You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I wanted to install torch 2.5.1 with CUDA 12.1 to use whisperX, however the current requirements.txt pins the version of ctranslate2==4.4.0. Running the default example present in the README.md file throws an error:
warnings.warn(
Could not load library libcudnn_ops_infer.so.8. Error: libcudnn_ops_infer.so.8: cannot open shared object file: No such file or directory
Aborted
Since the version of torch is not pinned in the requirements.txt file, the (currently) installed version will be 2.5.1 with CUDA 12.4, which is not compatible with ctranslate2==4.4.0. So far I've found that torch versions that work with ctranslate==4.4.0 are <2.4.0. The 4.5.0 release of ctranslate2 updates cuDNN to version 9.
What I'm proposing is to create a matrix of working combinations of ctranslate2, torch, and CUDA, or pin the version of torch such that it works with ctranslate2==4.4.0.
The text was updated successfully, but these errors were encountered:
Hi,
I wanted to install torch 2.5.1 with CUDA 12.1 to use whisperX, however the current
requirements.txt
pins the version ofctranslate2==4.4.0
. Running the default example present in theREADME.md
file throws an error:Since the version of torch is not pinned in the
requirements.txt
file, the (currently) installed version will be 2.5.1 with CUDA 12.4, which is not compatible withctranslate2==4.4.0
. So far I've found that torch versions that work withctranslate==4.4.0
are<2.4.0
. The 4.5.0 release of ctranslate2 updates cuDNN to version 9.What I'm proposing is to create a matrix of working combinations of ctranslate2, torch, and CUDA, or pin the version of torch such that it works with
ctranslate2==4.4.0
.The text was updated successfully, but these errors were encountered: