GitLab

PiperOrigin-RevId: 361957289

PiperOrigin-RevId: 361957289

PiperOrigin-RevId: 357974971

PiperOrigin-RevId: 357974971

PiperOrigin-RevId: 340580428

PiperOrigin-RevId: 340580428

PiperOrigin-RevId: 339071563

PiperOrigin-RevId: 339071563

Keras model serialization causes a lot of problems.

PiperOrigin-RevId: 328162551

Keras model serialization causes a lot of problems.

PiperOrigin-RevId: 328162551

PiperOrigin-RevId: 326286926

PiperOrigin-RevId: 326286926

PiperOrigin-RevId: 321817352

PiperOrigin-RevId: 321817352

PiperOrigin-RevId: 320124801

PiperOrigin-RevId: 320124801

This reverts commit 4bb13e61.

This reverts commit c3c2386c.

* root dir

* zone updated

* print mask

* preview emb

* tf print

* input only

* emb

* tf print

* emb after mask

* masked_softmax print

* print scores

* multi folder

* first pos emb

* check input shape

* add test temp

* import math

* two classes

* prints

* all get_pos replace

* make time scale private

* pos emb comments

* print input

* embedding_inputs

* tf shape

* dimention list

* tf_util

* print tf_util

* concise

* transformer pos change to layer

* keep length var

* length as input

* None as input

* print time signal

* print time signal

* remove print

* test input shape

* double check shape

* double check shape

* double check shape

* more test

* shape check

* shape check

* print 97 info

* print 97 info new

* test if sam

* assert same

* remove assert

* tf print same

* tf print diff

* output example

* output example

* output example

* formal test

* formal test length

* raise valurerror

* test valurerror

* double check

* comments

* remove prints

* rename relative

* delet naive test

* delete docs in xinliu branch

* code reformat

* import order

* indentation fix

* more files

* adjust char number

* disable not callable

* comment to length

* error of length unequal to input_shape

* root dir

* zone updated

* print mask

* preview emb

* tf print

* input only

* emb

* tf print

* emb after mask

* masked_softmax print

* print scores

* multi folder

* remove docs

* remove prints

* root dir

* zone updated

* print mask

* preview emb

* tf print

* input only

* emb

* tf print

* emb after mask

* masked_softmax print

* print scores

* multi folder

* remove docs

* apply revised 3 files

* rm prints

PiperOrigin-RevId: 314451720

* root dir

* zone updated

* print mask

* preview emb

* tf print

* input only

* emb

* tf print

* emb after mask

* masked_softmax print

* print scores

* multi folder

* first pos emb

* check input shape

* add test temp

* import math

* two classes

* prints

* all get_pos replace

* make time scale private

* pos emb comments

* print input

* embedding_inputs

* tf shape

* dimention list

* tf_util

* print tf_util

* concise

* transformer pos change to layer

* keep length var

* length as input

* None as input

* print time signal

* print time signal

* remove print

* test input shape

* double check shape

* double check shape

* double check shape

* more test

* shape check

* shape check

* print 97 info

* print 97 info new

* test if sam

* assert same

* remove assert

* tf print same

* tf print diff

* output example

* output example

* output example

* formal test

* formal test length

* raise valurerror

* test valurerror

* double check

* comments

* remove prints

* rename relative

* delet naive test

* delete docs in xinliu branch

* code reformat

* import order

* indentation fix

* more files

* adjust char number

* disable not callable

* comment to length

* error of length unequal to input_shape

PiperOrigin-RevId: 294997928

PiperOrigin-RevId: 286325224

PiperOrigin-RevId: 283266705

PiperOrigin-RevId: 281872406

Instead of needing to ensure variables are float32, casting inputs to float32, etc, instead dtype="float32" is passed to the layer constructor, which will do all that logic automatically.

The only difference is the output of LayerNorm is now float32 instead of float16, so an extra cast is needed elsewhere.

PiperOrigin-RevId: 273833286

PiperOrigin-RevId: 273371605

PiperOrigin-RevId: 267435985

PiperOrigin-RevId: 264853703

The old infer_float32_policies policy will be removed from TensorFlow soon.

Also, do Transformer inference in fp16, as well as training, when --dtype=fp16. In TF 2, layers now cannot run in multiple different dtypes, so we must use the same dtype for training and inference.

* trying fake merge call

* make metrics optional

* Remove extra print

* Fixes that make transformer run.

* Remove debug print statements.

* Changed the permissions to 644.

* Fix the rest of the permissions.

* enable static batch in all benchmarks

* Restrict dist strat hack to training mode

For now we will do predict/eval without dist strat, so remove that hack in non training cases.

* Use `inputs` instead of `x` as arg name for call

Keras has different behavior based on whether the inputs are called `inputs` or not. Using `inputs` gives expected behaviors.

* Avoid extra map fn on input in dist strat case

* Update how we handle custom metrics

This new approach works with and without dist strat. The previous one didn't work with dist strat. We need to fix that but this is reasonable in meantime (b/133724664).

* Update benchmarks

* typo in metrics code

* Revert metrics change

Didn't actually work in distributed case..

* Merged commit includes the following changes:
249776315  by tianlin<tianlin@google.com>:

    Internal change

249763206  by tianlin<tianlin@google.com>:

    For TF 2.0 (related to Beam Search), expand cond dims in tf.where(cond, x, y) to make all parameters broadcastable.

--
249392724  by hongkuny<hongkuny@google.com>:

    Internal change

PiperOrigin-RevId: 249776315

* Merged commit includes the following changes:
249823043  by tianlin<tianlin@google.com>:

    Bring back v2 test for predict and eval.

--

PiperOrigin-RevId: 249823043