GitLab

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.

251762562  by hongkuny<hongkuny@google.com>:

    Fix blue score inconsistency

--

PiperOrigin-RevId: 251762562

* Modify tweaked tests for better performance in no cloning mode

* Tweak trivial models

251681245  by hongkuny<hongkuny@google.com>:

    Update bert to use the new tf.distribute APIs

--
251575972  by A. Unique TensorFlower<gardener@tensorflow.org>:

    Remove `steps_per_run` when instantiating TPUStrategy.

--

PiperOrigin-RevId: 251681245

* Add multi-worker benchmarks to official resnet estimator_benchmark.py.

* fix super constructor calls

* set datasets_num_private_threads to 32 in multi worker tweaked benchmarks

251325964  by hongkuny<hongkuny@google.com>:

    Improve flags

--
250942274  by tobyboyd<tobyboyd@google.com>:

    Internal change

PiperOrigin-RevId: 251325964

* Add CTL benchmark

* Divide train loss by number of train steps

* increase num epochs to 10

* add benchmark for early stopping with CTL

* remove whitespace

Because we run warmup tests in all real data benchmarks, XLA bugs will cause non-XLA tests to fail as well.

After the change (https://github.com/tensorflow/models/pull/6846/files#diff-965780bf33f2aeca41a33f8eba197c79) I receive the following error:

File "./models/official/mnist/mnist_tpu.py", line 202, in <module>
    absl_app.run()
TypeError: run() missing 1 required positional argument: 'main'

I added main as an argument and it seems to be working fine now.

* Add mlperf like test.

* Final comments.

* docstring wording tweak.

* non-tweaked version

* Add step 1 instructions for MaskGAN.

* Fix various lint errors

* Fix logging format