collect_env.py 27.4 KB
Newer Older
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3

4
# ruff: noqa
5
6
7
8
# code borrowed from https://github.com/pytorch/pytorch/blob/main/torch/utils/collect_env.py

import datetime
import locale
9
import os
10
11
import subprocess
import sys
12

13
14
15
# Unlike the rest of the PyTorch this file must be python2 compliant.
# This script outputs relevant system environment info
# Run it with `python collect_env.py` or `python -m torch.utils.collect_env`
16
17
from collections import namedtuple

18
19
import regex as re

20
21
from vllm.envs import environment_variables

22
23
try:
    import torch
24

25
26
27
28
29
    TORCH_AVAILABLE = True
except (ImportError, NameError, AttributeError, OSError):
    TORCH_AVAILABLE = False

# System Environment Information
30
SystemEnv = namedtuple(
31
    "SystemEnv",
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
        "torch_version",
        "is_debug_build",
        "cuda_compiled_version",
        "gcc_version",
        "clang_version",
        "cmake_version",
        "os",
        "libc_version",
        "python_version",
        "python_platform",
        "is_cuda_available",
        "cuda_runtime_version",
        "cuda_module_loading",
        "nvidia_driver_version",
        "nvidia_gpu_models",
        "cudnn_version",
        "pip_version",  # 'pip' or 'pip3'
        "pip_packages",
        "conda_packages",
        "hip_compiled_version",
        "hip_runtime_version",
        "miopen_runtime_version",
        "caching_allocator_config",
        "is_xnnpack_available",
        "cpu_info",
        "rocm_version",  # vllm specific field
        "vllm_version",  # vllm specific field
        "vllm_build_flags",  # vllm specific field
        "gpu_topo",  # vllm specific field
        "env_vars",
    ],
)
65
66
67
68
69
70
71
72
73
74

DEFAULT_CONDA_PATTERNS = {
    "torch",
    "numpy",
    "cudatoolkit",
    "soumith",
    "mkl",
    "magma",
    "triton",
    "optree",
75
    "nccl",
76
    "transformers",
77
    "zmq",
78
79
    "nvidia",
    "pynvml",
80
    "flashinfer-python",
81
82
83
84
85
86
87
88
89
90
}

DEFAULT_PIP_PATTERNS = {
    "torch",
    "numpy",
    "mypy",
    "flake8",
    "triton",
    "optree",
    "onnx",
91
    "nccl",
92
    "transformers",
93
    "zmq",
94
95
    "nvidia",
    "pynvml",
96
    "flashinfer-python",
97
98
99
100
101
102
}


def run(command):
    """Return (return-code, stdout, stderr)."""
    shell = True if type(command) is str else False
103
    try:
104
105
106
        p = subprocess.Popen(
            command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=shell
        )
107
108
        raw_output, raw_err = p.communicate()
        rc = p.returncode
109
110
        if get_platform() == "win32":
            enc = "oem"
111
112
113
        else:
            enc = locale.getpreferredencoding()
        output = raw_output.decode(enc)
114
        if command == "nvidia-smi topo -m":
115
116
117
118
119
120
121
122
123
124
            # don't remove the leading whitespace of `nvidia-smi topo -m`
            #   because they are meaningful
            output = output.rstrip()
        else:
            output = output.strip()
        err = raw_err.decode(enc)
        return rc, output, err.strip()

    except FileNotFoundError:
        cmd_str = command if isinstance(command, str) else command[0]
125
        return 127, "", f"Command not found: {cmd_str}"
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145


def run_and_read_all(run_lambda, command):
    """Run command using run_lambda; reads and returns entire output if rc is 0."""
    rc, out, _ = run_lambda(command)
    if rc != 0:
        return None
    return out


def run_and_parse_first_match(run_lambda, command, regex):
    """Run command using run_lambda, returns the first regex match if it exists."""
    rc, out, _ = run_lambda(command)
    if rc != 0:
        return None
    match = re.search(regex, out)
    if match is None:
        return None
    return match.group(1)

146

147
148
149
150
151
def run_and_return_first_line(run_lambda, command):
    """Run command using run_lambda and returns first line if output is not empty."""
    rc, out, _ = run_lambda(command)
    if rc != 0:
        return None
152
    return out.split("\n")[0]
153
154
155
156
157


def get_conda_packages(run_lambda, patterns=None):
    if patterns is None:
        patterns = DEFAULT_CONDA_PATTERNS
158
159
    conda = os.environ.get("CONDA_EXE", "conda")
    out = run_and_read_all(run_lambda, [conda, "list"])
160
161
162
    if out is None:
        return out

163
164
165
166
167
    return "\n".join(
        line
        for line in out.splitlines()
        if not line.startswith("#") and any(name in line for name in patterns)
    )
168

169
170

def get_gcc_version(run_lambda):
171
    return run_and_parse_first_match(run_lambda, "gcc --version", r"gcc (.*)")
172

173

174
def get_clang_version(run_lambda):
175
176
177
    return run_and_parse_first_match(
        run_lambda, "clang --version", r"clang version (.*)"
    )
178
179
180


def get_cmake_version(run_lambda):
181
    return run_and_parse_first_match(run_lambda, "cmake --version", r"cmake (.*)")
182
183
184


def get_nvidia_driver_version(run_lambda):
185
186
187
188
189
    if get_platform() == "darwin":
        cmd = "kextstat | grep -i cuda"
        return run_and_parse_first_match(
            run_lambda, cmd, r"com[.]nvidia[.]CUDA [(](.*?)[)]"
        )
190
    smi = get_nvidia_smi()
191
    return run_and_parse_first_match(run_lambda, smi, r"Driver Version: (.*?) ")
192
193
194


def get_gpu_info(run_lambda):
195
196
197
198
199
    if get_platform() == "darwin" or (
        TORCH_AVAILABLE
        and hasattr(torch.version, "hip")
        and torch.version.hip is not None
    ):
200
201
202
203
204
205
206
207
208
209
210
211
        if TORCH_AVAILABLE and torch.cuda.is_available():
            if torch.version.hip is not None:
                prop = torch.cuda.get_device_properties(0)
                if hasattr(prop, "gcnArchName"):
                    gcnArch = " ({})".format(prop.gcnArchName)
                else:
                    gcnArch = "NoGCNArchNameOnOldPyTorch"
            else:
                gcnArch = ""
            return torch.cuda.get_device_name(None) + gcnArch
        return None
    smi = get_nvidia_smi()
212
213
    uuid_regex = re.compile(r" \(UUID: .+?\)")
    rc, out, _ = run_lambda(smi + " -L")
214
215
216
    if rc != 0:
        return None
    # Anonymize GPUs by removing their UUID
217
    return re.sub(uuid_regex, "", out)
218
219
220


def get_running_cuda_version(run_lambda):
221
    return run_and_parse_first_match(run_lambda, "nvcc --version", r"release .+ V(.*)")
222
223
224
225


def get_cudnn_version(run_lambda):
    """Return a list of libcudnn.so; it's hard to tell which one is being used."""
226
227
228
229
    if get_platform() == "win32":
        system_root = os.environ.get("SYSTEMROOT", "C:\\Windows")
        cuda_path = os.environ.get("CUDA_PATH", "%CUDA_PATH%")
        where_cmd = os.path.join(system_root, "System32", "where")
230
        cudnn_cmd = '{} /R "{}\\bin" cudnn*.dll'.format(where_cmd, cuda_path)
231
    elif get_platform() == "darwin":
232
233
234
235
        # CUDA libraries and drivers can be found in /usr/local/cuda/. See
        # https://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#install
        # https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#installmac
        # Use CUDNN_LIBRARY when cudnn library is installed elsewhere.
236
        cudnn_cmd = "ls /usr/local/cuda/lib/libcudnn*"
237
238
239
240
241
    else:
        cudnn_cmd = 'ldconfig -p | grep libcudnn | rev | cut -d" " -f1 | rev'
    rc, out, _ = run_lambda(cudnn_cmd)
    # find will return 1 if there are permission errors or if not found
    if len(out) == 0 or (rc != 1 and rc != 0):
242
        l = os.environ.get("CUDNN_LIBRARY")
243
244
245
246
        if l is not None and os.path.isfile(l):
            return os.path.realpath(l)
        return None
    files_set = set()
247
    for fn in out.split("\n"):
248
249
250
251
252
253
254
255
256
        fn = os.path.realpath(fn)  # eliminate symbolic links
        if os.path.isfile(fn):
            files_set.add(fn)
    if not files_set:
        return None
    # Alphabetize the result because the order is non-deterministic otherwise
    files = sorted(files_set)
    if len(files) == 1:
        return files[0]
257
258
    result = "\n".join(files)
    return "Probably one of the following:\n{}".format(result)
259
260
261
262


def get_nvidia_smi():
    # Note: nvidia-smi is currently available only on Windows and Linux
263
264
265
266
267
268
269
270
    smi = "nvidia-smi"
    if get_platform() == "win32":
        system_root = os.environ.get("SYSTEMROOT", "C:\\Windows")
        program_files_root = os.environ.get("PROGRAMFILES", "C:\\Program Files")
        legacy_path = os.path.join(
            program_files_root, "NVIDIA Corporation", "NVSMI", smi
        )
        new_path = os.path.join(system_root, "System32", smi)
271
272
273
274
275
276
277
278
279
280
        smis = [new_path, legacy_path]
        for candidate_smi in smis:
            if os.path.exists(candidate_smi):
                smi = '"{}"'.format(candidate_smi)
                break
    return smi


def get_rocm_version(run_lambda):
    """Returns the ROCm version if available, otherwise 'N/A'."""
281
282
283
    return run_and_parse_first_match(
        run_lambda, "hipcc --version", r"HIP version: (\S+)"
    )
284
285
286


def get_vllm_version():
287
288
289
290
    from vllm import __version__, __version_tuple__

    if __version__ == "dev":
        return "N/A (dev)"
291
    version_str = __version_tuple__[-1]
292
    if isinstance(version_str, str) and version_str.startswith("g"):
293
        # it's a dev build
294
        if "." in version_str:
295
            # it's a dev build containing local changes
296
297
            git_sha = version_str.split(".")[0][1:]
            date = version_str.split(".")[-1][1:]
298
299
300
301
302
            return f"{__version__} (git sha: {git_sha}, date: {date})"
        else:
            # it's a dev build without local changes
            git_sha = version_str[1:]  # type: ignore
            return f"{__version__} (git sha: {git_sha})"
303
    return __version__
304

305

306
307
def summarize_vllm_build_flags():
    # This could be a static method if the flags are constant, or dynamic if you need to check environment variables, etc.
308
309
310
    return "CUDA Archs: {}; ROCm: {}".format(
        os.environ.get("TORCH_CUDA_ARCH_LIST", "Not Set"),
        "Enabled" if os.environ.get("ROCM_HOME") else "Disabled",
311
312
313
314
    )


def get_gpu_topo(run_lambda):
315
316
    output = None

317
318
    if get_platform() == "linux":
        output = run_and_read_all(run_lambda, "nvidia-smi topo -m")
319
        if output is None:
320
            output = run_and_read_all(run_lambda, "rocm-smi --showtopo")
321
322

    return output
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


# example outputs of CPU infos
#  * linux
#    Architecture:            x86_64
#      CPU op-mode(s):        32-bit, 64-bit
#      Address sizes:         46 bits physical, 48 bits virtual
#      Byte Order:            Little Endian
#    CPU(s):                  128
#      On-line CPU(s) list:   0-127
#    Vendor ID:               GenuineIntel
#      Model name:            Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
#        CPU family:          6
#        Model:               106
#        Thread(s) per core:  2
#        Core(s) per socket:  32
#        Socket(s):           2
#        Stepping:            6
#        BogoMIPS:            5799.78
#        Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr
#                             sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl
#                             xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16
#                             pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand
#                             hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced
#                             fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap
#                             avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1
#                             xsaves wbnoinvd ida arat avx512vbmi pku ospke avx512_vbmi2 gfni vaes vpclmulqdq
#                             avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid md_clear flush_l1d arch_capabilities
#    Virtualization features:
#      Hypervisor vendor:     KVM
#      Virtualization type:   full
#    Caches (sum of all):
#      L1d:                   3 MiB (64 instances)
#      L1i:                   2 MiB (64 instances)
#      L2:                    80 MiB (64 instances)
#      L3:                    108 MiB (2 instances)
#    NUMA:
#      NUMA node(s):          2
#      NUMA node0 CPU(s):     0-31,64-95
#      NUMA node1 CPU(s):     32-63,96-127
#    Vulnerabilities:
#      Itlb multihit:         Not affected
#      L1tf:                  Not affected
#      Mds:                   Not affected
#      Meltdown:              Not affected
#      Mmio stale data:       Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
#      Retbleed:              Not affected
#      Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl and seccomp
#      Spectre v1:            Mitigation; usercopy/swapgs barriers and __user pointer sanitization
#      Spectre v2:            Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
#      Srbds:                 Not affected
#      Tsx async abort:       Not affected
#  * win32
#    Architecture=9
#    CurrentClockSpeed=2900
#    DeviceID=CPU0
#    Family=179
#    L2CacheSize=40960
#    L2CacheSpeed=
#    Manufacturer=GenuineIntel
#    MaxClockSpeed=2900
#    Name=Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
#    ProcessorType=3
#    Revision=27142
#
#    Architecture=9
#    CurrentClockSpeed=2900
#    DeviceID=CPU1
#    Family=179
#    L2CacheSize=40960
#    L2CacheSpeed=
#    Manufacturer=GenuineIntel
#    MaxClockSpeed=2900
#    Name=Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
#    ProcessorType=3
#    Revision=27142

400

401
def get_cpu_info(run_lambda):
402
403
404
405
    rc, out, err = 0, "", ""
    if get_platform() == "linux":
        rc, out, err = run_lambda("lscpu")
    elif get_platform() == "win32":
406
        rc, out, err = run_lambda(
407
408
            "wmic cpu get Name,Manufacturer,Family,Architecture,ProcessorType,DeviceID, \
        CurrentClockSpeed,MaxClockSpeed,L2CacheSize,L2CacheSpeed,Revision /VALUE"
409
        )
410
    elif get_platform() == "darwin":
411
        rc, out, err = run_lambda("sysctl -n machdep.cpu.brand_string")
412
    cpu_info = "None"
413
414
415
416
417
418
419
420
    if rc == 0:
        cpu_info = out
    else:
        cpu_info = err
    return cpu_info


def get_platform():
421
422
423
424
425
426
427
428
    if sys.platform.startswith("linux"):
        return "linux"
    elif sys.platform.startswith("win32"):
        return "win32"
    elif sys.platform.startswith("cygwin"):
        return "cygwin"
    elif sys.platform.startswith("darwin"):
        return "darwin"
429
430
431
432
433
    else:
        return sys.platform


def get_mac_version(run_lambda):
434
    return run_and_parse_first_match(run_lambda, "sw_vers -productVersion", r"(.*)")
435
436
437


def get_windows_version(run_lambda):
438
439
440
    system_root = os.environ.get("SYSTEMROOT", "C:\\Windows")
    wmic_cmd = os.path.join(system_root, "System32", "Wbem", "wmic")
    findstr_cmd = os.path.join(system_root, "System32", "findstr")
441
    return run_and_read_all(
442
443
        run_lambda, "{} os get Caption | {} /v Caption".format(wmic_cmd, findstr_cmd)
    )
444
445
446


def get_lsb_version(run_lambda):
447
448
449
    return run_and_parse_first_match(
        run_lambda, "lsb_release -a", r"Description:\t(.*)"
    )
450
451
452


def check_release_file(run_lambda):
453
454
455
    return run_and_parse_first_match(
        run_lambda, "cat /etc/*-release", r'PRETTY_NAME="(.*)"'
    )
456
457
458
459


def get_os(run_lambda):
    from platform import machine
460

461
462
    platform = get_platform()

463
    if platform == "win32" or platform == "cygwin":
464
465
        return get_windows_version(run_lambda)

466
    if platform == "darwin":
467
468
469
        version = get_mac_version(run_lambda)
        if version is None:
            return None
470
        return "macOS {} ({})".format(version, machine())
471

472
    if platform == "linux":
473
474
475
        # Ubuntu/Debian based
        desc = get_lsb_version(run_lambda)
        if desc is not None:
476
            return "{} ({})".format(desc, machine())
477
478
479
480

        # Try reading /etc/*-release
        desc = check_release_file(run_lambda)
        if desc is not None:
481
            return "{} ({})".format(desc, machine())
482

483
        return "{} ({})".format(platform, machine())
484
485
486
487
488
489
490

    # Unknown platform
    return platform


def get_python_platform():
    import platform
491

492
493
494
495
496
    return platform.platform()


def get_libc_version():
    import platform
497
498
499
500

    if get_platform() != "linux":
        return "N/A"
    return "-".join(platform.libc_ver())
501
502


503
504
505
def is_uv_venv():
    if os.environ.get("UV"):
        return True
506
    pyvenv_cfg_path = os.path.join(sys.prefix, "pyvenv.cfg")
507
    if os.path.exists(pyvenv_cfg_path):
508
509
        with open(pyvenv_cfg_path, "r") as f:
            return any(line.startswith("uv = ") for line in f)
510
511
512
    return False


513
514
515
516
517
def get_pip_packages(run_lambda, patterns=None):
    """Return `pip list` output. Note: will also find conda-installed pytorch and numpy packages."""
    if patterns is None:
        patterns = DEFAULT_PIP_PATTERNS

518
519
520
    def run_with_pip():
        try:
            import importlib.util
521
522

            pip_spec = importlib.util.find_spec("pip")
523
524
525
526
527
            pip_available = pip_spec is not None
        except ImportError:
            pip_available = False

        if pip_available:
528
            cmd = [sys.executable, "-mpip", "list", "--format=freeze"]
529
        elif is_uv_venv():
530
531
532
            print("uv is set")
            cmd = ["uv", "pip", "list", "--format=freeze"]
        else:
533
534
535
            raise RuntimeError(
                "Could not collect pip list output (pip or uv module not available)"
            )
536
537

        out = run_and_read_all(run_lambda, cmd)
538
539
540
        return "\n".join(
            line for line in out.splitlines() if any(name in line for name in patterns)
        )
541

542
    pip_version = "pip3" if sys.version[0] == "3" else "pip"
543
    out = run_with_pip()
544
545
546
547
    return pip_version, out


def get_cachingallocator_config():
548
    ca_config = os.environ.get("PYTORCH_CUDA_ALLOC_CONF", "")
549
550
551
552
553
554
    return ca_config


def get_cuda_module_loading_config():
    if TORCH_AVAILABLE and torch.cuda.is_available():
        torch.cuda.init()
555
        config = os.environ.get("CUDA_MODULE_LOADING", "")
556
557
558
559
560
561
562
563
        return config
    else:
        return "N/A"


def is_xnnpack_available():
    if TORCH_AVAILABLE:
        import torch.backends.xnnpack
564
565

        return str(torch.backends.xnnpack.enabled)  # type: ignore[attr-defined]
566
567
568
    else:
        return "N/A"

569

570
def get_env_vars():
571
572
573
574
575
576
577
578
579
580
581
582
583
    env_vars = ""
    secret_terms = ("secret", "token", "api", "access", "password")
    report_prefix = (
        "TORCH",
        "NCCL",
        "PYTORCH",
        "CUDA",
        "CUBLAS",
        "CUDNN",
        "OMP_",
        "MKL_",
        "NVIDIA",
    )
584
585
586
587
588
589
590
591
592
    for k, v in os.environ.items():
        if any(term in k.lower() for term in secret_terms):
            continue
        if k in environment_variables:
            env_vars = env_vars + "{}={}".format(k, v) + "\n"
        if k.startswith(report_prefix):
            env_vars = env_vars + "{}={}".format(k, v) + "\n"

    return env_vars
593

594

595
596
597
598
599
600
601
602
603
def get_env_info():
    run_lambda = run
    pip_version, pip_list_output = get_pip_packages(run_lambda)

    if TORCH_AVAILABLE:
        version_str = torch.__version__
        debug_mode_str = str(torch.version.debug)
        cuda_available_str = str(torch.cuda.is_available())
        cuda_version_str = torch.version.cuda
604
605
606
607
        if (
            not hasattr(torch.version, "hip") or torch.version.hip is None
        ):  # cuda version
            hip_compiled_version = hip_runtime_version = miopen_runtime_version = "N/A"
608
        else:  # HIP version
609

610
611
            def get_version_or_na(cfg, prefix):
                _lst = [s.rsplit(None, 1)[-1] for s in cfg if prefix in s]
612
                return _lst[0] if _lst else "N/A"
613

614
615
616
617
            cfg = torch._C._show_config().split("\n")
            hip_runtime_version = get_version_or_na(cfg, "HIP Runtime")
            miopen_runtime_version = get_version_or_na(cfg, "MIOpen")
            cuda_version_str = "N/A"
618
619
            hip_compiled_version = torch.version.hip
    else:
620
621
        version_str = debug_mode_str = cuda_available_str = cuda_version_str = "N/A"
        hip_compiled_version = hip_runtime_version = miopen_runtime_version = "N/A"
622
623
624
625
626
627
628
629
630
631
632
633
634

    sys_version = sys.version.replace("\n", " ")

    conda_packages = get_conda_packages(run_lambda)

    rocm_version = get_rocm_version(run_lambda)
    vllm_version = get_vllm_version()
    vllm_build_flags = summarize_vllm_build_flags()
    gpu_topo = get_gpu_topo(run_lambda)

    return SystemEnv(
        torch_version=version_str,
        is_debug_build=debug_mode_str,
635
636
637
        python_version="{} ({}-bit runtime)".format(
            sys_version, sys.maxsize.bit_length() + 1
        ),
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
        python_platform=get_python_platform(),
        is_cuda_available=cuda_available_str,
        cuda_compiled_version=cuda_version_str,
        cuda_runtime_version=get_running_cuda_version(run_lambda),
        cuda_module_loading=get_cuda_module_loading_config(),
        nvidia_gpu_models=get_gpu_info(run_lambda),
        nvidia_driver_version=get_nvidia_driver_version(run_lambda),
        cudnn_version=get_cudnn_version(run_lambda),
        hip_compiled_version=hip_compiled_version,
        hip_runtime_version=hip_runtime_version,
        miopen_runtime_version=miopen_runtime_version,
        pip_version=pip_version,
        pip_packages=pip_list_output,
        conda_packages=conda_packages,
        os=get_os(run_lambda),
        libc_version=get_libc_version(),
        gcc_version=get_gcc_version(run_lambda),
        clang_version=get_clang_version(run_lambda),
        cmake_version=get_cmake_version(run_lambda),
        caching_allocator_config=get_cachingallocator_config(),
        is_xnnpack_available=is_xnnpack_available(),
        cpu_info=get_cpu_info(run_lambda),
        rocm_version=rocm_version,
        vllm_version=vllm_version,
        vllm_build_flags=vllm_build_flags,
        gpu_topo=gpu_topo,
664
        env_vars=get_env_vars(),
665
666
    )

667

668
env_info_fmt = """
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
==============================
        System Info
==============================
OS                           : {os}
GCC version                  : {gcc_version}
Clang version                : {clang_version}
CMake version                : {cmake_version}
Libc version                 : {libc_version}

==============================
       PyTorch Info
==============================
PyTorch version              : {torch_version}
Is debug build               : {is_debug_build}
CUDA used to build PyTorch   : {cuda_compiled_version}
ROCM used to build PyTorch   : {hip_compiled_version}

==============================
      Python Environment
==============================
Python version               : {python_version}
Python platform              : {python_platform}

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : {is_cuda_available}
CUDA runtime version         : {cuda_runtime_version}
CUDA_MODULE_LOADING set to   : {cuda_module_loading}
GPU models and configuration : {nvidia_gpu_models}
Nvidia driver version        : {nvidia_driver_version}
cuDNN version                : {cudnn_version}
HIP runtime version          : {hip_runtime_version}
MIOpen runtime version       : {miopen_runtime_version}
Is XNNPACK available         : {is_xnnpack_available}

==============================
          CPU Info
==============================
708
709
{cpu_info}

710
711
712
==============================
Versions of relevant libraries
==============================
713
714
715
716
{pip_packages}
{conda_packages}
""".strip()

youkaichao's avatar
youkaichao committed
717
718
719
# both the above code and the following code use `strip()` to
# remove leading/trailing whitespaces, so we need to add a newline
# in between to separate the two sections
720
env_info_fmt += "\n\n"
youkaichao's avatar
youkaichao committed
721

722
env_info_fmt += """
723
724
725
726
727
==============================
         vLLM Info
==============================
ROCM Version                 : {rocm_version}
vLLM Version                 : {vllm_version}
728
vLLM Build Flags:
729
  {vllm_build_flags}
730
GPU Topology:
731
  {gpu_topo}
732

733
734
735
==============================
     Environment Variables
==============================
736
{env_vars}
737
738
739
740
""".strip()


def pretty_str(envinfo):
741
    def replace_nones(dct, replacement="Could not collect"):
742
743
744
745
746
747
        for key in dct.keys():
            if dct[key] is not None:
                continue
            dct[key] = replacement
        return dct

748
    def replace_bools(dct, true="Yes", false="No"):
749
750
751
752
753
754
755
        for key in dct.keys():
            if dct[key] is True:
                dct[key] = true
            elif dct[key] is False:
                dct[key] = false
        return dct

756
757
    def prepend(text, tag="[prepend]"):
        lines = text.split("\n")
758
        updated_lines = [tag + line for line in lines]
759
        return "\n".join(updated_lines)
760

761
    def replace_if_empty(text, replacement="No relevant packages"):
762
763
764
765
766
767
        if text is not None and len(text) == 0:
            return replacement
        return text

    def maybe_start_on_next_line(string):
        # If `string` is multiline, prepend a \n to it.
768
769
        if string is not None and len(string.split("\n")) > 1:
            return "\n{}\n".format(string)
770
771
772
773
774
        return string

    mutable_dict = envinfo._asdict()

    # If nvidia_gpu_models is multiline, start on the next line
775
776
777
    mutable_dict["nvidia_gpu_models"] = maybe_start_on_next_line(
        envinfo.nvidia_gpu_models
    )
778
779
780

    # If the machine doesn't have CUDA, report some fields as 'No CUDA'
    dynamic_cuda_fields = [
781
782
783
        "cuda_runtime_version",
        "nvidia_gpu_models",
        "nvidia_driver_version",
784
    ]
785
786
787
788
789
790
791
792
793
    all_cuda_fields = dynamic_cuda_fields + ["cudnn_version"]
    all_dynamic_cuda_fields_missing = all(
        mutable_dict[field] is None for field in dynamic_cuda_fields
    )
    if (
        TORCH_AVAILABLE
        and not torch.cuda.is_available()
        and all_dynamic_cuda_fields_missing
    ):
794
        for field in all_cuda_fields:
795
            mutable_dict[field] = "No CUDA"
796
        if envinfo.cuda_compiled_version is None:
797
            mutable_dict["cuda_compiled_version"] = "None"
798
799
800
801
802
803
804
805

    # Replace True with Yes, False with No
    mutable_dict = replace_bools(mutable_dict)

    # Replace all None objects with 'Could not collect'
    mutable_dict = replace_nones(mutable_dict)

    # If either of these are '', replace with 'No relevant packages'
806
807
    mutable_dict["pip_packages"] = replace_if_empty(mutable_dict["pip_packages"])
    mutable_dict["conda_packages"] = replace_if_empty(mutable_dict["conda_packages"])
808
809
810

    # Tag conda and pip packages with a prefix
    # If they were previously None, they'll show up as ie '[conda] Could not collect'
811
812
813
814
815
816
817
818
819
    if mutable_dict["pip_packages"]:
        mutable_dict["pip_packages"] = prepend(
            mutable_dict["pip_packages"], "[{}] ".format(envinfo.pip_version)
        )
    if mutable_dict["conda_packages"]:
        mutable_dict["conda_packages"] = prepend(
            mutable_dict["conda_packages"], "[conda] "
        )
    mutable_dict["cpu_info"] = envinfo.cpu_info
820
821
822
823
824
825
826
827
828
829
830
831
    return env_info_fmt.format(**mutable_dict)


def get_pretty_env_info():
    return pretty_str(get_env_info())


def main():
    print("Collecting environment information...")
    output = get_pretty_env_info()
    print(output)

832
833
834
835
836
    if (
        TORCH_AVAILABLE
        and hasattr(torch, "utils")
        and hasattr(torch.utils, "_crash_handler")
    ):
837
838
        minidump_dir = torch.utils._crash_handler.DEFAULT_MINIDUMP_DIR
        if sys.platform == "linux" and os.path.exists(minidump_dir):
839
            dumps = [
840
                os.path.join(minidump_dir, dump) for dump in os.listdir(minidump_dir)
841
            ]
842
843
            latest = max(dumps, key=os.path.getctime)
            ctime = os.path.getctime(latest)
844
            creation_time = datetime.datetime.fromtimestamp(ctime).strftime(
845
846
847
848
849
850
851
852
                "%Y-%m-%d %H:%M:%S"
            )
            msg = (
                "\n*** Detected a minidump at {} created on {}, ".format(
                    latest, creation_time
                )
                + "if this is related to your bug please include it when you file a report ***"
            )
853
854
855
            print(msg, file=sys.stderr)


856
if __name__ == "__main__":
857
    main()