The package limits the pytorch version to <2.0, which at this point is over 1 year old. This by itself runs the risk of running into issues with the installed CUDA version (unless pytorch is compiled from source), but it also can run into conflicts when used with other packages that require a more recent version of pytorch.
Is there a particular reason to have such a conservative version for the main dependency?
The package limits the pytorch version to <2.0, which at this point is over 1 year old. This by itself runs the risk of running into issues with the installed CUDA version (unless pytorch is compiled from source), but it also can run into conflicts when used with other packages that require a more recent version of pytorch. Is there a particular reason to have such a conservative version for the main dependency?