Download gmnet/code/journal_exp/scripts/check_env.py from YFanwang/Backup: direct link, hf CLI and curl.
- Browser
- Download file 7.95 kB
-
https://huggingface.co/datasets/YFanwang/Backup/resolve/main/gmnet/code/journal_exp/scripts/check_env.py
- Command line
-
hf download hf://datasets/YFanwang/Backup/gmnet/code/journal_exp/scripts/check_env.py
-
curl -L -o check_env.py https://huggingface.co/datasets/YFanwang/Backup/resolve/main/gmnet/code/journal_exp/scripts/check_env.py
7.95 kB
| #!/usr/bin/env python3 | |
| """Check the local GmNet runtime, storage, GPU, and optional S3 access.""" | |
| from __future__ import annotations | |
| import argparse | |
| import importlib | |
| import json | |
| import os | |
| import shutil | |
| import subprocess | |
| import sys | |
| import tempfile | |
| from pathlib import Path | |
| from typing import Any | |
| REQUIRED_MODULES = ( | |
| "torch", | |
| "torchvision", | |
| "timm", | |
| "yaml", | |
| "wandb", | |
| "boto3", | |
| "webdataset", | |
| "numpy", | |
| "scipy", | |
| ) | |
| CIFAR_SOURCE = Path("/s3-code/ywang29/datasets/cifar-10/cifar-10-python.tar.gz") | |
| CIFAR_BYTES = 170_498_071 | |
| IMAGENET_BUCKET = "snap-research-cv-code" | |
| IMAGENET_KEY = "ywang29/datasets/imagenet-1k/imagenet-1k.tar" | |
| IMAGENET_BYTES = 161_381_969_920 | |
| STRICT_VERSIONS = { | |
| "torch": "2.9.0+cu130", | |
| "torchvision": "0.24.0+cu130", | |
| "timm": "1.0.27", | |
| "yaml": "6.0.3", | |
| } | |
| def _module_version(module: Any) -> str: | |
| return str(getattr(module, "__version__", "unknown")) | |
| def _run(command: list[str], timeout: int = 30) -> subprocess.CompletedProcess[str]: | |
| return subprocess.run( | |
| command, | |
| check=False, | |
| stdout=subprocess.PIPE, | |
| stderr=subprocess.PIPE, | |
| text=True, | |
| timeout=timeout, | |
| ) | |
| def parse_args() -> argparse.Namespace: | |
| parser = argparse.ArgumentParser(description=__doc__) | |
| parser.add_argument("--require-cuda", action="store_true") | |
| parser.add_argument("--min-gpus", type=int, default=0) | |
| parser.add_argument("--check-s3", action="store_true") | |
| parser.add_argument("--require-imagenet-space", action="store_true") | |
| parser.add_argument("--strict-versions", action="store_true") | |
| parser.add_argument("--min-free-bytes", type=int, default=350_000_000_000) | |
| parser.add_argument("--json", action="store_true", dest="as_json") | |
| return parser.parse_args() | |
| def main() -> int: | |
| args = parse_args() | |
| report: dict[str, Any] = { | |
| "ok": True, | |
| "python": sys.version.split()[0], | |
| "executable": sys.executable, | |
| "modules": {}, | |
| "warnings": [], | |
| "errors": [], | |
| } | |
| errors: list[str] = report["errors"] | |
| warnings: list[str] = report["warnings"] | |
| if sys.version_info < (3, 10): | |
| errors.append(f"Python >=3.10 is required, found {sys.version.split()[0]}") | |
| loaded: dict[str, Any] = {} | |
| for module_name in REQUIRED_MODULES: | |
| try: | |
| module = importlib.import_module(module_name) | |
| except Exception as error: # import-time binary errors matter here | |
| errors.append(f"cannot import {module_name}: {error}") | |
| else: | |
| loaded[module_name] = module | |
| report["modules"][module_name] = _module_version(module) | |
| if args.strict_versions: | |
| if sys.version_info[:2] != (3, 12): | |
| errors.append( | |
| f"strict environment requires Python 3.12, found {sys.version.split()[0]}" | |
| ) | |
| for module_name, expected in STRICT_VERSIONS.items(): | |
| if module_name in loaded: | |
| actual = _module_version(loaded[module_name]) | |
| if actual != expected: | |
| errors.append( | |
| f"strict environment requires {module_name}=={expected}, found {actual}" | |
| ) | |
| torch = loaded.get("torch") | |
| if torch is not None: | |
| cuda_available = bool(torch.cuda.is_available()) | |
| gpu_count = int(torch.cuda.device_count()) if cuda_available else 0 | |
| report["torch_cuda"] = str(torch.version.cuda) | |
| report["cuda_available"] = cuda_available | |
| report["gpu_count"] = gpu_count | |
| report["cudnn_version"] = torch.backends.cudnn.version() | |
| if args.require_cuda and not cuda_available: | |
| errors.append("CUDA is required but torch.cuda.is_available() is false") | |
| if gpu_count < args.min_gpus: | |
| errors.append(f"at least {args.min_gpus} GPUs are required, found {gpu_count}") | |
| tmp_path = Path(os.environ.get("TMPDIR", "/tmp")) | |
| try: | |
| tmp_path.mkdir(parents=True, exist_ok=True) | |
| with tempfile.NamedTemporaryFile(dir=tmp_path, prefix="gmnet-env-", delete=True) as handle: | |
| handle.write(b"ok") | |
| handle.flush() | |
| usage = shutil.disk_usage(tmp_path) | |
| report["tmp"] = { | |
| "path": str(tmp_path.resolve()), | |
| "free_bytes": usage.free, | |
| "total_bytes": usage.total, | |
| } | |
| if args.require_imagenet_space and usage.free < args.min_free_bytes: | |
| errors.append( | |
| f"ImageNet staging needs {args.min_free_bytes} free bytes; {tmp_path} has {usage.free}" | |
| ) | |
| except OSError as error: | |
| errors.append(f"temporary directory is not writable: {tmp_path}: {error}") | |
| aws = shutil.which("aws") | |
| report["aws_cli"] = aws | |
| if args.check_s3: | |
| if aws is None: | |
| errors.append("aws CLI was not found") | |
| else: | |
| identity = _run([aws, "sts", "get-caller-identity", "--output", "json"]) | |
| if identity.returncode != 0: | |
| errors.append(f"AWS identity check failed: {identity.stderr.strip()}") | |
| else: | |
| try: | |
| identity_data = json.loads(identity.stdout) | |
| report["aws_identity"] = { | |
| "account": identity_data.get("Account"), | |
| "arn": identity_data.get("Arn"), | |
| } | |
| except json.JSONDecodeError: | |
| warnings.append("AWS identity output was not valid JSON") | |
| head = _run( | |
| [ | |
| aws, | |
| "s3api", | |
| "head-object", | |
| "--bucket", | |
| IMAGENET_BUCKET, | |
| "--key", | |
| IMAGENET_KEY, | |
| "--query", | |
| "ContentLength", | |
| "--output", | |
| "text", | |
| ] | |
| ) | |
| if head.returncode != 0: | |
| errors.append(f"ImageNet S3 head-object failed: {head.stderr.strip()}") | |
| else: | |
| try: | |
| remote_size = int(head.stdout.strip()) | |
| except ValueError: | |
| errors.append(f"invalid ImageNet ContentLength: {head.stdout.strip()!r}") | |
| else: | |
| report["imagenet_s3_bytes"] = remote_size | |
| if remote_size != IMAGENET_BYTES: | |
| errors.append( | |
| f"ImageNet archive size is {remote_size}, expected {IMAGENET_BYTES}" | |
| ) | |
| if not CIFAR_SOURCE.is_file(): | |
| errors.append(f"CIFAR-10 S3 mount object is missing: {CIFAR_SOURCE}") | |
| else: | |
| cifar_size = CIFAR_SOURCE.stat().st_size | |
| report["cifar_mount_bytes"] = cifar_size | |
| if cifar_size != CIFAR_BYTES: | |
| errors.append(f"CIFAR-10 archive size is {cifar_size}, expected {CIFAR_BYTES}") | |
| report["ok"] = not errors | |
| if args.as_json: | |
| print(json.dumps(report, indent=2, sort_keys=True)) | |
| else: | |
| print(f"Python: {report['python']} ({report['executable']})") | |
| for name, version in report["modules"].items(): | |
| print(f" {name}: {version}") | |
| if "gpu_count" in report: | |
| print( | |
| f"CUDA: available={report['cuda_available']} " | |
| f"version={report['torch_cuda']} GPUs={report['gpu_count']}" | |
| ) | |
| if "tmp" in report: | |
| print(f"Temporary storage: {report['tmp']['path']} ({report['tmp']['free_bytes']} bytes free)") | |
| for warning in warnings: | |
| print(f"WARNING: {warning}", file=sys.stderr) | |
| for error in errors: | |
| print(f"ERROR: {error}", file=sys.stderr) | |
| print("Environment check passed" if report["ok"] else "Environment check failed") | |
| return 0 if report["ok"] else 1 | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) | |