aboutsummaryrefslogtreecommitdiffhomepage
Commit message (Collapse)AuthorAge
...
* [tf.data] Fix broken implementation of `Dataset.from_generator()` on Windows.Gravatar Derek Murray2017-10-16
| | | | | | | | Due to a mix-up between NumPy's default array element type for a Python `int` on Windows and Linux, a tf.py_func() in `Dataset.from_generator()` would appear to return the wrong type on Windows (np.int32 instead of np.int64). All code using `Dataset.from_generator()` on Windows was previously broken. This change fixes both `tf.data.Dataset.from_generator()` and `tf.contrib.data.Dataset.from_generator()`. It also enables test coverage for this method on Windows, which should prevent future breakage. PiperOrigin-RevId: 172346533
* Fix xla_jit_compiled_cpu_function deps to pull in cpu_plugin.Gravatar A. Unique TensorFlower2017-10-16
| | | | | | | The intention was always for the user to only depend on xla_jit_compiled_cpu_function, and not need dependencies on internal targets. PiperOrigin-RevId: 172346257
* Added a cleaner mechanism to set the global constants in fisher_blocks.py ↵Gravatar A. Unique TensorFlower2017-10-16
| | | | | | | | and fisher_factors.py in the form of a function "set_global_constants". The old way of just manually setting these constants by importing the specific modules and accessing them directly should still work, but this new method is preferred. PiperOrigin-RevId: 172345996
* Proper use of convert_to_tensor in custom_gradientGravatar Alexandre Passos2017-10-16
| | | | PiperOrigin-RevId: 172342933
* Support a configurable TPU job nameGravatar Brennan Saeta2017-10-16
| | | | PiperOrigin-RevId: 172340173
* Support ClusterSpec propagation with XLA DevicesGravatar Brennan Saeta2017-10-16
| | | | | | Currently, you cannot use ClusterSpec propagation in conjunction with XLA devices, as the RenamedDevice wraps the underlying device and breaks the dynamic cast. PiperOrigin-RevId: 172339725
* Adds a host-memory GPU kernel for DestroyResourceOp.Gravatar Allen Lavoie2017-10-16
| | | | PiperOrigin-RevId: 172337312
* Automated g4 rollback of changelist 172039259Gravatar Allen Lavoie2017-10-16
| | | | PiperOrigin-RevId: 172336111
* [TF:XLA] Update xla_data comments for And, Or, and Not.Gravatar A. Unique TensorFlower2017-10-16
| | | | PiperOrigin-RevId: 172333451
* Fix typo (undefined variable `mean_absolute_error`, should refer to `error` ↵Gravatar A. Unique TensorFlower2017-10-16
| | | | | | previously defined). PiperOrigin-RevId: 172331504
* tfdbg doc: Fix minor typoGravatar Shanqing Cai2017-10-16
| | | | PiperOrigin-RevId: 172326303
* Automated g4 rollback of changelist 171877766Gravatar A. Unique TensorFlower2017-10-16
| | | | PiperOrigin-RevId: 172325692
* Fixing comment mismatch.Gravatar A. Unique TensorFlower2017-10-16
| | | | PiperOrigin-RevId: 172324333
* PiperOrigin-RevId: 172320984Gravatar A. Unique TensorFlower2017-10-16
|
* - Modified Jacobian computations in CurvatureMatrixVectorProductComputer to ↵Gravatar A. Unique TensorFlower2017-10-16
| | | | | | use true partial derivatives. This is done using the newly introduced stop_gradients argument to tf.gradients. PiperOrigin-RevId: 172315620
* Remove unused BUILD dependenciesGravatar A. Unique TensorFlower2017-10-16
| | | | PiperOrigin-RevId: 172314225
* Internal change.Gravatar Anna R2017-10-15
| | | | PiperOrigin-RevId: 172282778
* Replace NcclReduce/Broadcast ops during graph optimization so that we can ↵Gravatar A. Unique TensorFlower2017-10-15
| | | | | | | | generate gradients for Reduce/Broadcast. Changing _NcclBroadcastRecv shape input to int32 so that the corresponding Const op is outputting to HostMem. PiperOrigin-RevId: 172279684
* [XLA] Make pad shape inference error more informative.Gravatar Chris Leary2017-10-15
| | | | PiperOrigin-RevId: 172276292
* tfdbg: add persistent configGravatar Shanqing Cai2017-10-15
| | | | | | | | | | | | | * Add two persistent UI configurations backed by a file at ~/.tfdbg_config by default. * graph_recursion_depth, which controls the recursive output of li/lo commands. * mouse_mode, which controls the mouse state of the CursesUI. * Add `config` command to set and inspect the persistent configuration. E.g., * config show * config set graph_recursion_depth 3 * config set mouse_mode False Fixes: #13449 PiperOrigin-RevId: 172270804
* Add note pointing to master version of adding_an_op.Gravatar Mark Daoust2017-10-15
| | | | | | Fixes #13607 PiperOrigin-RevId: 172262174
* [XLA] Avoid unnecessary spaces in identifiers.Gravatar Chris Leary2017-10-14
| | | | PiperOrigin-RevId: 172224302
* Add streaming_false_{negative,positive}_rate and ↵Gravatar A. Unique TensorFlower2017-10-14
| | | | | | streaming_false_{negative,positive}_rate_at_thresholds. PiperOrigin-RevId: 172191462
* Fix case where broadcasting is not necessary.Gravatar Chris Ying2017-10-13
| | | | PiperOrigin-RevId: 172169909
* Optimized C++ and CUDA kernels for transposition.Gravatar A. Unique TensorFlower2017-10-13
| | | | | | | | * Shard fallback CPU implementation. * Optimize index calculations by trading 1 mod for 1 subtraction and 1 multiply (which have much lower combined latency). * Add optimized GPU kernels for on-the-fly conjugate transposition. PiperOrigin-RevId: 172167514
* Python wrapper to access the predicted peak memory usageGravatar Benoit Steiner2017-10-13
| | | | PiperOrigin-RevId: 172167437
* imperative_grad takes the tape instead of popping it.Gravatar Alexandre Passos2017-10-13
| | | | PiperOrigin-RevId: 172162006
* Make the HLO proto representation (hlo.proto) full fidelity. Hlo modules can ↵Gravatar Mark Heffernan2017-10-13
| | | | | | | | be serialized to HLO protos and deserialized without any information loss. As part of this change, a bug is fixed in NameUniquer. Previously, passing names with numeric suffixes could result in name collisions. PiperOrigin-RevId: 172161360
* Internal change.Gravatar Anna R2017-10-13
| | | | PiperOrigin-RevId: 172159815
* Enable Operation._add_control_inputs() with the C API and related improvementsGravatar Skye Wanderman-Milne2017-10-13
| | | | | | | | | | | | | | This change: - Implements the C API logic for Operation._add_control_inputs() - Adds type-checking to Operation._add_control_input() - Makes Graph::AddControlEdge() update the node def if necessary - Makes Graph::AddControlEdge() a no-op if the control edge already exists The AddControlEdge() changes may have a performance impact if anything is sensitive to AddControlEdge(), but nothing is to my knowledge. I'm not sure what benchmarks would confirm this. PiperOrigin-RevId: 172158589
* Fix crash if tf.nn.conv2d_backprop_filter or tf.nn.conv2d_backprop_input is ↵Gravatar Reed Wanderman-Milne2017-10-13
| | | | | | run with empty filter or input respectively. Resolves #13643. PiperOrigin-RevId: 172153646
* Ports the eager gradient tape to C.Gravatar Alexandre Passos2017-10-13
| | | | | | The tape stack is still in python as is the backprop code. PiperOrigin-RevId: 172151189
* Improve numerical accuracy of KL computationGravatar A. Unique TensorFlower2017-10-13
| | | | PiperOrigin-RevId: 172150350
* eager: Fix issue with custom_gradients and implicit_gradients.Gravatar Asim Shankar2017-10-13
| | | | | | While at it, clean up some dead code/comments in tape.py PiperOrigin-RevId: 172143125
* Define Evaluator class.Gravatar A. Unique TensorFlower2017-10-13
| | | | PiperOrigin-RevId: 172139804
* Automated g4 rollback of changelist 172050536Gravatar Russell Power2017-10-13
| | | | PiperOrigin-RevId: 172139466
* Automated g4 rollback of changelist 172018709Gravatar Andrew Harp2017-10-13
| | | | PiperOrigin-RevId: 172136820
* [XLA] DCE dead computations.Gravatar Justin Lebar2017-10-13
| | | | | | | We get a dead computation when e.g. we delete a reduction or remove a while loop. PiperOrigin-RevId: 172135511
* Internal change.Gravatar Anna R2017-10-13
| | | | PiperOrigin-RevId: 172134904
* [TF/XLA] Change GraphCompiler's to use a comparator based on node's name.Gravatar Yunxing Dai2017-10-13
| | | | | | | We realized that sorting the graph by id of a node is not always deterministic as ids themselves are randomly ordered by tensorflow. RELNOTES: n/a PiperOrigin-RevId: 172134671
* [Grappler] Remove reshapes whose source shape and destination shape are equal.Gravatar Jingyue Wu2017-10-13
| | | | | | | Also makes ArithmeticOptimizer::Optimize run shape inference at the beginning, and clear _output_shapes at the end. PiperOrigin-RevId: 172133948
* [TPU] Register gradient for CrossReplicaSum. Happily, the gradient for ↵Gravatar Peter Hawkins2017-10-13
| | | | | | CrossReplicaSum is just a CrossReplicaSum. PiperOrigin-RevId: 172132628
* Watch trainable variables in graph_callable decorated functionsGravatar Akshay Agrawal2017-10-13
| | | | PiperOrigin-RevId: 172131167
* Collapse backprop implementation of LSTMBlockCell into one CUDA kernel.Gravatar A. Unique TensorFlower2017-10-13
| | | | | | | | | | | Add benchmarks for backprop. Speed difference is minor - will need to move everything out of the graph for large speedups, I think. Also template the fprop kernel on use_peephole. Original change by @duckworthd PiperOrigin-RevId: 172131001
* Remove unnecessary reshape and get_shape from fused_batch_norm.Gravatar Chris Ying2017-10-13
| | | | PiperOrigin-RevId: 172130212
* Excluding test failing on windows with cmake.Gravatar Jiri Simsa2017-10-13
| | | | PiperOrigin-RevId: 172130104
* Automated g4 rollback of changelist 172041133Gravatar A. Unique TensorFlower2017-10-13
| | | | PiperOrigin-RevId: 172129075
* Fix build issue when tf/core:framework on an Android build.Gravatar Kiril Gorovoy2017-10-13
| | | | PiperOrigin-RevId: 172127789
* [XLA] The shape of the clone of a tuple should be the same as the shape of ↵Gravatar A. Unique TensorFlower2017-10-13
| | | | | | | | the original. This is true even if the layout of the tuple is weird - e.g. the subshapes of the output don't match the shape of the operands. PiperOrigin-RevId: 172124743
* Caching for variable scopes in LayersGravatar Allen Lavoie2017-10-13
| | | | PiperOrigin-RevId: 172122586
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-10 07:49:59 +0000