forked from horovod/horovod
-
Notifications
You must be signed in to change notification settings - Fork 0
/
docker-compose.test.yml
233 lines (227 loc) · 8.78 KB
/
docker-compose.test.yml
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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
version: '2.3'
services:
test-cpu-base:
build:
context: .
dockerfile: Dockerfile.test.cpu
args:
UBUNTU_VERSION: 18.04
GPP_VERSION: 7
MPI_KIND: None
PYTHON_VERSION: 3.8
TENSORFLOW_PACKAGE: tensorflow-cpu==2.4.1
KERAS_PACKAGE: keras==2.4.3
PYTORCH_PACKAGE: torch==1.7.1+cpu
TORCHVISION_PACKAGE: torchvision==0.8.2+cpu
# mxnet==1.7.0.post2 is the first version in 1.7.x that works with horovod
# https://github.com/apache/incubator-mxnet/issues/19575
MXNET_PACKAGE: mxnet==1.7.0.post2
PYSPARK_PACKAGE: pyspark==3.0.1
SPARK_PACKAGE: spark-3.0.1/spark-3.0.1-bin-hadoop2.7.tgz
HOROVOD_BUILD_FLAGS: HOROVOD_WITH_GLOO=1
privileged: true
shm_size: 8gb
test-cpu-gloo-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
test-cpu-mpich-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
MPI_KIND: MPICH
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-oneccl-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
MPI_KIND: ONECCL
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-openmpi-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
MPI_KIND: OpenMPI
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-openmpi-gloo-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
MPI_KIND: OpenMPI
test-cpu-gloo-py3_7-tf1_15_5-keras2_2_4-torch1_2_0-mxnet1_5_1_p0-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
PYTHON_VERSION: 3.7
# there is no tensorflow-cpu>1.15.0, so we use tensorflow==1.15.5
TENSORFLOW_PACKAGE: tensorflow==1.15.5
KERAS_PACKAGE: keras==2.2.4
PYTORCH_PACKAGE: torch==1.2.0+cpu
TORCHVISION_PACKAGE: torchvision==0.4.0+cpu
MXNET_PACKAGE: mxnet==1.5.1.post0
test-cpu-gloo-py3_7-tf2_0_4-keras2_3_1-torch1_3_1-mxnet1_5_1_p0-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
PYTHON_VERSION: 3.7
# there is no tensorflow-cpu==2.0.*, so we use tensorflow==2.0.4
TENSORFLOW_PACKAGE: tensorflow==2.0.4
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.3.1+cpu
TORCHVISION_PACKAGE: torchvision==0.4.2+cpu
MXNET_PACKAGE: mxnet==1.5.1.post0
test-cpu-gloo-py3_7-tf2_1_3-keras2_3_1-torch1_4_0-mxnet1_5_1_p0-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
PYTHON_VERSION: 3.7
TENSORFLOW_PACKAGE: tensorflow-cpu==2.1.3
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.4.0+cpu
TORCHVISION_PACKAGE: torchvision==0.5.0+cpu
MXNET_PACKAGE: mxnet==1.5.1.post0
test-cpu-gloo-py3_8-tf2_2_2-keras2_3_1-torch1_5_1-mxnet1_5_1_p0-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
TENSORFLOW_PACKAGE: tensorflow-cpu==2.2.2
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.5.1+cpu
TORCHVISION_PACKAGE: torchvision==0.6.1+cpu
MXNET_PACKAGE: mxnet==1.5.1.post0
test-cpu-gloo-py3_8-tf2_4_1-keras_none-torch1_7_1-mxnet1_7_0_p2-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
TENSORFLOW_PACKAGE: tensorflow-cpu==2.4.1
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch==1.7.1+cpu
TORCHVISION_PACKAGE: torchvision==0.8.2+cpu
test-cpu-gloo-py3_8-tfhead-keras_none-torchhead-mxnethead-pyspark_3_0_1:
extends: test-cpu-base
build:
args:
TENSORFLOW_PACKAGE: tf-nightly
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch-nightly
TORCHVISION_PACKAGE: torchvision
MXNET_PACKAGE: mxnet-nightly
test-cpu-gloo-py3_6-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_2_3_4:
extends: test-cpu-base
build:
args:
PYTHON_VERSION: 3.6
PYSPARK_PACKAGE: pyspark==2.3.4
SPARK_PACKAGE: spark-2.3.4/spark-2.3.4-bin-hadoop2.7.tgz
test-cpu-gloo-py3_7-tf2_4_1-keras2_4_3-torch1_7_1-mxnet1_7_0_p2-pyspark_2_4_7:
extends: test-cpu-base
build:
args:
PYTHON_VERSION: 3.7
PYSPARK_PACKAGE: pyspark==2.4.7
SPARK_PACKAGE: spark-2.4.7/spark-2.4.7-bin-hadoop2.7.tgz
test-gpu-base:
build:
context: .
dockerfile: Dockerfile.test.gpu
args:
CUDA_DOCKER_VERSION: 10.1-devel-ubuntu18.04
CUDNN_VERSION: 7.6.5.32-1+cuda10.1
NCCL_VERSION_OVERRIDE: 2.7.8-1+cuda10.1
GPP_VERSION: 7
MPI_KIND: None
PYTHON_VERSION: 3.8
TENSORFLOW_PACKAGE: tensorflow-gpu==2.3.2
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.6.0+cu101
TORCHVISION_PACKAGE: torchvision==0.7.0+cu101
# mxnet-cu101==1.7.0.post1 is the first version in 1.7.x that works with horovod
# https://github.com/apache/incubator-mxnet/issues/19575
MXNET_PACKAGE: mxnet-cu101==1.7.0.post1
PYSPARK_PACKAGE: pyspark==3.0.1
SPARK_PACKAGE: spark-3.0.1/spark-3.0.1-bin-hadoop2.7.tgz
HOROVOD_BUILD_FLAGS: HOROVOD_GPU_OPERATIONS=NCCL
HOROVOD_MIXED_INSTALL: 0
runtime: nvidia
# We plumb CUDA_VISIBLE_DEVICES instead of NVIDIA_VISIBLE_DEVICES because
# the latter does not work in privileged mode that we use in the containers.
environment:
- CUDA_VISIBLE_DEVICES
privileged: true
shm_size: 8gb
# torch==1.2.0+cu100 does not exist, torch==1.3.0+cu100 is the first +cu100
# torch==1.3.1+cu100 requires torchvision==0.4.2+cu100
test-gpu-gloo-py3_7-tf1_15_5-keras2_2_4-torch1_3_1-mxnet1_5_1_p0-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 10.0-devel-ubuntu18.04
PYTHON_VERSION: 3.7
TENSORFLOW_PACKAGE: tensorflow-gpu==1.15.5
KERAS_PACKAGE: keras==2.2.4
PYTORCH_PACKAGE: torch==1.3.1+cu100
TORCHVISION_PACKAGE: torchvision==0.4.2+cu100
MXNET_PACKAGE: mxnet-cu100==1.5.1.post0
# there is no mxnet-1.6.0.post0 and mxnet-1.6.0 does not work with horovod
# https://github.com/apache/incubator-mxnet/issues/16193
# however, there is an mxnet-cu101-1.6.0.post0, so we test this with gpu instead of cpu
test-gpu-gloo-py3_8-tf2_3_2-keras2_3_1-torch1_6_0-mxnet1_6_0_p0-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
TENSORFLOW_PACKAGE: tensorflow-gpu==2.3.2
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.6.0+cu101
TORCHVISION_PACKAGE: torchvision==0.7.0+cu101
MXNET_PACKAGE: mxnet-cu101==1.6.0.post0
# mxnet-cu101==1.7.0.post1 is the first version in 1.7.x that works with horovod
# https://github.com/apache/incubator-mxnet/issues/19575
test-gpu-gloo-py3_8-tf2_3_2-keras2_3_1-torch1_6_0-mxnet1_7_0_p1-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
TENSORFLOW_PACKAGE: tensorflow-gpu==2.3.2
KERAS_PACKAGE: keras==2.3.1
PYTORCH_PACKAGE: torch==1.6.0+cu101
TORCHVISION_PACKAGE: torchvision==0.7.0+cu101
MXNET_PACKAGE: mxnet-cu101==1.7.0.post1
# mxnet-cu110 does not exist so we use mxnet_head
test-gpu-openmpi-gloo-py3_8-tf2_4_1-keras_none-torch1_7_1-mxnethead-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.0-devel-ubuntu18.04
CUDNN_VERSION: 8.0.5.39-1+cuda11.0
NCCL_VERSION_OVERRIDE: 2.7.8-1+cuda11.0
MPI_KIND: OpenMPI
TENSORFLOW_PACKAGE: tensorflow-gpu==2.4.1
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch==1.7.1+cu110
TORCHVISION_PACKAGE: torchvision==0.8.2+cu110
MXNET_PACKAGE: mxnet-nightly-cu110
test-gpu-gloo-py3_8-tfhead-keras_none-torchhead-mxnethead-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.0-devel-ubuntu18.04
CUDNN_VERSION: 8.0.5.39-1+cuda11.0
NCCL_VERSION_OVERRIDE: 2.7.8-1+cuda11.0
TENSORFLOW_PACKAGE: tf-nightly-gpu
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch-nightly
TORCHVISION_PACKAGE: torchvision
MXNET_PACKAGE: mxnet-nightly-cu110
# mxnet-cu110 does not exist so we use mxnet_head
test-mixed-openmpi-gloo-py3_8-tf2_4_1-keras2_4_3-torch1_7_1-mxnethead-pyspark_3_0_1:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.0-devel-ubuntu18.04
CUDNN_VERSION: 8.0.5.39-1+cuda11.0
NCCL_VERSION_OVERRIDE: 2.7.8-1+cuda11.0
MPI_KIND: OpenMPI
TENSORFLOW_PACKAGE: tensorflow-gpu==2.4.1
KERAS_PACKAGE: keras==2.4.3
PYTORCH_PACKAGE: torch==1.7.1+cu110
TORCHVISION_PACKAGE: torchvision==0.8.2+cu110
MXNET_PACKAGE: mxnet-nightly-cu110
HOROVOD_BUILD_FLAGS: ""
HOROVOD_MIXED_INSTALL: 1