100 lines
3.2 KiB
Python
100 lines
3.2 KiB
Python
|
# test_cuda.py
|
|||
|
|
"""
|
|||
|
|
CUDA 环境全面检测脚本
|
|||
|
|
Author: Qwen
|
|||
|
|
功能：检测 NVIDIA 驱动、CUDA、PyTorch 与 GPU 是否正常工作
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
import torch
|
|||
|
|
import sys
|
|||
|
|
import os
|
|||
|
|
|
|||
|
|
|
|||
|
|
def print_separator():
|
|||
|
|
print("\n" + "=" * 60 + "\n")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def check_nvidia_smi():
|
|||
|
|
print("🔍 正在检查 nvidia-smi（NVIDIA 驱动和 GPU 状态）...")
|
|||
|
|
try:
|
|||
|
|
result = os.popen("nvidia-smi").read()
|
|||
|
|
print(result)
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"❌ 执行 nvidia-smi 失败: {e}")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def check_torch_cuda():
|
|||
|
|
print("🔍 正在检查 PyTorch 与 CUDA 的集成...")
|
|||
|
|
print(f"➡️ PyTorch 版本: {torch.__version__}")
|
|||
|
|
|
|||
|
|
# 检查 CUDA 是否可用
|
|||
|
|
is_available = torch.cuda.is_available()
|
|||
|
|
print(f"➡️ torch.cuda.is_available(): {is_available}")
|
|||
|
|
|
|||
|
|
if is_available:
|
|||
|
|
print(f"➡️ CUDA 版本 (PyTorch 使用): {torch.version.cuda}")
|
|||
|
|
print(f"➡️ GPU 数量: {torch.cuda.device_count()}")
|
|||
|
|
print(f"➡️ 当前设备: {torch.cuda.current_device()}")
|
|||
|
|
print(f"➡️ 设备名称: {torch.cuda.get_device_name(0)}")
|
|||
|
|
|
|||
|
|
# 尝试创建一个 GPU 张量
|
|||
|
|
try:
|
|||
|
|
x = torch.randn(3, 3).cuda()
|
|||
|
|
print("✅ GPU 张量创建成功！CUDA 可用且正常工作。")
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"❌ GPU 张量创建失败: {e}")
|
|||
|
|
else:
|
|||
|
|
print("❌ CUDA 不可用！PyTorch 无法使用 GPU。")
|
|||
|
|
print("💡 常见原因：")
|
|||
|
|
print(" - 没有安装 NVIDIA 驱动")
|
|||
|
|
print(" - PyTorch 安装的是 CPU 版本")
|
|||
|
|
print(" - CUDA Toolkit 与 PyTorch 不匹配")
|
|||
|
|
print(" - 环境变量 CUDA_VISIBLE_DEVICES 设置错误")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def check_environment():
|
|||
|
|
print("🔍 检查环境变量...")
|
|||
|
|
cuda_visible_devices = os.environ.get("CUDA_VISIBLE_DEVICES", None)
|
|||
|
|
print(f"➡️ CUDA_VISIBLE_DEVICES: {cuda_visible_devices}")
|
|||
|
|
|
|||
|
|
print(f"➡️ Python 路径: {sys.executable}")
|
|||
|
|
print(f"➡️ Python 版本: {sys.version}")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def check_torch_install_type():
|
|||
|
|
print("🔍 检查 PyTorch 安装类型...")
|
|||
|
|
if "cpu" in torch.__version__ or "cpu" in str(torch.version.cuda):
|
|||
|
|
print("⚠️ PyTorch 可能是 CPU 版本")
|
|||
|
|
else:
|
|||
|
|
print("✅ PyTorch 应该是 GPU 版本（含 CUDA 支持）")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def main():
|
|||
|
|
print("🚀 开始检测 CUDA 与 PyTorch 环境...\n")
|
|||
|
|
|
|||
|
|
print_separator()
|
|||
|
|
check_environment()
|
|||
|
|
|
|||
|
|
print_separator()
|
|||
|
|
check_nvidia_smi()
|
|||
|
|
|
|||
|
|
print_separator()
|
|||
|
|
check_torch_install_type()
|
|||
|
|
|
|||
|
|
print_separator()
|
|||
|
|
check_torch_cuda()
|
|||
|
|
|
|||
|
|
print_separator()
|
|||
|
|
if torch.cuda.is_available():
|
|||
|
|
print("🎉 恭喜！CUDA 和 PyTorch 集成正常，可以使用 GPU 训练！")
|
|||
|
|
else:
|
|||
|
|
print("💔 问题严重：CUDA 不可用，请按以下步骤排查：")
|
|||
|
|
print(" 1. 运行 `nvidia-smi` 看是否显示 GPU 和驱动版本")
|
|||
|
|
print(" 2. 如果没有，安装 NVIDIA 驱动")
|
|||
|
|
print(" 3. 重新安装 PyTorch GPU 版本：")
|
|||
|
|
print(" pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121")
|
|||
|
|
print(" 4. 确保没有设置 CUDA_VISIBLE_DEVICES= 或设为无效值")
|
|||
|
|
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
main()
|