Move sorting of variables from the implicit_grad family of functions
authorA. Unique TensorFlower <gardener@tensorflow.org>
Wed, 21 Feb 2018 23:54:54 +0000 (15:54 -0800)
committerTensorFlower Gardener <gardener@tensorflow.org>
Wed, 21 Feb 2018 23:58:49 +0000 (15:58 -0800)
commit2e22c7c9b02d9153aac12c0483d6baeb0cada318
treebe871a88954a00ea3b5703081954e89418fb67fd
parent6907c18e8f4e9318f0546f2558992711fcfc02da
Move sorting of variables from the implicit_grad family of functions
up to GradientTape.watched_variables() so we also get deterministic
behavior when using the GradientTape as is done in
Optimizer.compute_gradients().

PiperOrigin-RevId: 186533323
tensorflow/python/eager/backprop.py