Closed windhxs closed 2 years ago
Hi @windhxs ,
fastest solution would be to set CUDA_VISIBLE_DEVICES
environment variable (incl. the id of your desired GPU to use), such as: export CUDA_VISIBLE_DEVICES=1
or you can prepend this variable when calling python: CUDA_VISIBLE_DEVICES=1 python3 cli.py
.
More infos here :hugs:
thanks a lot!
It seems that your code will utilize all GPU,but I share a server with others, and it turns out to be something like
cuda out of memory
. Is there any parameters that can assign which GPU to use?