Repository navigation
Expand file tree
/
Copy pathsetup.py
648 lines (526 loc) · 22.2 KB
/
Copy pathsetup.py
File metadata and controls
648 lines (526 loc) · 22.2 KB
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
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
4
8000
48
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
import importlib.util
import logging
import os
import re
import subprocess
import sys
from pathlib import Path
from shutil import which
from typing import Dict, List
import torch
from packaging.version import Version, parse
from setuptools import Extension, find_packages, setup
from setuptools.command.build_ext import build_ext
from setuptools_scm import get_version
from torch.utils.cpp_extension import CUDA_HOME
def load_module_from_path(module_name, path):
spec = importlib.util.spec_from_file_location(module_name, path)
module = importlib.util.module_from_spec(spec)
sys.modules[module_name] = module
spec.loader.exec_module(module)
return module
ROOT_DIR = os.path.dirname(__file__)
logger = logging.getLogger(__name__)
# cannot import envs directly because it depends on vllm,
# which is not installed yet
envs = load_module_from_path('envs', os.path.join(ROOT_DIR, 'vllm', 'envs.py'))
VLLM_TARGET_DEVICE = envs.VLLM_TARGET_DEVICE
if not sys.platform.startswith("linux"):
logger.warning(
"vLLM only supports Linux platform (including WSL). "
"Building on %s, "
"so vLLM may not be able to run correctly", sys.platform)
VLLM_TARGET_DEVICE = "empty"
MAIN_CUDA_VERSION = "12.1"
def is_sccache_available() -> bool:
return which("sccache") is not None
def is_ccache_available() -> bool:
return which("ccache") is not None
def is_ninja_available() -> bool:
return which("ninja") is not None
class CMakeExtension(Extension):
def __init__(self, name: str, cmake_lists_dir: str = '.', **kwa) -> None:
super().__init__(name, sources=[], py_limited_api=True, **kwa)
self.cmake_lists_dir = os.path.abspath(cmake_lists_dir)
class cmake_build_ext(build_ext):
# A dict of extension directories that have been configured.
did_config: Dict[str, bool] = {}
#
# Determine number of compilation jobs and optionally nvcc compile threads.
#
def compute_num_jobs(self):
# `num_jobs` is either the value of the MAX_JOBS environment variable
# (if defined) or the number of CPUs available.
num_jobs = envs.MAX_JOBS
if num_jobs is not None:
num_jobs = int(num_jobs)
logger.info("Using MAX_JOBS=%d as the number of jobs.", num_jobs)
else:
try:
# os.sched_getaffinity() isn't universally available, so fall
# back to os.cpu_count() if we get an error here.
num_jobs = len(os.sched_getaffinity(0))
except AttributeError:
num_jobs = os.cpu_count()
nvcc_threads = None
if _is_cuda() and get_nvcc_cuda_version() >= Version("11.2"):
# `nvcc_threads` is either the value of the NVCC_THREADS
# environment variable (if defined) or 1.
# when it is set, we reduce `num_jobs` to avoid
# overloading the system.
nvcc_threads = envs.NVCC_THREADS
if nvcc_threads is not None:
nvcc_threads = int(nvcc_threads)
logger.info(
"Using NVCC_THREADS=%d as the number of nvcc threads.",
nvcc_threads)
else:
nvcc_threads = 1
num_jobs = max(1, num_jobs // nvcc_threads)
return num_jobs, nvcc_threads
#
# Perform cmake configuration for a single extension.
#
def configure(self, ext: CMakeExtension) -> None:
# If we've already configured using the CMakeLists.txt for
# this extension, exit early.
if ext.cmake_lists_dir in cmake_build_ext.did_config:
return
cmake_build_ext.did_config[ext.cmake_lists_dir] = True
# Select the build type.
# Note: optimization level + debug info are set by the build type
default_cfg = "Debug" if self.debug else "RelWithDebInfo"
cfg = envs.CMAKE_BUILD_TYPE or default_cfg
cmake_args = [
'-DCMAKE_BUILD_TYPE={}'.format(cfg),
'-DVLLM_TARGET_DEVICE={}'.format(VLLM_TARGET_DEVICE),
]
verbose = envs.VERBOSE
if verbose:
cmake_args += ['-DCMAKE_VERBOSE_MAKEFILE=ON']
if is_sccache_available():
cmake_args += [
'-DCMAKE_C_COMPILER_LAUNCHER=sccache',
'-DCMAKE_CXX_COMPILER_LAUNCHER=sccache',
'-DCMAKE_CUDA_COMPILER_LAUNCHER=sccache',
'-DCMAKE_HIP_COMPILER_LAUNCHER=sccache',
]
elif is_ccache_available():
cmake_args += [
'-DCMAKE_C_COMPILER_LAUNCHER=ccache',
'-DCMAKE_CXX_COMPILER_LAUNCHER=ccache',
'-DCMAKE_CUDA_COMPILER_LAUNCHER=ccache',
'-DCMAKE_HIP_COMPILER_LAUNCHER=ccache',
]
# Pass the python executable to cmake so it can find an exact
# match.
cmake_args += ['-DVLLM_PYTHON_EXECUTABLE={}'.format(sys.executable)]
# Pass the python path to cmake so it can reuse the build dependencies
# on subsequent calls to python.
cmake_args += ['-DVLLM_PYTHON_PATH={}'.format(":".join(sys.path))]
# Override the base directory for FetchContent downloads to $ROOT/.deps
# This allows sharing dependencies between profiles,
# and plays more nicely with sccache.
# To override this, set the FETCHCONTENT_BASE_DIR environment variable.
fc_base_dir = os.path.join(ROOT_DIR, ".deps")
fc_base_dir = os.environ.get("FETCHCONTENT_BASE_DIR", fc_base_dir)
cmake_args += ['-DFETCHCONTENT_BASE_DIR={}'.format(fc_base_dir)]
#
# Setup parallelism and build tool
#
num_jobs, nvcc_threads = self.compute_num_jobs()
if nvcc_threads:
cmake_args += ['-DNVCC_THREADS={}'.format(nvcc_threads)]
if is_ninja_available():
build_tool = ['-G', 'Ninja']
cmake_args += [
'-DCMAKE_JOB_POOL_COMPILE:STRING=compile',
'-DCMAKE_JOB_POOLS:STRING=compile={}'.format(num_jobs),
]
else:
# Default build tool to whatever cmake picks.
build_tool = []
subprocess.check_call(
['cmake', ext.cmake_lists_dir, *build_tool, *cmake_args],
cwd=self.build_temp)
def build_extensions(self) -> None:
# Ensure that CMake is present and working
try:
subprocess.check_output(['cmake', '--version'])
except OSError as e:
raise RuntimeError('Cannot find CMake executable') from e
# Create build directory if it does not exist.
if not os.path.exists(self.build_temp):
os.makedirs(self.build_temp)
targets = []
def target_name(s: str) -> str:
return s.removeprefix("vllm.").removeprefix("vllm_flash_attn.")
# Build all the extensions
for ext in self.extensions:
self.configure(ext)
targets.append(target_name(ext.name))
num_jobs, _ = self.compute_num_jobs()
build_args = [
"--build",
".",
f"-j={num_jobs}",
*[f"--target={name}" for name in targets],
]
subprocess.check_call(["cmake", *build_args], cwd=self.build_temp)
# Install the libraries
for ext in self.extensions:
# Install the extension into the proper location
outdir = Path(self.get_ext_fullpath(ext.name)).parent.absolute()
# Skip if the install directory is the same as the build directory
if outdir == self.build_temp:
continue
# CMake appends the extension prefix to the install path,
# and outdir already contains that prefix, so we need to remove it.
prefix = outdir
for i in range(ext.name.count('.')):
prefix = prefix.parent
# prefix here should actually be the same for all components
install_args = [
"cmake", "--install", ".", "--prefix", prefix, "--component",
target_name(ext.name)
]
subprocess.check_call(install_args, cwd=self.build_temp)
def run(self):
# First, run the standard build_ext command to compile the extensions
super().run()
# copy vllm/vllm_flash_attn/*.py from self.build_lib to current
# directory so that they can be included in the editable build
import glob
files = glob.glob(
os.path.join(self.build_lib, "vllm", "vllm_flash_attn", "*.py"))
for file in files:
dst_file = os.path.join("vllm/vllm_flash_attn",
os.path.basename(file))
print(f"Copying {file} to {dst_file}")
self.copy_file(file, dst_file)
class repackage_wheel(build_ext):
"""Extracts libraries and other files from an existing wheel."""
default_wheel = "https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl"
def run(self) -> None:
wheel_location = os.getenv("VLLM_PRECOMPILED_WHEEL_LOCATION",
self.default_wheel)
assert _is_cuda(
), "VLLM_USE_PRECOMPILED is only supported for CUDA builds"
import zipfile
if os.path.isfile(wheel_location):
wheel_path = wheel_location
print(f"Using existing wheel={wheel_path}")
else:
# Download the wheel from a given URL, assume
# the filename is the last part of the URL
wheel_filename = wheel_location.split("/")[-1]
import tempfile
# create a temporary directory to store the wheel
temp_dir = tempfile.mkdtemp(prefix="vllm-wheels")
wheel_path = os.path.join(temp_dir, wheel_filename)
print(f"Downloading wheel from {wheel_location} to {wheel_path}")
from urllib.request import urlretrieve
try:
urlretrieve(wheel_location, filename=wheel_path)
except Exception as e:
from setuptools.errors import SetupError
raise SetupError(
f"Failed to get vLLM wheel from {wheel_location}") from e
with zipfile.ZipFile(wheel_path) as wheel:
files_to_copy = [
"vllm/_C.abi3.so",
"vllm/_moe_C.abi3.so",
"vllm/vllm_flash_attn/vllm_flash_attn_c.abi3.so",
"vllm/vllm_flash_attn/flash_attn_interface.py",
"vllm/vllm_flash_attn/__init__.py",
# "vllm/_version.py", # not available in nightly wheels yet
]
file_members = filter(lambda x: x.filename in files_to_copy,
wheel.filelist)
for file in file_members:
print(f"Extracting and including {file.filename} "
"from existing wheel")
package_name = os.path.dirname(file.filename).replace("/", ".")
file_name = os.path.basename(file.filename)
if package_name not in package_data:
package_data[package_name] = []
wheel.extract(file)
if file_name.endswith(".py"):
# python files shouldn't be added to package_data
continue
package_data[package_name].append(file_name)
def _is_hpu() -> bool:
is_hpu_available = True
try:
subprocess.run(["hl-smi"], capture_output=True, check=True)
except (FileNotFoundError, PermissionError, subprocess.CalledProcessError):
if not os.path.exists('/dev/accel/accel0') and not os.path.exists(
'/dev/accel/accel_controlD0'):
# last resort...
try:
output = subprocess.check_output(
'lsmod | grep habanalabs | wc -l', shell=True)
is_hpu_available = int(output) > 0
except (ValueError, FileNotFoundError, PermissionError,
subprocess.CalledProcessError):
is_hpu_available = False
return is_hpu_available or VLLM_TARGET_DEVICE == "hpu"
def _no_device() -> bool:
return VLLM_TARGET_DEVICE == "empty"
def _is_cuda() -> bool:
has_cuda = torch.version.cuda is not None
return (VLLM_TARGET_DEVICE == "cuda" and has_cuda
and not (_is_neuron() or _is_tpu() or _is_hpu()))
def _is_hip() -> bool:
return (VLLM_TARGET_DEVICE == "cuda"
or VLLM_TARGET_DEVICE == "rocm") and torch.version.hip is not None
def _is_neuron() -> bool:
torch_neuronx_installed = True
try:
subprocess.run(["neuron-ls"], capture_output=True, check=True)
except (FileNotFoundError, PermissionError, subprocess.CalledProcessError):
torch_neuronx_installed = False
return torch_neuronx_installed or VLLM_TARGET_DEVICE == "neuron"
def _is_tpu() -> bool:
return VLLM_TARGET_DEVICE == "tpu"
def _is_cpu() -> bool:
return VLLM_TARGET_DEVICE == "cpu"
def _is_openvino() -> bool:
return VLLM_TARGET_DEVICE == "openvino"
def _is_xpu() -> bool:
return VLLM_TARGET_DEVICE == "xpu"
def _build_custom_ops() -> bool:
return _is_cuda() or _is_hip() or _is_cpu()
def get_hipcc_rocm_version():
# Run the hipcc --version command
result = subprocess.run(['hipcc', '--version'],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
text=True)
# Check if the command was executed successfully
if result.returncode != 0:
print("Error running 'hipcc --version'")
return None
# Extract the version using a regular expression
match = re.search(r'HIP version: (\S+)', result.stdout)
if match:
# Return the version string
return match.group(1)
else:
print("Could not find HIP version in the output")
return None
def get_neuronxcc_version():
import sysconfig
site_dir = sysconfig.get_paths()["purelib"]
version_file = os.path.join(site_dir, "neuronxcc", "version",
"__init__.py")
# Check if the command was executed successfully
with open(version_file) as fp:
content = fp.read()
# Extract the version using a regular expression
match = re.search(r"__version__ = '(\S+)'", content)
if match:
# Return the version string
return match.group(1)
else:
raise RuntimeError("Could not find Neuron version in the output")
def get_nvcc_cuda_version() -> Version:
"""Get the CUDA version from nvcc.
Adapted from https://github.com/NVIDIA/apex/blob/8b7a1ff183741dd8f9b87e7bafd04cfde99cea28/setup.py
"""
assert CUDA_HOME is not None, "CUDA_HOME is not set"
nvcc_output = subprocess.check_output([CUDA_HOME + "/bin/nvcc", "-V"],
universal_newlines=True)
output = nvcc_output.split()
release_idx = output.index("release") + 1
nvcc_cuda_version = parse(output[release_idx].split(",")[0])
return nvcc_cuda_version
def get_path(*filepath) -> str:
return os.path.join(ROOT_DIR, *filepath)
def get_gaudi_sw_version():
"""
Returns the driver version.
"""
# Enable console printing for `hl-smi` check
output = subprocess.run("hl-smi",
shell=True,
text=True,
capture_output=True,
env={"ENABLE_CONSOLE": "true"})
if output.returncode == 0 and output.stdout:
return output.stdout.split("\n")[2].replace(
" ", "").split(":")[1][:-1].split("-")[0]
return "0.0.0" # when hl-smi is not available
def get_vllm_version() -> str:
# TODO: Revisit this temporary approach: https://github.com/vllm-project/vllm/issues/9182#issuecomment-2404860236
try:
version = get_version(
write_to="vllm/_version.py", # TODO: move this to pyproject.toml
)
except LookupError:
version = "0.0.0"
sep = "+" if "+" not in version else "." # dev versions might contain +
if _no_device():
if envs.VLLM_TARGET_DEVICE == "empty":
version += f"{sep}empty"
elif _is_cuda():
if envs.VLLM_USE_PRECOMPILED:
version += f"{sep}precompiled"
else:
cuda_version = str(get_nvcc_cuda_version())
if cuda_version != MAIN_CUDA_VERSION:
cuda_version_str = cuda_version.replace(".", "")[:3]
# skip this for source tarball, required for pypi
if "sdist" not in sys.argv:
version += f"{sep}cu{cuda_version_str}"
elif _is_hip():
# Get the HIP version
hipcc_version = get_hipcc_rocm_version()
if hipcc_version != MAIN_CUDA_VERSION:
rocm_version_str = hipcc_version.replace(".", "")[:3]
version += f"{sep}rocm{rocm_version_str}"
elif _is_neuron():
# Get the Neuron version
neuron_version = str(get_neuronxcc_version())
if neuron_version != MAIN_CUDA_VERSION:
neuron_version_str = neuron_version.replace(".", "")[:3]
version += f"{sep}neuron{neuron_version_str}"
elif _is_hpu():
# Get the Intel Gaudi Software Suite version
gaudi_sw_version = str(get_gaudi_sw_version())
if gaudi_sw_version != MAIN_CUDA_VERSION:
gaudi_sw_version = gaudi_sw_version.replace(".", "")[:3]
version += f"{sep}gaudi{gaudi_sw_version}"
elif _is_openvino():
version += f"{sep}openvino"
elif _is_tpu():
version += f"{sep}tpu"
elif _is_cpu():
version += f"{sep}cpu"
elif _is_xpu():
version += f"{sep}xpu"
else:
raise RuntimeError("Unknown runtime environment")
return version
def read_readme() -> str:
"""Read the README file if present."""
p = get_path("README.md")
if os.path.isfile(p):
with open(get_path("README.md"), encoding="utf-8") as f:
return f.read()
else:
return ""
def get_requirements() -> List[str]:
"""Get Python package dependencies from requirements.txt."""
def _read_requirements(filename: str) -> List[str]:
with open(get_path(filename)) as f:
requirements = f.read().strip().split("\n")
resolved_requirements = []
for line in requirements:
if line.startswith("-r "):
resolved_requirements += _read_requirements(line.split()[1])
elif line.startswith("--"):
continue
else:
resolved_requirements.append(line)
return resolved_requirements
if _no_device():
requirements = _read_requirements("requirements-cuda.txt")
elif _is_cuda():
requirements = _read_requirements("requirements-cuda.txt")
cuda_major, cuda_minor = torch.version.cuda.split(".")
modified_requirements = []
for req in requirements:
if ("vllm-flash-attn" in req
and not (cuda_major == "12" and cuda_minor == "1")):
# vllm-flash-attn is built only for CUDA 12.1.
# Skip for other versions.
continue
modified_requirements.append(req)
requirements = modified_requirements
elif _is_hip():
requirements = _read_requirements("requirements-rocm.txt")
elif _is_neuron():
requirements = _read_requirements("requirements-neuron.txt")
elif _is_hpu():
requirements = _read_requirements("requirements-hpu.txt")
elif _is_openvino():
requirements = _read_requirements("requirements-openvino.txt")
elif _is_tpu():
requirements = _read_requirements("requirements-tpu.txt")
elif _is_cpu():
requirements = _read_requirements("requirements-cpu.txt")
elif _is_xpu():
requirements = _read_requirements("requirements-xpu.txt")
else:
raise ValueError(
"Unsupported platform, please use CUDA, ROCm, Neuron, HPU, "
"OpenVINO, or CPU.")
return requirements
ext_modules = []
if _is_cuda() or _is_hip():
ext_modules.append(CMakeExtension(name="vllm._moe_C"))
if _is_hip():
ext_modules.append(CMakeExtension(name="vllm._rocm_C"))
if _is_cuda():
ext_modules.append(
CMakeExtension(name="vllm.vllm_flash_attn.vllm_flash_attn_c"))
if _build_custom_ops():
ext_modules.append(CMakeExtension(name="vllm._C"))
package_data = {
"vllm": ["py.typed", "model_executor/layers/fused_moe/configs/*.json"]
}
if _no_device():
ext_modules = []
if not ext_modules:
cmdclass = {}
else:
cmdclass = {
"build_ext":
repackage_wheel if envs.VLLM_USE_PRECOMPILED else cmake_build_ext
}
setup(
name="vllm",
version=get_vllm_version(),
author="vLLM Team",
license="Apache 2.0",
description=("A high-throughput and memory-efficient inference and "
"serving engine for LLMs"),
long_description=read_readme(),
long_description_content_type="text/markdown",
url="https://github.com/vllm-project/vllm",
project_urls={
"Homepage": "https://github.com/vllm-project/vllm",
"Documentation": "https://vllm.readthedocs.io/en/latest/",
},
classifiers=[
"Programming Language :: Python :: 3.9",
"Programming Language :: Python :: 3.10",
"Programming Language :: Python :: 3.11",
"Programming Language :: Python :: 3.12",
"License :: OSI Approved :: Apache Software License",
"Intended Audience :: Developers",
"Intended Audience :: Information Technology",
"Intended Audience :: Science/Research",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
"Topic :: Scientific/Engineering :: Information Analysis",
],
packages=find_packages(exclude=("benchmarks", "csrc", "docs", "examples",
"tests*")),
python_requires=">=3.9",
install_requires=get_requirements(),
ext_modules=ext_modules,
extras_require={
"tensorizer": ["tensorizer>=2.9.0"],
"runai": ["runai-model-streamer", "runai-model-streamer-s3", "boto3"],
"audio": ["librosa", "soundfile"], # Required for audio processing
"video": ["decord"] # Required for video processing
},
cmdclass=cmdclass,
package_data=package_data,
entry_points={
"console_scripts": [
"vllm=vllm.scripts:main",
],
},
)