| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 208575390
|
|
|
|
|
|
| |
Fixes #16660
PiperOrigin-RevId: 208573227
|
|
|
|
| |
PiperOrigin-RevId: 208567907
|
|
|
|
|
|
|
|
|
|
| |
Removes a bit of boilerplate from training loops (making a prefix from a directory).
Also clarifies the recovery of checkpoint lists (like tf.train.Saver.recover_last_checkpoints, but automatic and more thorough). Adds a couple fields to the CheckpointState proto to support this.
Should live in contrib until I make it work well with tf.keras.Model.save_weights. When used together, save_weights needs to number its checkpoints. (There's a TODO for this.)
PiperOrigin-RevId: 208566198
|
|\
| |
| |
| | |
PiperOrigin-RevId: 208565050
|
| |
| |
| |
| | |
PiperOrigin-RevId: 208561390
|
| |
| |
| |
| | |
PiperOrigin-RevId: 208560644
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
- replacing `OpKernelContext` with newly introduced `DatasetContext` in `DatasetBase` constructor to make it possible to instantiate `DatasetBase` in places where an instance of `OpKernelContext` is not available
- replacing `dataset::MakeIteratorContext(OpKernelContext* ctx)` factory with `IteratorContext(OpKernelContext *ctx)` constructor.
- folding `GraphDatasetBase` into `DataseBase` and removing the default implementation of `AsGraphDefInternal`, making it the responsibility of the derived class to implement it to encourage/hint developers to provide serialization logic
PiperOrigin-RevId: 208560010
|
| |
| |
| |
| | |
PiperOrigin-RevId: 208558224
|
| |
| |
| |
| |
| |
| |
| |
| | |
Removes build-time compiler warnings due to overriding a member function
without an explicit `override` marker due to `-Winconsistent-missing-override`
directive.
PiperOrigin-RevId: 208556615
|
| |
| |
| |
| | |
PiperOrigin-RevId: 208556290
|
| |
| |
| |
| | |
PiperOrigin-RevId: 208548301
|
|\ \
| | |
| | |
| | | |
PiperOrigin-RevId: 208540342
|
| | |
| | |
| | |
| | | |
PiperOrigin-RevId: 208538037
|
|\ \ \
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 208532594
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
More of the shape inference can be done in C++-land, which may help grappler
do its thing. Also fix a bug where keys.dim_size(0) was being requested even
when keys.dims() == 0
[this should probably lead to DCHECK failure, but doesn't seem to].
PiperOrigin-RevId: 208529368
|
|\ \ \ \
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 208526581
|
|\ \ \ \ \
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 208525964
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 208519851
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Allow '/*...*/' and '//' comments to appear anywhere in HLO text, including multi-line comments. Previously only '/*...*/' comments were only allowed and only in certain locations in a serialized Literal.
PiperOrigin-RevId: 208519204
|
|\ \ \ \ \ \
| | | | | | |
| | | | | | |
| | | | | | | |
PiperOrigin-RevId: 208518776
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
- Make RunImpl private and expose it to the tests using a friend.
- Improve (my subjective opinion) the logging somewhat:
- Be briefer
- Don't print redundant information
- Print more detailed rationale for why a node wasn't clustered
PiperOrigin-RevId: 208518576
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
Also remove MobileNetV2 due to a compatibility issue (will be re-enabled in the next version).
PiperOrigin-RevId: 208517089
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
Fixes a potential data race in which the tensor backing a variable can be
read by one thread while another thread is writing to it.
PiperOrigin-RevId: 208516878
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
PiperOrigin-RevId: 208514584
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
domain.
PiperOrigin-RevId: 208513708
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
with fastmath.
- pow(x, -.5) => 1/sqrt(x)
- tanh(x) => fast llvm-ir implementation, instead of __nv_tanh from
libdevice.
Both of these should be safe. The main thing we're concerned about when
fastmath is disabled is getting proper NaN/Inf behavior, and these
changes don't affect that.
PiperOrigin-RevId: 208511562
|
|\ \ \ \ \ \ \
| | | | | | | |
| | | | | | | |
| | | | | | | | |
PiperOrigin-RevId: 208509699
|
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
During copy elision, copies which were previously added during copy insertion might end up with the same source and destination buffer. Previous logic did not remove them. This CL handles this corner case.
PiperOrigin-RevId: 208509171
|
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
PiperOrigin-RevId: 208508212
|
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
PiperOrigin-RevId: 208507857
|
|\ \ \ \ \ \ \ \
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208507804
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208505669
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208503071
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208502552
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
The motivation for this change is to allow us to change the default for CPU and
GPU independently. In particular, we want to disable fastmath on both
platforms, but I think it will be much easier on GPU, and I don't want that
blocked on getting things perfect on CPU.
The primary motivation for turning off fastmath is that on GPU with f16, model
authors may use automatic loss scaling, which looks for NaNs in the loss and
tweaks the loss scale factor. If fastmath is enabled, the compiler is free to
produce any fp value instead of NaN, so this is unsound.
The reason I expect this will be easier to do without regression on GPU is that
on CPU some of the important vectorization we do relies on fastmath being
enabled, but we have nothing of the sort on GPU.
No functional change, since this patch leaves the defaults as they currently
are (fastmath on for both platforms).
PiperOrigin-RevId: 208502191
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208497573
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208495688
|
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
PiperOrigin-RevId: 208495685
|
|\ \ \ \ \ \ \ \ \
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208493257
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208491690
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208491390
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
- Fixes breakage in tests using nnapi
PiperOrigin-RevId: 208485430
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208482434
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208481668
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
On Android stderr is not captured for applications, so developers do not see
the errors from tflite. This adds logcat output.
Output to stderr is kept, as it is convenient for unit tests run through the
shell.
PiperOrigin-RevId: 208463170
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
1) if either of the side is always zero, you never want to do the multiplication
2) because of the zero at the end, the if clause was never hit, not saving those flops.
PiperOrigin-RevId: 208457125
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208447033
|
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
PiperOrigin-RevId: 208446011
|
|\ \ \ \ \ \ \ \ \ \
| | | | | | | | | | |
| | | | | | | | | | |
| | | | | | | | | | | |
PiperOrigin-RevId: 208412584
|