Sync replicas distributed training example with two strategies:
authorA. Unique TensorFlower <gardener@tensorflow.org>
Wed, 4 Apr 2018 12:31:48 +0000 (05:31 -0700)
committerTensorFlower Gardener <gardener@tensorflow.org>
Wed, 4 Apr 2018 12:34:52 +0000 (05:34 -0700)
commit71b19430e8484b136e0b872f6a543aff8a242587
treeb783147361aafd2a412d7c6b0c7ad1a578410e1e
parent1f5324ca69bc1017972eef8e418691cff9a86dd7
Sync replicas distributed training example with two strategies:
1) Interleave covariance and inverse update ops with training op.
2) Run the inverse and covariance ops on separate dedicated workers.

PiperOrigin-RevId: 191579634
tensorflow/contrib/kfac/examples/BUILD
tensorflow/contrib/kfac/examples/convnet.py
tensorflow/contrib/kfac/examples/convnet_mnist_distributed_main.py [new file with mode: 0644]
tensorflow/contrib/kfac/examples/convnet_mnist_multi_tower_main.py [new file with mode: 0644]
tensorflow/contrib/kfac/examples/convnet_mnist_single_main.py [moved from tensorflow/contrib/kfac/examples/convnet_mnist_main.py with 57% similarity]
tensorflow/contrib/kfac/examples/tests/convnet_test.py