aboutsummaryrefslogtreecommitdiffhomepage
Commit message (Collapse)AuthorAge
...
* Fix broken link in code example, pointed out by:Gravatar A. Unique TensorFlower2017-03-04
| | | | | #8041 Change: 149202172
* Fix code that ignores tensorflow::Status.Gravatar Justin Lebar2017-03-03
| | | | Change: 149170109
* Android: move makefile libtensorflow_inference.so build to ↵Gravatar Andrew Harp2017-03-03
| | | | | | | sub_makefile/android dir, and add libtensorflow_demo.so target to remove the necessity of using Bazel to build the demo. Partial solution to #8059, followup will add gradle/Android Studio integration. Change: 149167229
* Disable flaky pandas_io_testGravatar Gunhan Gulsoy2017-03-03
| | | | Change: 149165553
* Set the default flag for using nonfused winograd convolution in fowrad ↵Gravatar A. Unique TensorFlower2017-03-03
| | | | | | convolution to be false. Change: 149161107
* Mark Exporter and SessionBundle as deprecated. Add recommendation to switch toGravatar Sukriti Ramesh2017-03-03
| | | | | SavedModel. Change: 149158386
* Drop "x","y" args in favor of "input_fn",Gravatar A. Unique TensorFlower2017-03-03
| | | | | | | Also Add a data download step. Will need to sync with "get_started/monitors.md" Change: 149158319
* Make bidirectional_dynamic_rnn sequence_length argument optional (Fixes #5588).Gravatar Eugene Brevdo2017-03-03
| | | | | | If the argument is not provided, the time reversal is applied to all batch entries the same: time is reversed from 0 to max_time for each entry. Change: 149155863
* Clipping into [0.0, 1.0] before converting back to original data type in ↵Gravatar A. Unique TensorFlower2017-03-03
| | | | | | 'adjust_brightness'. The current implementation (i.e. without clipping before conversion) introduces different behavior for images with different original data types, i.e. uint8 or float32. When converting back to the original data type, if the original type is uint8, there is a automatic clipping effect, since all underflow/overflow numbers will be constrained to [0, 255]. But if the original data type is float32, no clipping will happen. Change: 149155199
* Add a script to run xla tests.Gravatar Gunhan Gulsoy2017-03-03
| | | | Change: 149152982
* Add the graphdef version to InferenceContext and to ShapeRefiner::AddNode.Gravatar A. Unique TensorFlower2017-03-03
| | | | | | | Use this to allow loading reductions saved with older graphdefs. Change GraphConstructor to not increase the version when importing, but instead take the min of all versions. Change: 149152437
* [TF:XLA] Clarify order of operations for ReduceWindow and SelectAndScatter.Gravatar Peter Hawkins2017-03-03
| | | | Change: 149150316
* [TF:XLA] Implement SparseSoftmaxCrossEntropyWithLogits.Gravatar Peter Hawkins2017-03-03
| | | | Change: 149149782
* Fix the error message returned from _init_from_args function in QueueRunner ↵Gravatar A. Unique TensorFlower2017-03-03
| | | | | | to state that the input argument queue_closed_exception_types, if passed, should be of type 'tuple'. Change: 149149270
* In ShapeRefiner::ExtractConstantSubgraph, don't includeGravatar A. Unique TensorFlower2017-03-03
| | | | | | subgraphs that use a Merge node. Back edges for merges are not on the graph when it is imported by GraphConstructor. Change: 149145899
* Factor-out dryrun functionalities from graph_transfer_utils to ↵Gravatar A. Unique TensorFlower2017-03-03
| | | | | | remote_fused_graph_execute_op in order to cache shapes in RemtoeFusedGraphExecuteInfo Change: 149143066
* Restore the adding_an_op code examples that used to live underGravatar A. Unique TensorFlower2017-03-03
| | | | | g3doc/, now under examples/. Partial fix of #8029. Change: 149142119
* Fix python paths in windows cmake build.Gravatar A. Unique TensorFlower2017-03-03
| | | | Change: 149132162
* Increase the number of retries to 10.Gravatar Alexey Surkov2017-03-03
| | | | | Also makes file upload retries use the common retry logic. Change: 149131497
* Doc generator refactoring.Gravatar A. Unique TensorFlower2017-03-03
| | | | Change: 149127634
* Does pandas import check in each caller file directly to avoid flaky tests.Gravatar Jianwei Xie2017-03-03
| | | | Change: 149118694
* Update build-file to expose framework/types.proto temporarily.Gravatar A. Unique TensorFlower2017-03-03
| | | | Change: 149108576
* Create contrib folder for new boosted_trees library.Gravatar A. Unique TensorFlower2017-03-03
| | | | Change: 149105174
* Fully qualify symbols to prevent clashes in directories named 'tensorflow'.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149085750
* Interal change.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149084406
* Moves build.gradle to use gradle plugin 2.2Gravatar Mark McDonald2017-03-02
| | | | Change: 149075765
* Avoid creating a tensorflow::Status::OK() object and destroying it onGravatar Jeffrey A. Dean2017-03-02
| | | | | | | | | | | | | | | | every call to TF_CHECK_OK(). Speeds up a microbenchmark that is added in this change to status_test.cc from 1.19 ns per TF_CHECK_OP (before the changes to status.{h,cc}) to 0.587ns per TF_CHECK_OK (51% improvement). Since Status::operator== method generates quite a lot of code, and we now avoid calling that, code size is considerably smaller by about 352 bytes per TF_CHECK_OK. The size of BM_TF_CHECK_OK routine in status_test.cc is reduced from 699 bytes to 347 bytes, as measured by nm --print-size --radix=d ...status_test_binary... | grep BM_TF Change: 149073899
* Android: add support for building libtensorflow_inference.so with make. This ↵Gravatar Andrew Harp2017-03-02
| | | | | | | will enable Android Studio builds of the demo on Windows machines in follow-up changes. Note that the resulting .so file is currently armeabi-v7a only and 43mb, which will also be optimized in later CLs. Change: 149073764
* Fix the cmake failure: ambiguous symbolGravatar Yuefeng Zhou2017-03-02
| | | | Change: 149072356
* [XLA] [StreamExecutor] Tune GEMMs when possible.Gravatar Justin Lebar2017-03-02
| | | | | | | | | | | | | cublas 8 adds the cublasGemmEx function, which lets you specify an explicit "algorithm" for the computation. This functions as an opaque tuning hint to cublas. This patch adds support for cublasGemmEx to StreamExecutor, and wires up XLA's GemmThunk to use the new function. This patch does not add GEMM autotuning support in TensorFlow proper, only XLA. Change: 149068961
* [StreamExecutor] Minor comment cleanups.Gravatar Justin Lebar2017-03-02
| | | | Change: 149066697
* [XLA:GPU] Cache GPU substreams across executionsGravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149063035
* Upon any HTTP error, clear the response buffer.Gravatar Jonathan Hseu2017-03-02
| | | | Change: 149062390
* Add tfcompile target triple for android_x86.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149060929
* Error out if a subclass of Estimator overrides a member of Estimator.Gravatar Mustafa Ispir2017-03-02
| | | | Change: 149060568
* Ensure that QuantizeAndDequantizeV2Op doesn't alter modify input tensors ↵Gravatar Suharsh Sivakumar2017-03-02
| | | | | | when range_given is false. Change: 149059522
* Fix bug in matmul gradients for complex types.Gravatar A. Unique TensorFlower2017-03-02
| | | | | Clean up matmul gradient test code and improve test coverage. Change: 149057186
* Added interfaces for the messages in contrib/../graph_explorer.proto.Gravatar Vijay D'Silva2017-03-02
| | | | Change: 149051396
* Fix typos and inconsistencies in layers docstrings.Gravatar Francois Chollet2017-03-02
| | | | Change: 149050553
* [TF:XLA] Reduce sequential memory usage via better ordering and simulated heap.Gravatar A. Unique TensorFlower2017-03-02
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The choice of instruction ordering, and the minimization of fragmentation once we've chosen an order, are two large inter-related factors wrt overall memory usage. The approach in this CL uses heuristics to do better on both, but neither problem is completely solved. To pick a better an ordering (the larger factor), the approach is to try the original list-scheduler based ordering, and to also try a DFS based ordering. We pick the ordering that yields a smaller minimum memory, computed with the simulated heap, ignoring fragmentation. Note that this is the absolute minimum memory for a given ordering. To minimize fragmentation, the approach is to run a heap simulation on temporary buffers. We still try to re-use existing allocations when possible, but instead of creating new allocations for temp buffers, we collect all the leftovers and use a heap to pack them. The heap algorithm that gave the best results is "lazy best-fit"; a variant of traditional best-fit that sometimes delays offset assignment until Free is called, in the hopes of yielding larger free chunks. Here's some measurements of the temp buffer sizes for GNMT encoder training (a stacked LSTM). Lower is better. I've tried various combinations of instruction ordering and heap simulation, to show the joint impact of these two factors. List-scheduler order, no heap simulation 33.33GiB List-scheduler order, with heap simulation 25.09GiB Minimized DFS order, no heap simulation 16.59GiB Arbitrary DFS order, no heap simulation 15.05GiB (old) Arbitrary DFS order, with heap simulation 12.57GiB Minimized DFS order, with heap simulation 11.71GiB (new) Note that the original list scheduler order is much worse than DFS on stacked LSTMs, but (not shown here) is much better than DFS on convolutions like Inception. Also note that heap simulation packs things tighter for all instruction orders in this example, but to varying degrees. Change: 149049028
* Tracking memory usage of queues.Gravatar Yuefeng Zhou2017-03-02
| | | | Change: 149048750
* Improve speed of depthwise conv backward on GPU.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149047908
* Implement sharding by input row for depthwiseconv2d, allowing multithreading ↵Gravatar Andrew Harp2017-03-02
| | | | | | when batch size is 1. This resulted in reducing the wall time spent in the op by 42% during testing on a Pixel phone. Change: 149047766
* Add BenchmarkType to TestResults proto and fix logging of C++ microbenchmarks.Gravatar Eugene Brevdo2017-03-02
| | | | | C++ microbenchmarks results are now normalized by iters (like python benchmarks). Change: 149045367
* Print whether a node is_dead in the executor log statement.Gravatar Suharsh Sivakumar2017-03-02
| | | | Change: 149043862
* Go: Update generated wrapper functions for TensorFlow ops.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149040924
* Update ops-related pbtxt files.Gravatar A. Unique TensorFlower2017-03-02
| | | | Change: 149040750
* When a list of strings is passed to one of the read_batch_* methods, allow eachGravatar A. Unique TensorFlower2017-03-02
| | | | | | element to contain wildcards. Before we did wildcard expansion only if a single element is given. Change: 149039100
* Retrying FileExists errors in case of errors to overcome intermittent GCS ↵Gravatar A. Unique TensorFlower2017-03-02
| | | | | | errors. Change: 149038920
* Tracking lookup table memory usage.Gravatar Yuefeng Zhou2017-03-02
| | | | Change: 149036604
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-01 03:44:13 +0000