1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""TPU Distribution Strategy.
This is experimental. It's not ready for general use.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.contrib import tpu
from tensorflow.contrib.distribute.python import one_device_strategy
from tensorflow.contrib.distribute.python import values
from tensorflow.contrib.tpu.python.ops import tpu_ops
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.util import nest
class TPUStrategy(one_device_strategy.OneDeviceStrategy):
"""Experimental TPU distribution strategy implementation."""
def __init__(self, num_cores_per_host=2):
# TODO(isaprykin): Generalize the defaults. They are currently tailored for
# the unit test.
super(TPUStrategy, self).__init__('/cpu:0')
# TODO(isaprykin): Auto-detect number of cores and hosts.
self._num_cores_per_host = num_cores_per_host
# TODO(priyag): This should not be hardcoded here.
self._host = '/task:0/device:CPU:0'
def distribute_dataset(self, dataset_fn):
# TODO(priyag): Perhaps distribute across cores here.
return self._call_dataset_fn(dataset_fn)
# TODO(priyag): Deal with OutOfRange errors.
# TODO(sourabhbajaj): Remove the initial_loop_values parameter when we have
# a mechanism to infer the outputs of `fn`. Pending b/110550782.
def _run_steps_on_dataset(self, fn, iterator, iterations,
initial_loop_values=None):
# Enqueue ops
shapes = nest.flatten(iterator.output_shapes)
if any([not s.is_fully_defined() for s in shapes]):
raise ValueError(
'TPU currently requires fully defined shapes. Either use '
'set_shape() on the input tensors or use '
'dataset.apply(map_and_batch(..., drop_remainder=True)).')
types = nest.flatten(iterator.output_types)
def enqueue_ops_fn():
"""Enqueue ops for one iteration."""
control_deps = []
sharded_inputs = []
with ops.device(self._host):
for _ in range(self._num_cores_per_host):
# Use control dependencies to ensure a deterministic ordering.
with ops.control_dependencies(control_deps):
inputs = nest.flatten(iterator.get_next())
control_deps.extend(inputs)
sharded_inputs.append(inputs)
enqueue_ops = []
for core_id, shard_input in enumerate(sharded_inputs):
enqueue_ops.append(
tpu_ops.infeed_enqueue_tuple(
inputs=shard_input, shapes=shapes, device_ordinal=core_id))
return enqueue_ops
def enqueue_ops_loop_body(i):
with ops.control_dependencies(enqueue_ops_fn()):
return i + 1
with ops.device(self._host):
enqueue_ops = control_flow_ops.while_loop(
lambda i: i < iterations,
enqueue_ops_loop_body,
[constant_op.constant(0)],
parallel_iterations=1)
# Dequeue ops
def dequeue_fn():
dequeued = tpu.infeed_dequeue_tuple(dtypes=types, shapes=shapes)
return nest.pack_sequence_as(iterator.output_shapes, dequeued)
# Wrap `fn` for repeat.
if initial_loop_values is None:
initial_loop_values = []
ctx = values.MultiStepContext(initial_loop_values)
def run_fn(*args, **kwargs):
del args, kwargs
fn_result = fn(ctx, dequeue_fn())
if ctx.last_step_outputs is None:
ctx.last_step_outputs = []
with ops.control_dependencies([fn_result]):
return array_ops.identity(ctx.last_step_outputs)
# Repeat
# TODO(sourabhbajaj): The input to while loop should be based on the output
# type of the step_fn
def iterate_on_tpu():
return tpu.repeat(iterations, run_fn, [initial_loop_values])
# Re-write and distribute computation.
# TODO(sourabhbajaj): Convert the output to PerDevice variable and
# implement support for that in reduce.
last_step_tensor_outputs = tpu.batch_parallel(
iterate_on_tpu, [], num_shards=self._num_cores_per_host)
# Take index [0] of last_step_tensor_outputs as we wrapped
# initial_loop_values in a list in the `repeat` call.
return (control_flow_ops.group(last_step_tensor_outputs, enqueue_ops),
last_step_tensor_outputs[0], ctx)
def _call_for_each_tower(self, fn, *args, **kwargs):
kwargs.pop('run_concurrently', None)
with one_device_strategy._OneDeviceTowerContext(self): # pylint: disable=protected-access
return fn(*args, **kwargs)
def get_initialization_ops(self):
return [tpu.initialize_system()]
def get_finalize_ops(self):
return [tpu.shutdown_system()]
def _reduce(self, method_string, value, destinations):
del destinations # TPU is graph mode only. Rely on implicit Send/Recv.
if method_string == 'mean':
# TODO(jhseu): Revisit once we support model-parallelism.
value *= (1. / self._num_cores_per_host)
return tpu_ops.cross_replica_sum(value)
@property
def num_towers(self):
return self._num_cores_per_host
|
