RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you

最新推荐文章于 2024-05-07 17:19:42 发布

原创

最新推荐文章于 2024-05-07 17:19:42 发布 · 1.9k 阅读

3 ·

CC 4.0 BY-SA版权

文章标签：

#python

在运行PyTorch程序时遇到RuntimeError，原因是CUDA不能在子进程中重新初始化。为了解决这个问题，需要将数据加载器的num_workers参数设置为0，以避免使用多进程。这样可以确保CUDA与multiprocessing的兼容性。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

报错如下:
Traceback (most recent call last):
File “E2Capsnet.py”, line 777, in
main()
File “E2Capsnet.py”, line 619, in main
train(epoch)
File “E2Capsnet.py”, line 470, in train
for batch_id, (data, label) in enumerate(raf.train_loader):
File “/home/nuaa301/anaconda3/envs/yang_torch/lib/python3.6/site-packages/torch/utils/data/dataloader.py”, line 517, in next
data = self._next_data()
File “/home/nuaa301/anaconda3/envs/yang_torch/lib/python3.6/site-packages/torch/utils/data/dataloader.py”, line 1199, in _next_data
return self._process_data(data)
File “/home/nuaa301/anaconda3/envs/yang_torch/lib/python3.6/site-packages/torch/utils/data/dataloader.py”, line 1225, in _process_data
data.reraise()
File “/home/nuaa301/anaconda3/envs/yang_torch/lib/python3.6/site-packages/torch/_utils.py”, line 429, in reraise
raise self.exc_type(msg)
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original