust solution for starting multiple processes and connecting them; `PickUnusedPortOrDie()` is not robust to multiple concurrent tests on the same host. A possible approach would be to add a portserver, or add a handshake between the parent test process and the server processes. Change: 117293607
-
由 A. Unique TensorFlower 提交于
Change: 117285646
-
由 Geoffrey Irving 提交于
Also fix MakePartialShape to correctly set is_unknown_ = false. Change: 117285169
-
由 David G. Andersen 提交于
advantage of additional test support for the use_locking path. END_PUBLIC Automated g4 rollback of changelist 117208315. *** Reason for rollback *** Fixed bug introduced in original CL *** Original change description *** Automated g4 rollback of changelist 117196108. BEGIN_PUBLIC Reverting use_locking change temporarily to fix test failure. Change: 117284547
-
由 A. Unique TensorFlower 提交于
Removes gradient error tests, and makes small change to create ranges with math_ops (and to avoid Python 3 errors). Change: 117284043
-
- 16 3月, 2016 26 次提交
-
-
由 A. Unique TensorFlower 提交于
Change: 117280722
-
由 A. Unique TensorFlower 提交于
Do input size inference based on inputs to LSTMCell, the same as for other cells. Leave the broken input_size @property in place as for the other cells. Change: 117276425
-
由 A. Unique TensorFlower 提交于
Change: 117273353
-
由 Eugene Brevdo 提交于
Also some updates to run_and_gather_logs. Reduce the set of tests that run_and_gather_logs_test runs to speed up the tests. Change: 117271064
-
由 Andrew Harp 提交于
Change: 117263085
-
由 Derek Murray 提交于
This change prevents feeding a tensor if its constant value has been accessed. For example, the constant value is often used in shape functions (e.g. in `tf.reshape(data, indices)`, `indices` is a tensor, but it is often constant) in order to infer more precise shapes. It is also used in the `tf.image.*` to generate simpler, more efficient graphs. However, doing so is potentially unsafe, because the tensor can be fed with a different value, which invalidates the previously obtained constant value, and can lead to subtle bugs. IF THIS BREAKS YOU ------------------ Replace the tensor that you are feeding with a `tf.placeholder()` of the appropriate dtype and shape. Change: 117263031
-
由 A. Unique TensorFlower 提交于
Change: 117260468
-
由 Manjunath Kudlur 提交于
- Return the result from GetOpList as uninterpreted bytes object. - Write a input typemap for GetPythonWrappers to receive python 'bytes' object and convert to const char* pointer and length. Change: 117258253
-
由 Geoffrey Irving 提交于
Also move some functions from header to C++. Change: 117255894
-
由 Eugene Brevdo 提交于
calling select() at each step. conditionals add a bunch of extra ops that slow things down; and generally the size of the input tensor into dynamic_rnn matches max_sequence_length so they provide no benefit. Before change, benchmarks: Calculation: Static Unroll with Dynamic Flow LSTM vs. Dynamic Unroll LSTM batch max_t units gpu dt(static) dt(dynamic) dt(dynamic)/dt(static) 256 50 512 False 1.795002 1.774248 0.988437 256 50 512 True 0.186828 0.200752 1.074525 256 50 256 False 0.597320 0.750226 1.255986 256 50 256 True 0.082047 0.091411 1.114130 256 50 128 False 0.250596 0.238233 0.950666 256 50 128 True 0.056480 0.063086 1.116960 After change, benchmarks: Calculation: Static Unroll with Dynamic Flow LSTM vs. Dynamic Unroll LSTM batch max_t units gpu dt(static) dt(dynamic) dt(dynamic)/dt(static) 256 50 512 False 1.723348 1.763019 1.023020 256 50 512 True 0.186794 0.196334 1.051072 256 50 256 False 0.644540 0.704506 1.093036 256 50 256 True 0.082274 0.087785 1.066985 256 50 128 False 0.241971 0.234559 0.969368 256 50 128 True 0.056356 0.059771 1.060611 Basically expect a more significant decrease in GPU step time when the matrices are smaller. Change: 117254684
-
由 Yutaka Leon 提交于
updates to the keys and values tensors. Change: 117252477
-
由 David G. Andersen 提交于
Change: 117250550
-
由 A. Unique TensorFlower 提交于
This makes it possible to apply regularizers like the Elastic Net, which incorporates separate L1 and L2 regularization components. Change: 117246793
-
由 Manjunath Kudlur 提交于
Change: 117245363
-
由 Dan Smilkov 提交于
Was able to reproduce the issue in python3. The problem comes down to 'str' and 'bytes' being equivalent types in python2, but in python3 they are not, since python3 encodes all strings as unicode. Because of this, they need to explicitly be converted to bytes using an encoding, which in this case is utf-8 since protobuf uses utf-8 for encoding strings. Change: 117244230
-
由 Derek Murray 提交于
Adds a typemap for the `const char* target` argument to `TF_SetTarget()` that accepts any object that may be converted to a null-terminated string. Change: 117241091
-
由 Martin Wicke 提交于
Change: 117209040
-
由 David G. Andersen 提交于
Change: 117208315
-
由 Josh Levenberg 提交于
private: tensorflow/core/platform/google/build_config:cuda_runtime_extra Change: 117197032
-
由 David G. Andersen 提交于
to the ref tensor itself. With this change, all locks are grabbed, and duplicates are explicitly skipped. Locks are acquired in address order to provide a lock discipline for avoiding deadlock. (Note that this is not guaranteed effective unless other multi-input code in TensorFlow follows it, something to add to the TODO list for the future. But the training ops are one of the primary locations where this pattern is used.) Change: 117196108
-
由 Yutaka Leon 提交于
updates to the keys tensor. Change: 117194227
-
由 Jianmin Chen 提交于
Change: 117191014
-
由 Manjunath Kudlur 提交于
build_config:cuda. This makes it possible to expand build_config:cuda to include other cuda dependencies not compatible with tf_custom_op_library. Change: 117186282
-
由 Eugene Brevdo 提交于
Change: 117185329
-
由 Derek Murray 提交于
Change: 117181098
-
由 A. Unique TensorFlower 提交于
is a SymbolicGradient op present. Fine-grained registration is not needed presently as the production inference models will not have gradients. Change: 117178512
-
- 15 3月, 2016 5 次提交
-
-
由 Zongheng Yang 提交于
Change: 117171668
-
由 Manjunath Kudlur 提交于
load_op_library. * In C++, check if the library is loaded, and if so, return an error status. * In Python, memoize library filename to module mapping. If library already loaded, return the memoized module. Fixes #1378 Change: 117171149
-
由 Eugene Brevdo 提交于
* includes a unit test that runs a benchmark and collects the results. * CC tests in tf/core/kernels now set linkstatic so GPU-based CC tests can be run by this test runner. * refactored the py test declarations in tensorflow.bzl to allow users to set main and args directly. necessary for the test runner. Change: 117170938
-
由 Sherry Moore 提交于
Change: 117170777
-
由 Geoffrey Irving 提交于
Also move some functions from header to C++. Change: 117167529
-