forked from microsoft/DeepSpeed
-
Notifications
You must be signed in to change notification settings - Fork 1
/
setup.py
executable file
·197 lines (159 loc) · 6.82 KB
/
setup.py
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
"""
Copyright 2020 The Microsoft DeepSpeed Team
DeepSpeed library
Create a new wheel via the following command: python setup.py bdist_wheel
The wheel will be located at: dist/*.whl
"""
import os
import shutil
import subprocess
import warnings
from setuptools import setup, find_packages
try:
import torch
from torch.utils.cpp_extension import BuildExtension
except ImportError:
raise ImportError('Unable to import torch, please visit https://pytorch.org/ '
'to see how to properly install torch on your system.')
from op_builder import ALL_OPS, get_default_compute_capatabilities
def fetch_requirements(path):
with open(path, 'r') as fd:
return [r.strip() for r in fd.readlines()]
install_requires = fetch_requirements('requirements/requirements.txt')
extras_require = {
'1bit_adam': fetch_requirements('requirements/requirements-1bit-adam.txt'),
'readthedocs': fetch_requirements('requirements/requirements-readthedocs.txt'),
'dev': fetch_requirements('requirements/requirements-dev.txt'),
}
# If MPI is available add 1bit-adam requirements
if torch.cuda.is_available():
if shutil.which('ompi_info') or shutil.which('mpiname'):
cupy = f"cupy-cuda{torch.version.cuda.replace('.','')[:3]}"
extras_require['1bit_adam'].append(cupy)
# Make an [all] extra that installs all needed dependencies
all_extras = set()
for extra in extras_require.items():
for req in extra[1]:
all_extras.add(req)
extras_require['all'] = list(all_extras)
cmdclass = {}
# For any pre-installed ops force disable ninja
cmdclass['build_ext'] = BuildExtension.with_options(use_ninja=False)
TORCH_MAJOR = torch.__version__.split('.')[0]
TORCH_MINOR = torch.__version__.split('.')[1]
if not torch.cuda.is_available():
# Fix to allow docker builds, similar to https://github.com/NVIDIA/apex/issues/486
print(
"[WARNING] Torch did not find cuda available, if cross-compiling or running with cpu only "
"you can ignore this message. Adding compute capability for Pascal, Volta, and Turing "
"(compute capabilities 6.0, 6.1, 6.2)")
if os.environ.get("TORCH_CUDA_ARCH_LIST", None) is None:
os.environ["TORCH_CUDA_ARCH_LIST"] = get_default_compute_capatabilities()
ext_modules = []
# Default to pre-install kernels to false so we rely on JIT
BUILD_OP_DEFAULT = int(os.environ.get('DS_BUILD_OPS', 0))
print(f"DS_BUILD_OPS={BUILD_OP_DEFAULT}")
def command_exists(cmd):
result = subprocess.Popen(f'type {cmd}', stdout=subprocess.PIPE, shell=True)
return result.wait() == 0
def op_enabled(op_name):
assert hasattr(ALL_OPS[op_name], 'BUILD_VAR'), \
f"{op_name} is missing BUILD_VAR field"
env_var = ALL_OPS[op_name].BUILD_VAR
return int(os.environ.get(env_var, BUILD_OP_DEFAULT))
install_ops = dict.fromkeys(ALL_OPS.keys(), False)
for op_name, builder in ALL_OPS.items():
op_compatible = builder.is_compatible()
# If op is compatible update install reqs so it can potentially build/run later
if op_compatible:
reqs = builder.python_requirements()
install_requires += builder.python_requirements()
# If op install enabled, add builder to extensions
if op_enabled(op_name) and op_compatible:
install_ops[op_name] = op_enabled(op_name)
ext_modules.append(builder.builder())
compatible_ops = {op_name: op.is_compatible() for (op_name, op) in ALL_OPS.items()}
print(f'Install Ops={install_ops}')
# Write out version/git info
git_hash_cmd = "git rev-parse --short HEAD"
git_branch_cmd = "git rev-parse --abbrev-ref HEAD"
if command_exists('git') and 'DS_BUILD_STRING' not in os.environ:
try:
result = subprocess.check_output(git_hash_cmd, shell=True)
git_hash = result.decode('utf-8').strip()
result = subprocess.check_output(git_branch_cmd, shell=True)
git_branch = result.decode('utf-8').strip()
except subprocess.CalledProcessError:
git_hash = "unknown"
git_branch = "unknown"
else:
git_hash = "unknown"
git_branch = "unknown"
# Parse the DeepSpeed version string from version.txt
version_str = open('version.txt', 'r').read().strip()
# Build specifiers like .devX can be added at install time. Otherwise, add the git hash.
# example: DS_BUILD_STR=".dev20201022" python setup.py sdist bdist_wheel
#version_str += os.environ.get('DS_BUILD_STRING', f'+{git_hash}')
# Building wheel for distribution, update version file
if 'DS_BUILD_STRING' in os.environ:
# Build string env specified, probably building for distribution
with open('build.txt', 'w') as fd:
fd.write(os.environ.get('DS_BUILD_STRING'))
version_str += os.environ.get('DS_BUILD_STRING')
elif os.path.isfile('build.txt'):
# build.txt exists, probably installing from distribution
with open('build.txt', 'r') as fd:
version_str += fd.read().strip()
else:
# None of the above, probably installing from source
version_str += f'+{git_hash}'
torch_version = ".".join([TORCH_MAJOR, TORCH_MINOR])
# Set cuda_version to 0.0 if cpu-only
cuda_version = "0.0"
if torch.version.cuda is not None:
cuda_version = ".".join(torch.version.cuda.split('.')[:2])
torch_info = {"version": torch_version, "cuda_version": cuda_version}
print(f"version={version_str}, git_hash={git_hash}, git_branch={git_branch}")
with open('deepspeed/git_version_info_installed.py', 'w') as fd:
fd.write(f"version='{version_str}'\n")
fd.write(f"git_hash='{git_hash}'\n")
fd.write(f"git_branch='{git_branch}'\n")
fd.write(f"installed_ops={install_ops}\n")
fd.write(f"compatible_ops={compatible_ops}\n")
fd.write(f"torch_info={torch_info}\n")
print(f'install_requires={install_requires}')
print(f'compatible_ops={compatible_ops}')
print(f'ext_modules={ext_modules}')
# Parse README.md to make long_description for PyPI page.
thisdir = os.path.abspath(os.path.dirname(__file__))
with open(os.path.join(thisdir, 'README.md'), encoding='utf-8') as fin:
readme_text = fin.read()
setup(name='deepspeed',
version=version_str,
description='DeepSpeed library',
long_description=readme_text,
long_description_content_type='text/markdown',
author='DeepSpeed Team',
author_email='[email protected]',
url='http://deepspeed.ai',
install_requires=install_requires,
extras_require=extras_require,
packages=find_packages(exclude=["docker",
"third_party"]),
include_package_data=True,
scripts=[
'bin/deepspeed',
'bin/deepspeed.pt',
'bin/ds',
'bin/ds_ssh',
'bin/ds_report',
'bin/ds_elastic'
],
classifiers=[
'Programming Language :: Python :: 3.6',
'Programming Language :: Python :: 3.7',
'Programming Language :: Python :: 3.8'
],
license='MIT',
ext_modules=ext_modules,
cmdclass=cmdclass)