forked from horovod/horovod
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdocker-compose.test.yml
177 lines (163 loc) · 6.21 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
version: '2.3'
services:
test-cpu-base:
build:
context: .
dockerfile: Dockerfile.test.cpu
args:
UBUNTU_VERSION: 20.04
GPP_VERSION: 7
MPI_KIND: None
PYTHON_VERSION: 3.9
TENSORFLOW_PACKAGE: tensorflow-cpu==2.14.0
KERAS_PACKAGE: keras==2.14.0
PYTORCH_PACKAGE: torch==2.1.2+cpu
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision==0.16.2+cpu
MXNET_PACKAGE: mxnet==1.9.1
PYSPARK_PACKAGE: pyspark==3.4.0
SPARK_PACKAGE: spark-3.4.0/spark-3.4.0-bin-hadoop3.tgz
HOROVOD_BUILD_FLAGS: HOROVOD_WITH_GLOO=1
privileged: true
shm_size: 8gb
# our baseline first
test-cpu-gloo-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-cpu-base
# permute MPI kinds
test-cpu-mpich-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-cpu-base
build:
args:
MPI_KIND: MPICH
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-oneccl-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-cpu-base
build:
args:
MPI_KIND: ONECCL
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-openmpi-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-cpu-base
build:
args:
MPI_KIND: OpenMPI
HOROVOD_BUILD_FLAGS: HOROVOD_WITHOUT_GLOO=1
test-cpu-openmpi-gloo-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-cpu-base
build:
args:
MPI_KIND: OpenMPI
test-cpu-gloo-py3_9-tf2_12_1-keras2_12_0-torch1_13_1-mxnet1_7_0_p2-pyspark3_4_0:
extends: test-cpu-base
build:
args:
TENSORFLOW_PACKAGE: tensorflow-cpu==2.12.1
KERAS_PACKAGE: keras==2.12.0
PYTORCH_PACKAGE: torch==1.13.1+cpu
TORCHVISION_PACKAGE: torchvision==0.14.1+cpu
MXNET_PACKAGE: mxnet==1.7.0.post2
# then our baseline again, omitted ...
test-cpu-openmpi-gloo-py3_9-tfhead-keras_none-torchhead-mxnethead-pyspark3_4_0:
extends: test-cpu-base
build:
args:
MPI_KIND: OpenMPI
TENSORFLOW_PACKAGE: tf-nightly
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch-nightly
TORCHVISION_PACKAGE: torchvision
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
MXNET_PACKAGE: mxnet-nightly
test-gpu-base:
build:
context: .
dockerfile: Dockerfile.test.gpu
args:
GPP_VERSION: 7
MPI_KIND: None
PYTHON_VERSION: 3.9
PYSPARK_PACKAGE: pyspark==3.4.0
SPARK_PACKAGE: spark-3.4.0/spark-3.4.0-bin-hadoop3.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
# The container isn't provided for CUDA 10 anymore. The lowest version of mxnet available for cu112 is 1.8.0.post0.
test-gpu-gloo-py3_9-tf2_12_1-keras2_12_0-torch1_13_1-mxnet1_8_0_p0-pyspark3_4_0:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.6.2-devel-ubuntu20.04
CUDNN_VERSION: 8.4.1.50-1+cuda11.6
NCCL_VERSION_OVERRIDE: 2.11.4-1+cuda11.6
TENSORFLOW_PACKAGE: tensorflow==2.12.1
KERAS_PACKAGE: keras==2.12.0
PYTORCH_PACKAGE: torch==1.13.1+cu116
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision==0.14.1+cu116
MXNET_PACKAGE: mxnet-cu112==1.8.0.post0
test-gpu-gloo-py3_9-tf2_13_0-keras2_13_1-torch2_0_1-mxnet1_8_0_p0-pyspark3_4_0:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.6.2-devel-ubuntu20.04
CUDNN_VERSION: 8.4.1.50-1+cuda11.6
NCCL_VERSION_OVERRIDE: 2.11.4-1+cuda11.6
# tensorflow package supports GPU from 2.11.1 and 2.12.0 on
TENSORFLOW_PACKAGE: tensorflow==2.13.0
KERAS_PACKAGE: keras==2.13.1
PYTORCH_PACKAGE: torch==2.0.1+cu118
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision==0.15.2+cu118
MXNET_PACKAGE: mxnet-cu112==1.8.0.post0
test-gpu-openmpi-gloo-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.8.0-devel-ubuntu20.04
CUDNN_VERSION: 8.6.0.163-1+cuda11.8
NCCL_VERSION_OVERRIDE: 2.16.5-1+cuda11.8
MPI_KIND: OpenMPI
# tensorflow package supports GPU from 2.11.1 and 2.12.0 on
TENSORFLOW_PACKAGE: tensorflow==2.14.0
KERAS_PACKAGE: keras==2.14.0
PYTORCH_PACKAGE: torch==2.1.2+cu118
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision==0.16.2+cu118
MXNET_PACKAGE: mxnet-cu112==1.9.1
test-gpu-openmpi-gloo-py3_9-tfhead-keras_none-torchhead-mxnethead-pyspark3_4_0:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.8.0-devel-ubuntu20.04
CUDNN_VERSION: 8.6.0.163-1+cuda11.8
NCCL_VERSION_OVERRIDE: 2.16.5-1+cuda11.8
MPI_KIND: OpenMPI
TENSORFLOW_PACKAGE: tf-nightly
KERAS_PACKAGE: None
PYTORCH_PACKAGE: torch-nightly-cu118
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision
MXNET_PACKAGE: mxnet-nightly-cu112
test-mixed-openmpi-gloo-py3_9-tf2_14_0-keras2_14_0-torch2_1_2-mxnet1_9_1-pyspark3_4_0:
extends: test-gpu-base
build:
args:
CUDA_DOCKER_VERSION: 11.8.0-devel-ubuntu20.04
CUDNN_VERSION: 8.6.0.163-1+cuda11.8
NCCL_VERSION_OVERRIDE: 2.16.5-1+cuda11.8
MPI_KIND: OpenMPI
# tensorflow package supports GPU from 2.11.1 and 2.12.0 on
TENSORFLOW_PACKAGE: tensorflow==2.14.0
KERAS_PACKAGE: keras==2.14.0
PYTORCH_PACKAGE: torch==2.1.2+cu118
PYTORCH_LIGHTNING_PACKAGE: pytorch-lightning==1.5.9
TORCHVISION_PACKAGE: torchvision==0.16.2+cu118
MXNET_PACKAGE: mxnet-cu112==1.9.1
HOROVOD_BUILD_FLAGS: ""
HOROVOD_MIXED_INSTALL: 1