cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the ‘sp

cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the 'spawn' start method.

报错原因：

在使用sanic实现多线程操作的时候，报出这样的错误。

解决办法：

网上搜索资料发现的三种方法：

1、在main函数中设置工作模式，通过添加下面代码实现：

torch.multiprocessing.set_start_method('spawn')

2、用main包起来

import torchdef main():for i, data in enumerate(dataloader):# do something hereif __name__ == '__main__':main()

3、设置num_workers为0解决

但是，上面三个方面对我的问题不起作用。

我的解决方案：

sanic版本需要设置正确，我之前用pip自动安装的版本是20.9.3，换成20.6.3版本后，问题解决。

惊不惊喜，意不意外！折磨了好多天的bug，终于解决了！！！

cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the ‘sp相关推荐

Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you
RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you ...
cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the ...
cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the 'sp ...
RuntimeError: Cannot re-initialize CUDA in forked subprocess
问题就在于训练所使用的num_workers数量问题,直接将 num_workers=0 就不会报错了
目的：使用CUDA环境变量CUDA_VISIBLE_DEVICES来限定CUDA程序所能使用的GPU
目的:使用CUDA环境变量CUDA_VISIBLE_DEVICES来限定CUDA程序所能使用的GPU 目的:使用CUDA环境变量CUDA_VISIBLE_DEVICES来限定CUDA程序所能使用的GP ...
ubuntu14.04.5装cuda7.5记录(解决unable to locate the kernel source，装cuda黑屏问题，装cuda循环登录问题）
ubuntu14.04.5装cuda7.5记录(解决unable to locate the kernel source,装cuda黑屏问题,装cuda循环登录问题) 参考文章: (1)ubuntu1 ...
type torch.cuda.FloatTensor but found type torch.cuda.ByteTensor
type torch.cuda.FloatTensor but found type torch.cuda.ByteTensor train_label_batch = torch.from_nump ...
pytorch CUDA driver version is insufficient for CUDA runtime version解决
pytorch CUDA driver version is insufficient for CUDA runtime version 最后发现pytorch下载时选择cuda版本,选错了安装就会报 ...
'CUDA driver version is insufficient for CUDA runtime version
ubuntu系统中,运行报错: 'CUDA driver version is insufficient for CUDA runtime version nvidia驱动是384.111, cuda ...
Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime
Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime ...
cuda 安装_win10+VS 2017 安装 CUDA（Visual Studio Integration失败）
安装参考:windows 10系统nvidia的cuda软件的安装 windows 安装 CUDA可能遇到这个问题:选择自定义安装,但是失败,如下图所示,Visual Studio Integrati ...

cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the ‘sp

cannot re-initialize CUDA in forked subprocess.To use CUDA with multiprocessing,you must use the ‘sp相关推荐

最新文章

热门文章