Replies: 1 comment
-
Thanks, @cromz22, for TIPS! |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Currently, ESPnet installation automatically detects the CUDA versions from the
nvcc
command.However, in some environments, the
nvcc
command is not available or shows a different version from the onenvidia-smi
shows. (cf. Different CUDA versions shown by nvcc and NVIDIA-smi)In such cases, even though PyTorch can indeed work with GPUs, the Makefile installs a CPU version of PyTorch.
My current workaround is to finish the installation process as usual and then execute the following commands:
(The key is to remove the
cpuonly
package, which tells conda to install CPU version of pytorch. If you just reinstall pytorch conda will install CPU version again. The package exists only in conda and not in pip.)This is not really an issue of ESPnet if CUDA environment is configured properly, so I'm just commenting this on a discussion. I hope this could be helpful to someone who stumbled into the same issue.
Beta Was this translation helpful? Give feedback.
All reactions