GitLab

* set default steps to 300K.

* Log flags to perfzero.

* Add XLA support to transformer

- Moved config logic to keras_utils
- Added enable_xla flag to _performance flags
- Did not refactor enable_xla flag from keras resnet due to
  reliance on calling FLAGs in estimator keras and that is
  a needed refactor for another time.

* fix g3 lint complaint.

* Refactor set config into keras_utils.

* Move flags out of main.

* pipe through enable_xla

* Update official/transformer/v2/misc.py
Co-Authored-By: Reed <reedwm@google.com>

Before, there was a global default loss scale for all models. Currently, only resnet uses loss scaling, but this will be useful once more models support it.

* Support pure eager execution in ResNet50

* Use smaller batch size

* Add enable_get_next_as_optional flag.

* Set enable_get_next_as_optional to strategy.

* Add comments to explain the flag.

* Remove trailing whitespace.

* Remove trailing space.

* Added 'tfdata_exp' version of all benchmarks which set
FLAGS.tf_data_experimental_slack = True. Renamed
`data_prefetch_with_slack` to `data_delay_prefetch` (haoyu's change)
to make the names more distinct.

* Add flag to resnet input pipeline and surface through
keras_imagenet_main.py

* Fix trivial model to work properly with fp16

* Add comment on manual casting

* Do not report metrics in performance benchmarks

* Rename flag

* Add learning rate tensor. This makes training slower

* Improve LearningRateSchedule with better efficiency

* Fix lint error

* Replace constant definition with existing one

* Enable CuDNN BatchNorm spatial persistent by default; Remove 2nd zero padding layer

* Apply scale=False and fused=True consistently to BatchNorm layers

* Undo remove padding layer

* Replace zero padding with padding attribute in max pooling for better performance

* Resolve comments

* Revert "Replace zero padding with padding attribute in max pooling for better performance"

This reverts commit ad49db057c800ecac008eec1057005bd2c08ac73.

* Add benchmarks with the --cloning flag to Resnet and NFC.

* Renamed cloning to clone_model_in_keras_dist_strat. Dropped a few tests that aren't essential.

* Fixed up the formatting after re-naming the flag to a much longer  name.  Thanks, lint.
* Fixed the lint error in nfc_common.py

Reason: test failures because contrib is not available in V2

This reverts commit 325dd761.

* Introduce a short sleep before ds.prefetch in tf.data.

* Further limit dataset threads to reduce CPU contention

* Tuned dataset sleep time

* Rename dataset sleep flag; enable it only for Keras Graph mode

* Revert "Set input layer `batch_size` in multi-replica mode (#6578)"

This reverts commit f1a59682.

* Rename variables.

* Revert "Revert " Ensure static shapes when enabling XLA in Resnet Keras model (#6508)" (#6517)"

This reverts commit cc9eef76.

* Set `batch_size` to keras.Input in non-eager mode.

Eager mode currently has OOM problem.

* Add comments for enable_eager flag.

* Always set drop_remainder=True.

* Only set drop_remainder=True for XLA.

* add ds support for ncf

* remove comments for in_top_k

* avoid expanding the input layers

* resolve comments and fix lint

* Added some comments in code and fix lint

* fix lint

* add some documentation

* add tensorflow imports

* Add profiler callback for Keras models

* Update build stats to identify time callback by type

* Add warning message when both TensorBoard and profiler callbacks are used

Reason: break 1-gpu nightly test.

This reverts commit 371645fc.

Don't pass `batch_size` to keras.layers.Input in DS multi-replica case. There is currently a bug in Keras side which will cause a batch size incompatible error.

* Update resnet_model.py

* Ensure static shapes when enabling XLA.

* Define `drop_remainder` as a variable.

* Handles per_replica_batch_size in non-XLA mode

* Remove trailing whitespace.

required by multi-node collective ops in eager mode.

* Mixed precision support

* Add TODOs

* fix resnet breakage and add keras end-to-end tests

* delint

* address PR comments

* tmp commit

* tmp commit

* first attempt (without eval)

* Bug fixes

* bug fixes

* training done

* Loss NAN, no eval

* Loss weight problem solved

* resolve the NAN loss problem

* Problem solved. Clean up needed

* Added a todo

* Remove debug prints

* Extract get_optimizer to ncf_common

* Move metrics computation back to neumf; use DS.scope api

* Extract DS.scope code to utils

* lint fixes

* Move obtaining DS above producer.start to avoid race condition

* move pt 1

* move pt 2

* Update the run script

* Wrap keras_model related code into functions

* Update the doc for softmax_logitfy and change the method name

* Resolve PR comments

* working version with: eager, DS, batch and no masks

* Remove git conflict indicator

* move reshape to neumf_model

* working version, not converge

* converged

* fix a test

* more lint fix

* more lint fix

* more lint fixes

* more lint fix

* Removed unused imports

* fix test

* dummy commit for kicking of checks

* fix lint issue

* dummy input to kick off checks

* dummy input to kick off checks

* add collective to dist strat

* addressed review comments

* add a doc string

* Add flag to enable XLA in Keras models

* Fix lint errors (some of them are old errors)

* Add a flag to specify distribution strategies.

* Fix a small error.

* Address comments.

* Address comments.

* Fix typos.

- Modest speedup for CIFAR-10
- Slightly greater speedup expected for ImageNet ResNet50.