解决问题Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insuffi

plaintextCopy code
Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime

shellCopy code
$ nvcc --version

pythonCopy code
import torch
import subprocess
def check_cuda_driver_version():
    # 获取当前CUDA驱动程序版本
    try:
        output = subprocess.check_output(["nvcc", "--version"])
        version_str = output.decode("utf-8")
        version_lines = version_str.split("\n")
        for line in version_lines:
            if "release" in line:
                version = line.split()[-1]
                return version
    except (subprocess.CalledProcessError, FileNotFoundError):
        print("Failed to check CUDA driver version.")
        return None
def install_cuda_driver():
    # 下载并安装适当版本的CUDA驱动程序
    cuda_version = "11.2"  # 需要的CUDA版本
    driver_url = f"https://www.nvidia.com/Download/driverResults.aspx/{cuda_version}"
    print(f"Please manually download and install CUDA driver version {cuda_version} from {driver_url}")
def main():
    cuda_driver_version = check_cuda_driver_version()
    if cuda_driver_version is not None:
        required_driver_version = "11.2"  # 需要的最低驱动程序版本
        if cuda_driver_version < required_driver_version:
            print(f"Your CUDA driver version ({cuda_driver_version}) is insufficient for CUDA runtime.")
            install_cuda_driver()
        else:
            print("Your CUDA driver version is compatible with CUDA runtime.")
            # 进行后续的CUDA计算操作
    else:
        print("Failed to check CUDA driver version.")
if __name__ == "__main__":
    main()

pythonCopy code
import torch
import torchvision
def cuda_example():
    # 创建GPU设备
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    
    # 加载数据集
    dataset = torchvision.datasets.CIFAR10("data/", train=True, download=True)
    
    # 创建数据加载器
    data_loader = torch.utils.data.DataLoader(dataset, batch_size=64, shuffle=True)
    
    # 创建模型并将其移动到GPU设备上
    model = torchvision.models.resnet18().to(device)
    
    # 定义损失函数
    criterion = torch.nn.CrossEntropyLoss().to(device)
    
    # 定义优化器
    optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
    
    # 训练循环
    for images, labels in data_loader:
        # 将输入数据和标签移动到GPU设备上
        images = images.to(device)
        labels = labels.to(device)
        
        # 前向传播
        outputs = model(images)
        
        # 计算损失
        loss = criterion(outputs, labels)
        
        # 反向传播
        optimizer.zero_grad()
        loss.backward()
        
        # 更新参数
        optimizer.step()
        
        # 检查CUDA操作的状态码
        status = torch.cuda.synchronize()
        if status != torch.cuda.success:
            print("CUDA operation failed.")
    
    # 清理GPU缓存
    torch.cuda.empty_cache()