| Commit message (Collapse) | Author | Age |
... | |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
by extending the core/platform API with some basic functionality.
The new functions allow:
1. Determining how many NUMA nodes are available.
2. Setting the executing thread to be bound to a particular node,
or not bound at all.
3. Allocating memory affiliated with a particular node.
This change introduces the API only, there is not yet a real
implementation.
PiperOrigin-RevId: 202724194
|
|
|
|
| |
PiperOrigin-RevId: 202724146
|
|
|
|
| |
PiperOrigin-RevId: 202724096
|
|
|
|
| |
PiperOrigin-RevId: 202720777
|
|
|
|
|
|
| |
Return the first unsupported op with detailed error message instead of all unsupported ops without details.
PiperOrigin-RevId: 202720375
|
|
|
|
|
|
| |
back.
PiperOrigin-RevId: 202719394
|
|
|
|
| |
PiperOrigin-RevId: 202716942
|
|
|
|
|
|
| |
StatusOr<TensorShape>.
PiperOrigin-RevId: 202711909
|
|
|
|
|
|
| |
tf.contrib.data.bucket_by_sequence_length.
PiperOrigin-RevId: 202711095
|
|
|
|
| |
PiperOrigin-RevId: 202706517
|
|
|
|
|
|
|
|
| |
The overall approach is to teach the gradients code how to traverse
the implicit edges between captured external tensors and ops inside
the function body.
PiperOrigin-RevId: 202705929
|
|
|
|
| |
PiperOrigin-RevId: 202705179
|
|
|
|
| |
PiperOrigin-RevId: 202703970
|
|
|
|
|
|
|
|
|
|
|
|
| |
Conceptually lists just get replaced with a list-like wrapper. A shallow copy is maintained for error checking (since appends to it aren't monitored, we can't do restore-on-create for variables unless it's being modified through the wrapper).
There are lots of other details. I gave up on generalizing our isinstance(obj, (list, tuple)) checks and just subclassed list. Behaving like a list means the type should be unhashable, which requires some workarounds when we're collecting objects (object-identity collections, and object-identity versions of weak reference containers).
Adds a decorator for exempting whole methods from automatic dependency tracking so we don't need to track down every last self.inputs = [] statement to avoid polluting dependencies.
There's a TODO for tuples and dictionaries.
PiperOrigin-RevId: 202703271
|
|
|
|
| |
PiperOrigin-RevId: 202701234
|
|
|
|
| |
PiperOrigin-RevId: 202698606
|
|
|
|
| |
PiperOrigin-RevId: 202698287
|
|
|
|
| |
PiperOrigin-RevId: 202696277
|
|
|
|
|
|
|
| |
Keras test_function, and to write histogram outputs with a batch-level
global step.
PiperOrigin-RevId: 202696047
|
|
|
|
| |
PiperOrigin-RevId: 202695310
|
|
|
|
|
|
|
|
| |
documentation.
Fixes #20265.
PiperOrigin-RevId: 202695249
|
|
|
|
| |
PiperOrigin-RevId: 202693036
|
|
|
|
|
|
| |
batches correctly.
PiperOrigin-RevId: 202688283
|
|
|
|
|
|
|
|
|
| |
This is at the least useful for testing behavioral differences between
a wrapped Python function and the corresponding graph functions. Prior to this change,
decorating a Python function with `@function.defun` would render the Python function
inaccessible.
PiperOrigin-RevId: 202685407
|
|
|
|
|
|
| |
version.
PiperOrigin-RevId: 202683951
|
|
|
|
| |
PiperOrigin-RevId: 202682712
|
|
|
|
|
|
| |
Use Sort to implement R1 TopK for an arbitrary dimension size, and more types.
PiperOrigin-RevId: 202681175
|
|
|
|
| |
PiperOrigin-RevId: 202681043
|
|
|
|
| |
PiperOrigin-RevId: 202679902
|
|
|
|
| |
PiperOrigin-RevId: 202673820
|
|
|
|
|
|
| |
This is only currently implemented in the evaluator backend, and even that implementation is partial - the key and value type must match.
PiperOrigin-RevId: 202673122
|
|
|
|
| |
PiperOrigin-RevId: 202671299
|
|
|
|
| |
PiperOrigin-RevId: 202670638
|
|
|
|
|
|
| |
The most expensive part of this kernel is the index construction. The optimized implementation builds the new index matrix at most once, rather than performing up to 3 passes (adding a leading dimension, `SparseTensor::Concat()` and `Reshape()`), and adds a specialized codepath for the common case of stacking together rank-1 SparseTensors.
PiperOrigin-RevId: 202669432
|
|
|
|
| |
PiperOrigin-RevId: 202668511
|
|
|
|
| |
PiperOrigin-RevId: 202668227
|
|
|
|
|
|
|
|
| |
`stats_aggreagtor`.
Also collects metrics for `examples_count`, `features_count`, `feature_values_count`, `feature_lists_count` and `sequence_examples_count` when `feature_stats()` transformation is applied to the dataset.
PiperOrigin-RevId: 202667632
|
|
|
|
| |
PiperOrigin-RevId: 202667025
|
|
|
|
|
|
|
| |
This change allows TensorFlow users to stream data directly from
Cloud Bigtable into the TensorFlow runtime using tf.data.
PiperOrigin-RevId: 202664219
|
|
|
|
| |
PiperOrigin-RevId: 202663814
|
|
|
|
|
|
|
|
|
| |
separate from the dataflow analysis. Extend it to cover return statements, nested functions and finally blocks.
Note: AutoGraph doesn't support exceptions and will reject try/finally constructs, but they were easy enough to add.
This is not used yet.
PiperOrigin-RevId: 202661509
|
|
|
|
|
|
|
|
|
|
| |
their dependencies. Use the monadic structure of XlaOp instead. Remove XlaBuilder* arguments to many utility functions.
Various small cleanups. Rename PrependMajorDims to ConcatVectors to better reflect what it does.
No functional changes intended.
PiperOrigin-RevId: 202655690
|
|
|
|
| |
PiperOrigin-RevId: 202645535
|
|
|
|
|
|
|
|
| |
This became necessary when the TOKEN primitive type was added.
In some models, an existing tuple T is extended to (T, token[]).
Also add the TOKEN case to a switch statement where it was missing.
PiperOrigin-RevId: 202643759
|
|
|
|
|
|
|
|
| |
Start a new client library "numeric", after the C++ <numeric> header where std::iota lives.
[TF:XLA] Replace uses of XlaHelpers::Iota() with xla::Iota(). Add a helper to get the XLA type of an operator input.
PiperOrigin-RevId: 202636221
|
|
|
|
|
|
| |
users to use the free functions (or the operator overloads) in namespace xla:: instead.
PiperOrigin-RevId: 202631789
|
|
|
|
| |
PiperOrigin-RevId: 202624150
|
|
|
|
|
|
| |
maintaining forward compatibility of Python API calls.
PiperOrigin-RevId: 202618021
|
|
|
|
| |
PiperOrigin-RevId: 202613754
|
|
|
|
| |
PiperOrigin-RevId: 202598404
|