 View on TensorFlow.org
|
 Run in Google Colab
|
 Download notebook
|
 View source on GitHub
|
This tutorial trains a TensorFlow model to classify the MNIST dataset, where the training function is compiled using XLA.
First, load TensorFlow and enable eager execution.
import tensorflow as tf
tf.compat.v1.enable_eager_execution()
Then define some necessary constants and prepare the MNIST dataset.
# Size of each input image, 28 x 28 pixels
IMAGE_SIZE = 28 * 28
# Number of distinct number labels, [0..9]
NUM_CLASSES = 10
# Number of examples in each training batch (step)
TRAIN_BATCH_SIZE = 100
# Number of training steps to run
TRAIN_STEPS = 1000
# Loads MNIST dataset.
train, test = tf.keras.datasets.mnist.load_data()
train_ds = tf.data.Dataset.from_tensor_slices(train).batch(TRAIN_BATCH_SIZE).repeat()
# Casting from raw data to the required datatypes.
def cast(images, labels):
images = tf.cast(
tf.reshape(images, [-1, IMAGE_SIZE]), tf.float32)
labels = tf.cast(labels, tf.int64)
return (images, labels)
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz 11493376/11490434 [==============================] - 0s 0us/step 11501568/11490434 [==============================] - 0s 0us/step
Finally, define the model and the optimizer. The model uses a single dense layer.
layer = tf.keras.layers.Dense(NUM_CLASSES)
optimizer = tf.keras.optimizers.Adam()
Define the training function
In the training function, you get the predicted labels using the layer defined above, and then minimize the gradient of the loss using the optimizer. In order to compile the computation using XLA, place it inside tf.function with jit_compile=True.
@tf.function(jit_compile=True)
def train_mnist(images, labels):
images, labels = cast(images, labels)
with tf.GradientTape() as tape:
predicted_labels = layer(images)
loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=predicted_labels, labels=labels
))
layer_variables = layer.trainable_variables
grads = tape.gradient(loss, layer_variables)
optimizer.apply_gradients(zip(grads, layer_variables))
Train and test the model
Once you have defined the training function, define the model.
for images, labels in train_ds:
if optimizer.iterations > TRAIN_STEPS:
break
train_mnist(images, labels)
And, finally, check the accuracy:
images, labels = cast(test[0], test[1])
predicted_labels = layer(images)
correct_prediction = tf.equal(tf.argmax(predicted_labels, 1), labels)
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print("Prediction accuracy after training: %s" % accuracy)
Prediction accuracy after training: tf.Tensor(0.8721, shape=(), dtype=float32)
Behind the scenes, the XLA compiler has compiled the entire TF function to HLO, which has enabled fusion optimizations. Using the introspection facilities, we can see the HLO code (other interesting possible values for "stage" are optimized_hlo for HLO after optimizations and optimized_hlo_dot for a Graphviz graph):
print(train_mnist.experimental_get_compiler_ir(images, labels)(stage='hlo'))
HloModule a_inference_train_mnist_5284__.204, input_output_alias={ {0}: (2, {}, may-alias), {1}: (3, {}, may-alias), {2}: (5, {}, may-alias), {3}: (8, {}, may-alias), {4}: (9, {}, may-alias), {5}: (10, {}, may-alias), {6}: (11, {}, may-alias) }
%max_float_.65 (x.66: f32[], y.67: f32[]) -> f32[] {
%x.66 = f32[] parameter(0)
%y.67 = f32[] parameter(1)
ROOT %maximum.68 = f32[] maximum(f32[] %x.66, f32[] %y.67)
}
%add_float_.75 (x.76: f32[], y.77: f32[]) -> f32[] {
%x.76 = f32[] parameter(0)
%y.77 = f32[] parameter(1)
ROOT %add.78 = f32[] add(f32[] %x.76, f32[] %y.77)
}
%add_float_.94 (x.95: f32[], y.96: f32[]) -> f32[] {
%x.95 = f32[] parameter(0)
%y.96 = f32[] parameter(1)
ROOT %add.97 = f32[] add(f32[] %x.95, f32[] %y.96)
}
%Mean-reduction.106 (x.107: f32[], y.108: f32[]) -> f32[] {
%x.107 = f32[] parameter(0)
%y.108 = f32[] parameter(1)
ROOT %add.109 = f32[] add(f32[] %x.107, f32[] %y.108)
}
%region_0.123 (Arg_0.124: f32[], Arg_1.125: f32[]) -> f32[] {
%Arg_0.124 = f32[] parameter(0)
%Arg_1.125 = f32[] parameter(1)
ROOT %add.126 = f32[] add(f32[] %Arg_0.124, f32[] %Arg_1.125), metadata={op_name="gradient_tape/dense/BiasAdd/BiasAddGrad"}
}
ENTRY %a_inference_train_mnist_5284__.204 (arg0.1: f32[10000,784], arg1.2: s64[10000], arg2.3: f32[784,10], arg3.4: f32[10], arg4.5: f32[], arg5.6: s64[], arg6.7: f32[], arg7.8: f32[], arg8.9: f32[784,10], arg9.10: f32[784,10], arg10.11: f32[10], arg11.12: f32[10]) -> (f32[784,10], f32[10], s64[], f32[784,10], f32[784,10], /*index=5*/f32[10], f32[10]) {
%constant.15 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_2" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=141}
%constant.16 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_2" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=141}
%arg6.7 = f32[] parameter(6), parameter_replication={false}, metadata={op_name="XLA_Args"}
%subtract.17 = f32[] subtract(f32[] %constant.16, f32[] %arg6.7), metadata={op_type="Sub" op_name="Adam/sub_2" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=141}
%constant.18 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_3" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=144}
%constant.19 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_3" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=144}
%arg7.8 = f32[] parameter(7), parameter_replication={false}, metadata={op_name="XLA_Args"}
%subtract.20 = f32[] subtract(f32[] %constant.19, f32[] %arg7.8), metadata={op_type="Sub" op_name="Adam/sub_3" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=144}
%constant.21 = s64[] constant(1), metadata={op_type="AddV2" op_name="Adam/add" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=127}
%constant.26 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_1" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%constant.30 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%arg4.5 = f32[] parameter(4), parameter_replication={false}, metadata={op_name="XLA_Args"}
%constant.31 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%arg5.6 = s64[] parameter(5), parameter_replication={false}, metadata={op_name="XLA_Args"}
%constant.22 = s64[] constant(1), metadata={op_type="AddV2" op_name="Adam/add" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=127}
%add.23 = s64[] add(s64[] %arg5.6, s64[] %constant.22), metadata={op_type="AddV2" op_name="Adam/add" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=127}
%convert.24 = f32[] convert(s64[] %add.23), metadata={op_type="Cast" op_name="Adam/Cast_1" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=127}
%power.29 = f32[] power(f32[] %arg7.8, f32[] %convert.24), metadata={op_type="Pow" op_name="Adam/Pow_1" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=131}
%subtract.32 = f32[] subtract(f32[] %constant.31, f32[] %power.29), metadata={op_type="Sub" op_name="Adam/sub" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%sqrt.33 = f32[] sqrt(f32[] %subtract.32), metadata={op_name="Adam/Sqrt"}
%constant.27 = f32[] constant(1), metadata={op_type="Sub" op_name="Adam/sub_1" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%power.25 = f32[] power(f32[] %arg6.7, f32[] %convert.24), metadata={op_type="Pow" op_name="Adam/Pow" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=130}
%subtract.28 = f32[] subtract(f32[] %constant.27, f32[] %power.25), metadata={op_type="Sub" op_name="Adam/sub_1" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%divide.34 = f32[] divide(f32[] %sqrt.33, f32[] %subtract.28), metadata={op_type="RealDiv" op_name="Adam/truediv" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%multiply.35 = f32[] multiply(f32[] %arg4.5, f32[] %divide.34), metadata={op_type="Mul" op_name="Adam/mul" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=133}
%constant.36 = s32[2]{0} constant({-1, 784}), metadata={op_type="Reshape" op_name="Reshape" source_file="/tmp/ipykernel_31814/494562224.py" source_line=16}
%arg1.2 = s64[10000]{0} parameter(1), parameter_replication={false}, metadata={op_name="XLA_Args"}
%reshape.14 = s64[10000]{0} reshape(s64[10000]{0} %arg1.2)
%broadcast.47 = s64[10000,10]{1,0} broadcast(s64[10000]{0} %reshape.14), dimensions={0}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%iota.46 = s64[10000,10]{1,0} iota(), iota_dimension=1, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%compare.48 = pred[10000,10]{1,0} compare(s64[10000,10]{1,0} %broadcast.47, s64[10000,10]{1,0} %iota.46), direction=EQ, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.43 = f32[] constant(1), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.45 = f32[10000,10]{1,0} broadcast(f32[] %constant.43), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.42 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.44 = f32[10000,10]{1,0} broadcast(f32[] %constant.42), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%select.49 = f32[10000,10]{1,0} select(pred[10000,10]{1,0} %compare.48, f32[10000,10]{1,0} %broadcast.45, f32[10000,10]{1,0} %broadcast.44), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.57 = s64[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.58 = s64[10000]{0} broadcast(s64[] %constant.57), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%compare.59 = pred[10000]{0} compare(s64[10000]{0} %broadcast.58, s64[10000]{0} %reshape.14), direction=LE, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.54 = s64[] constant(10), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.55 = s64[10000]{0} broadcast(s64[] %constant.54), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%compare.56 = pred[10000]{0} compare(s64[10000]{0} %reshape.14, s64[10000]{0} %broadcast.55), direction=LT, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%and.60 = pred[10000]{0} and(pred[10000]{0} %compare.59, pred[10000]{0} %compare.56), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.52 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.53 = f32[10000]{0} broadcast(f32[] %constant.52), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.50 = f32[] constant(nan), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.51 = f32[10000]{0} broadcast(f32[] %constant.50), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%select.61 = f32[10000]{0} select(pred[10000]{0} %and.60, f32[10000]{0} %broadcast.53, f32[10000]{0} %broadcast.51), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.62 = f32[10000,10]{1,0} broadcast(f32[10000]{0} %select.61), dimensions={0}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%add.63 = f32[10000,10]{1,0} add(f32[10000,10]{1,0} %select.49, f32[10000,10]{1,0} %broadcast.62), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%negate.90 = f32[10000,10]{1,0} negate(f32[10000,10]{1,0} %add.63), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.86 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.87 = f32[10000,10]{1,0} broadcast(f32[] %constant.86), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%compare.88 = pred[10000,10]{1,0} compare(f32[10000,10]{1,0} %add.63, f32[10000,10]{1,0} %broadcast.87), direction=EQ, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.84 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.85 = f32[10000,10]{1,0} broadcast(f32[] %constant.84), dimensions={}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%arg0.1 = f32[10000,784]{1,0} parameter(0), parameter_replication={false}, metadata={op_name="XLA_Args"}
%reshape.13 = f32[10000,784]{1,0} reshape(f32[10000,784]{1,0} %arg0.1)
%reshape.37 = f32[10000,784]{1,0} reshape(f32[10000,784]{1,0} %reshape.13), metadata={op_type="Reshape" op_name="Reshape" source_file="/tmp/ipykernel_31814/494562224.py" source_line=16}
%arg2.3 = f32[784,10]{1,0} parameter(2), parameter_replication={false}, metadata={op_name="XLA_Args"}
%dot.38 = f32[10000,10]{1,0} dot(f32[10000,784]{1,0} %reshape.37, f32[784,10]{1,0} %arg2.3), lhs_contracting_dims={1}, rhs_contracting_dims={0}, metadata={op_type="MatMul" op_name="dense/MatMul" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/layers/core.py" source_line=1229}
%transpose.39 = f32[10000,10]{1,0} transpose(f32[10000,10]{1,0} %dot.38), dimensions={0,1}, metadata={op_type="MatMul" op_name="dense/MatMul" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/layers/core.py" source_line=1229}
%arg3.4 = f32[10]{0} parameter(3), parameter_replication={false}, metadata={op_name="XLA_Args"}
%broadcast.40 = f32[10000,10]{1,0} broadcast(f32[10]{0} %arg3.4), dimensions={1}, metadata={op_type="BiasAdd" op_name="dense/BiasAdd" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/layers/core.py" source_line=1240}
%add.41 = f32[10000,10]{1,0} add(f32[10000,10]{1,0} %transpose.39, f32[10000,10]{1,0} %broadcast.40), metadata={op_type="BiasAdd" op_name="dense/BiasAdd" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/layers/core.py" source_line=1240}
%constant.64 = f32[] constant(-inf), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%reduce.69 = f32[10000]{0} reduce(f32[10000,10]{1,0} %add.41, f32[] %constant.64), dimensions={1}, to_apply=%max_float_.65, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.70 = f32[10000,10]{1,0} broadcast(f32[10000]{0} %reduce.69), dimensions={0}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%subtract.71 = f32[10000,10]{1,0} subtract(f32[10000,10]{1,0} %add.41, f32[10000,10]{1,0} %broadcast.70), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%exponential.72 = f32[10000,10]{1,0} exponential(f32[10000,10]{1,0} %subtract.71), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.73 = f32[10000,10]{1,0} convert(f32[10000,10]{1,0} %exponential.72), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.74 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%reduce.79 = f32[10000]{0} reduce(f32[10000,10]{1,0} %convert.73, f32[] %constant.74), dimensions={1}, to_apply=%add_float_.75, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.80 = f32[10000]{0} convert(f32[10000]{0} %reduce.79), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%log.81 = f32[10000]{0} log(f32[10000]{0} %convert.80), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%broadcast.82 = f32[10000,10]{1,0} broadcast(f32[10000]{0} %log.81), dimensions={0}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%subtract.83 = f32[10000,10]{1,0} subtract(f32[10000,10]{1,0} %subtract.71, f32[10000,10]{1,0} %broadcast.82), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%select.89 = f32[10000,10]{1,0} select(pred[10000,10]{1,0} %compare.88, f32[10000,10]{1,0} %broadcast.85, f32[10000,10]{1,0} %subtract.83), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%multiply.91 = f32[10000,10]{1,0} multiply(f32[10000,10]{1,0} %negate.90, f32[10000,10]{1,0} %select.89), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.93 = f32[10000,10]{1,0} convert(f32[10000,10]{1,0} %multiply.91), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.92 = f32[] constant(0), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%reduce.98 = f32[10000]{0} reduce(f32[10000,10]{1,0} %convert.93, f32[] %constant.92), dimensions={1}, to_apply=%add_float_.94, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.99 = f32[10000]{0} convert(f32[10000]{0} %reduce.98), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.103 = f32[10000]{0} convert(f32[10000]{0} %convert.99), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.104 = f32[] constant(0), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.105 = f32[] convert(f32[] %constant.104), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%reduce.110 = f32[] reduce(f32[10000]{0} %convert.103, f32[] %convert.105), dimensions={0}, to_apply=%Mean-reduction.106, metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.111 = s32[] constant(10000), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.112 = f32[] convert(s32[] %constant.111), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%divide.113 = f32[] divide(f32[] %reduce.110, f32[] %convert.112), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%convert.114 = f32[] convert(f32[] %divide.113), metadata={op_type="Mean" op_name="Mean" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%constant.115 = f32[] constant(0.0001), metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%broadcast.116 = f32[10000,1]{1,0} broadcast(f32[] %constant.115), dimensions={}, metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%arg8.9 = f32[784,10]{1,0} parameter(8), parameter_replication={false}, metadata={op_name="XLA_Args"}
%constant.117 = f32[] constant(0.0001), metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%broadcast.118 = f32[10000,1]{1,0} broadcast(f32[] %constant.117), dimensions={}, metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%reshape.119 = f32[10000]{0} reshape(f32[10000,1]{1,0} %broadcast.118), metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%broadcast.120 = f32[10000,10]{1,0} broadcast(f32[10000]{0} %reshape.119), dimensions={0}, metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%broadcast.100 = f32[10000,10]{1,0} broadcast(f32[10000]{0} %convert.80), dimensions={0}, metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%divide.101 = f32[10000,10]{1,0} divide(f32[10000,10]{1,0} %exponential.72, f32[10000,10]{1,0} %broadcast.100), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%subtract.102 = f32[10000,10]{1,0} subtract(f32[10000,10]{1,0} %divide.101, f32[10000,10]{1,0} %add.63), metadata={op_type="SparseSoftmaxCrossEntropyWithLogits" op_name="SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=7}
%multiply.121 = f32[10000,10]{1,0} multiply(f32[10000,10]{1,0} %broadcast.120, f32[10000,10]{1,0} %subtract.102), metadata={op_type="Mul" op_name="gradient_tape/SparseSoftmaxCrossEntropyWithLogits/mul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%dot.153 = f32[784,10]{1,0} dot(f32[10000,784]{1,0} %reshape.37, f32[10000,10]{1,0} %multiply.121), lhs_contracting_dims={0}, rhs_contracting_dims={0}, metadata={op_type="MatMul" op_name="gradient_tape/dense/MatMul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%transpose.154 = f32[784,10]{1,0} transpose(f32[784,10]{1,0} %dot.153), dimensions={0,1}, metadata={op_type="MatMul" op_name="gradient_tape/dense/MatMul" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%subtract.163 = f32[784,10]{1,0} subtract(f32[784,10]{1,0} %transpose.154, f32[784,10]{1,0} %arg8.9), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%constant.156 = f32[] constant(1), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.162 = f32[] subtract(f32[] %constant.156, f32[] %arg6.7), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.164 = f32[784,10]{1,0} broadcast(f32[] %subtract.162), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.165 = f32[784,10]{1,0} multiply(f32[784,10]{1,0} %subtract.163, f32[784,10]{1,0} %broadcast.164), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.166 = f32[784,10]{1,0} add(f32[784,10]{1,0} %arg8.9, f32[784,10]{1,0} %multiply.165), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.158 = f32[] subtract(f32[] %constant.156, f32[] %power.29), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%sqrt.159 = f32[] sqrt(f32[] %subtract.158), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.160 = f32[] multiply(f32[] %arg4.5, f32[] %sqrt.159), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.157 = f32[] subtract(f32[] %constant.156, f32[] %power.25), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%divide.161 = f32[] divide(f32[] %multiply.160, f32[] %subtract.157), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.176 = f32[784,10]{1,0} broadcast(f32[] %divide.161), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.177 = f32[784,10]{1,0} multiply(f32[784,10]{1,0} %add.166, f32[784,10]{1,0} %broadcast.176), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%arg9.10 = f32[784,10]{1,0} parameter(9), parameter_replication={false}, metadata={op_name="XLA_Args"}
%multiply.168 = f32[784,10]{1,0} multiply(f32[784,10]{1,0} %transpose.154, f32[784,10]{1,0} %transpose.154), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.169 = f32[784,10]{1,0} subtract(f32[784,10]{1,0} %multiply.168, f32[784,10]{1,0} %arg9.10), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.167 = f32[] subtract(f32[] %constant.156, f32[] %arg7.8), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.170 = f32[784,10]{1,0} broadcast(f32[] %subtract.167), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.171 = f32[784,10]{1,0} multiply(f32[784,10]{1,0} %subtract.169, f32[784,10]{1,0} %broadcast.170), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.172 = f32[784,10]{1,0} add(f32[784,10]{1,0} %arg9.10, f32[784,10]{1,0} %multiply.171), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%sqrt.173 = f32[784,10]{1,0} sqrt(f32[784,10]{1,0} %add.172), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%constant.155 = f32[] constant(1e-07), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.174 = f32[784,10]{1,0} broadcast(f32[] %constant.155), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.175 = f32[784,10]{1,0} add(f32[784,10]{1,0} %sqrt.173, f32[784,10]{1,0} %broadcast.174), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%divide.178 = f32[784,10]{1,0} divide(f32[784,10]{1,0} %multiply.177, f32[784,10]{1,0} %add.175), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.179 = f32[784,10]{1,0} subtract(f32[784,10]{1,0} %arg2.3, f32[784,10]{1,0} %divide.178), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%reshape.182 = f32[784,10]{1,0} reshape(f32[784,10]{1,0} %subtract.179), metadata={op_name="XLA_Retvals"}
%tuple.183 = (f32[784,10]{1,0}) tuple(f32[784,10]{1,0} %reshape.182), metadata={op_name="XLA_Retvals"}
%get-tuple-element.184 = f32[784,10]{1,0} get-tuple-element((f32[784,10]{1,0}) %tuple.183), index=0, metadata={op_name="XLA_Retvals"}
%arg10.11 = f32[10]{0} parameter(10), parameter_replication={false}, metadata={op_name="XLA_Args"}
%constant.122 = f32[] constant(0), metadata={op_type="BiasAddGrad" op_name="gradient_tape/dense/BiasAdd/BiasAddGrad" source_file="/tmp/ipykernel_31814/3922067182.py" source_line=11}
%reduce.127 = f32[10]{0} reduce(f32[10000,10]{1,0} %multiply.121, f32[] %constant.122), dimensions={0}, to_apply=%region_0.123, metadata={op_name="gradient_tape/dense/BiasAdd/BiasAddGrad"}
%subtract.136 = f32[10]{0} subtract(f32[10]{0} %reduce.127, f32[10]{0} %arg10.11), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%constant.129 = f32[] constant(1), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.135 = f32[] subtract(f32[] %constant.129, f32[] %arg6.7), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.137 = f32[10]{0} broadcast(f32[] %subtract.135), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.138 = f32[10]{0} multiply(f32[10]{0} %subtract.136, f32[10]{0} %broadcast.137), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.139 = f32[10]{0} add(f32[10]{0} %arg10.11, f32[10]{0} %multiply.138), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.131 = f32[] subtract(f32[] %constant.129, f32[] %power.29), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%sqrt.132 = f32[] sqrt(f32[] %subtract.131), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.133 = f32[] multiply(f32[] %arg4.5, f32[] %sqrt.132), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.130 = f32[] subtract(f32[] %constant.129, f32[] %power.25), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%divide.134 = f32[] divide(f32[] %multiply.133, f32[] %subtract.130), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.149 = f32[10]{0} broadcast(f32[] %divide.134), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.150 = f32[10]{0} multiply(f32[10]{0} %add.139, f32[10]{0} %broadcast.149), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%arg11.12 = f32[10]{0} parameter(11), parameter_replication={false}, metadata={op_name="XLA_Args"}
%multiply.141 = f32[10]{0} multiply(f32[10]{0} %reduce.127, f32[10]{0} %reduce.127), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.142 = f32[10]{0} subtract(f32[10]{0} %multiply.141, f32[10]{0} %arg11.12), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.140 = f32[] subtract(f32[] %constant.129, f32[] %arg7.8), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.143 = f32[10]{0} broadcast(f32[] %subtract.140), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%multiply.144 = f32[10]{0} multiply(f32[10]{0} %subtract.142, f32[10]{0} %broadcast.143), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.145 = f32[10]{0} add(f32[10]{0} %arg11.12, f32[10]{0} %multiply.144), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%sqrt.146 = f32[10]{0} sqrt(f32[10]{0} %add.145), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%constant.128 = f32[] constant(1e-07), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%broadcast.147 = f32[10]{0} broadcast(f32[] %constant.128), dimensions={}, metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%add.148 = f32[10]{0} add(f32[10]{0} %sqrt.146, f32[10]{0} %broadcast.147), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%divide.151 = f32[10]{0} divide(f32[10]{0} %multiply.150, f32[10]{0} %add.148), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%subtract.152 = f32[10]{0} subtract(f32[10]{0} %arg3.4, f32[10]{0} %divide.151), metadata={op_type="ResourceApplyAdam" op_name="Adam/Adam/update_1/ResourceApplyAdam" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/adam.py" source_line=176}
%reshape.185 = f32[10]{0} reshape(f32[10]{0} %subtract.152), metadata={op_name="XLA_Retvals"}
%tuple.186 = (f32[10]{0}) tuple(f32[10]{0} %reshape.185), metadata={op_name="XLA_Retvals"}
%get-tuple-element.187 = f32[10]{0} get-tuple-element((f32[10]{0}) %tuple.186), index=0, metadata={op_name="XLA_Retvals"}
%constant.180 = s64[] constant(1), metadata={op_type="AssignAddVariableOp" op_name="Adam/Adam/AssignAddVariableOp" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/optimizer_v2.py" source_line=725}
%add.181 = s64[] add(s64[] %arg5.6, s64[] %constant.180), metadata={op_type="AssignAddVariableOp" op_name="Adam/Adam/AssignAddVariableOp" source_file="/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/keras/optimizer_v2/optimizer_v2.py" source_line=725}
%reshape.188 = s64[] reshape(s64[] %add.181), metadata={op_name="XLA_Retvals"}
%tuple.189 = (s64[]) tuple(s64[] %reshape.188), metadata={op_name="XLA_Retvals"}
%get-tuple-element.190 = s64[] get-tuple-element((s64[]) %tuple.189), index=0, metadata={op_name="XLA_Retvals"}
%reshape.191 = f32[784,10]{1,0} reshape(f32[784,10]{1,0} %add.166), metadata={op_name="XLA_Retvals"}
%tuple.192 = (f32[784,10]{1,0}) tuple(f32[784,10]{1,0} %reshape.191), metadata={op_name="XLA_Retvals"}
%get-tuple-element.193 = f32[784,10]{1,0} get-tuple-element((f32[784,10]{1,0}) %tuple.192), index=0, metadata={op_name="XLA_Retvals"}
%reshape.194 = f32[784,10]{1,0} reshape(f32[784,10]{1,0} %add.172), metadata={op_name="XLA_Retvals"}
%tuple.195 = (f32[784,10]{1,0}) tuple(f32[784,10]{1,0} %reshape.194), metadata={op_name="XLA_Retvals"}
%get-tuple-element.196 = f32[784,10]{1,0} get-tuple-element((f32[784,10]{1,0}) %tuple.195), index=0, metadata={op_name="XLA_Retvals"}
%reshape.197 = f32[10]{0} reshape(f32[10]{0} %add.139), metadata={op_name="XLA_Retvals"}
%tuple.198 = (f32[10]{0}) tuple(f32[10]{0} %reshape.197), metadata={op_name="XLA_Retvals"}
%get-tuple-element.199 = f32[10]{0} get-tuple-element((f32[10]{0}) %tuple.198), index=0, metadata={op_name="XLA_Retvals"}
%reshape.200 = f32[10]{0} reshape(f32[10]{0} %add.145), metadata={op_name="XLA_Retvals"}
%tuple.201 = (f32[10]{0}) tuple(f32[10]{0} %reshape.200), metadata={op_name="XLA_Retvals"}
%get-tuple-element.202 = f32[10]{0} get-tuple-element((f32[10]{0}) %tuple.201), index=0, metadata={op_name="XLA_Retvals"}
ROOT %tuple.203 = (f32[784,10]{1,0}, f32[10]{0}, s64[], f32[784,10]{1,0}, f32[784,10]{1,0}, /*index=5*/f32[10]{0}, f32[10]{0}) tuple(f32[784,10]{1,0} %get-tuple-element.184, f32[10]{0} %get-tuple-element.187, s64[] %get-tuple-element.190, f32[784,10]{1,0} %get-tuple-element.193, f32[784,10]{1,0} %get-tuple-element.196, /*index=5*/f32[10]{0} %get-tuple-element.199, f32[10]{0} %get-tuple-element.202), metadata={op_name="XLA_Retvals"}
}