diffusers
diffusers copied to clipboard
A bug in multiple gpu training
Describe the bug
when train this code with multiple gpu
I got a bug like this

Reproduction
I just mutify this code
change num_processes=1 to num_processes=2
before this bug another bug is like that
RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method
I change the 'fork' to 'spawn' like that

Logs
No response
System Info
platform in jupyter notebook