| Commit message (Collapse) | Author | Age |
|
|
|
|
|
| |
- Acts as indicator for the TF runtime to make possible optimizations by treating the input tensor as a constant.
PiperOrigin-RevId: 177656212
|
|
|
|
| |
PiperOrigin-RevId: 177655994
|
|
|
|
| |
PiperOrigin-RevId: 177655417
|
|
|
|
|
|
|
|
| |
which could be used elsewhere in the graph; SliceProcessorConcatOffset is a
historical implementation anyway, and the same functionality could be
provided by more recently developed SliceProcessor and SliceProcessorConst.
PiperOrigin-RevId: 177653218
|
|
|
|
| |
PiperOrigin-RevId: 177652538
|
|
|
|
| |
PiperOrigin-RevId: 177651937
|
|
|
|
| |
PiperOrigin-RevId: 177647906
|
|
|
|
|
|
|
|
|
|
| |
Fix some inaccurate estimates exposed by this approach:
- propagate the inaccuracy flag when merging device stats;
- estimate Const as no-op;
- estimate RandomUniform, Relu and Softmax as element-wise;
- consider estimates accurate for known element-wise ops in op_level_cost_estimator.
PiperOrigin-RevId: 177643976
|
|
|
|
| |
PiperOrigin-RevId: 177640956
|
|
|
|
|
|
|
|
| |
allows for releasing the state associated with the handle.
Also simplifies the state owned by the FunctionLibraryRuntimeImpl. Instead of having a vector of ref counted Item objects and a separate vector of function bodies, we merge it into one object that holds the entire instantiated state for the function.
PiperOrigin-RevId: 177639560
|
|
|
|
| |
PiperOrigin-RevId: 177636427
|
|
|
|
| |
PiperOrigin-RevId: 177633993
|
|
|
|
| |
PiperOrigin-RevId: 177633858
|
|
|
|
|
| |
RELNOTES: N/A
PiperOrigin-RevId: 177631104
|
|
|
|
| |
PiperOrigin-RevId: 177629372
|
|
|
|
| |
PiperOrigin-RevId: 177625918
|
|
|
|
|
|
|
| |
This would help debug issues like this:
#14759
PiperOrigin-RevId: 177625164
|
|
|
|
|
|
| |
run environment in tf_op_stats.proto.
PiperOrigin-RevId: 177624829
|
|
|
|
| |
PiperOrigin-RevId: 177619402
|
|
|
|
|
|
|
|
|
| |
variables.
This additional parameter to computations was only needed for the CPU backend, but
it is not used anymore. The thread pool it holds is already passed via xla::ExecutableRunOptions. Also, remove unused members from XlaCompiler::CompilationResult.
PiperOrigin-RevId: 177618693
|
|
|
|
| |
PiperOrigin-RevId: 177616799
|
|
|
|
|
|
|
|
|
|
| |
This changes exposes a minor functional change: unknown attrs cause an
import error with the C API enabled, but only cause an error when the
graph is run in the original Python code (the error will occur even if
the problematic op isn't being run). It seems OK to fail faster in
error cases.
PiperOrigin-RevId: 177615841
|
|
|
|
|
|
| |
This change makes `__array__` conform to numpy's API.
PiperOrigin-RevId: 177614929
|
|
|
|
| |
PiperOrigin-RevId: 177614538
|
|
|
|
| |
PiperOrigin-RevId: 177612830
|
|
|
|
| |
PiperOrigin-RevId: 177611775
|
|
|
|
| |
PiperOrigin-RevId: 177611737
|
|
|
|
|
|
| |
that controls the type of pi-based damping adjustment to use. Current options are "tracenorm", which is what we were already doing and is the default, and "off".
PiperOrigin-RevId: 177610677
|
|
|
|
| |
PiperOrigin-RevId: 177606245
|
|
|
|
| |
PiperOrigin-RevId: 177605983
|
|
|
|
| |
PiperOrigin-RevId: 177596564
|
|
|
|
|
|
|
| |
This is a step towards supporting custom Dataset ops built as external
libraries.
PiperOrigin-RevId: 177595688
|
|
|
|
|
|
|
|
| |
(generated_input, generated_data) tuples in the pool.
2) Rename `tensor_pool` function to `random_tensor_pool` so the function name does not collide with the pkg name.
PiperOrigin-RevId: 177594443
|
|
|
|
|
|
| |
Use tf.contrib.summary.create_file_writer, instead.
PiperOrigin-RevId: 177588097
|
|
|
|
| |
PiperOrigin-RevId: 177559103
|
|
|
|
|
|
|
|
| |
`FullyConnectedSeriesFB` uses an approximation to the Fisher information
matrix designed for RNNs. This CL also adds support for dtypes
other than `float32` to `fisher_factors`.
PiperOrigin-RevId: 177558080
|
|
|
|
| |
PiperOrigin-RevId: 177555877
|
|
|
|
|
|
| |
by regarding several things as identifier.
PiperOrigin-RevId: 177548483
|
|
|
|
| |
PiperOrigin-RevId: 177545934
|
|
|
|
| |
PiperOrigin-RevId: 177545499
|
|
|
|
| |
PiperOrigin-RevId: 177543170
|
|
|
|
| |
PiperOrigin-RevId: 177540002
|
|
|
|
| |
PiperOrigin-RevId: 177537667
|
|
|
|
| |
PiperOrigin-RevId: 177535370
|
|
|
|
|
|
|
|
| |
int8, uint8, and bool, in addition to float ,double, int32, and int64, which were already supported.
Add unit test for all types.
PiperOrigin-RevId: 177533200
|
|
|
|
| |
PiperOrigin-RevId: 177526301
|
|
|
|
|
|
| |
backwards filter
PiperOrigin-RevId: 177522710
|
|
|
|
| |
PiperOrigin-RevId: 177519992
|
|
|
|
|
|
| |
Adds _cpu_jit to tf_cnn_benchmarks_xla BUILD rule and fixes an issue in XLA bridge triggered by XLA CPU compilation of whole graphs. In particular, modifies mark_for_compilation_pass.cc to skip _Retval nodes when looking for compilation candidates in the top level function. _Retval nodes are introduced in the input subgraph as a replacement for fetches. Including _Retval nodes into XLA clusters confuses encapsulate subgraph pass that expects a graph with no pre-existing _Retval nodes.
PiperOrigin-RevId: 177518178
|
|
|
|
| |
PiperOrigin-RevId: 177518046
|