- 1 year ago
pip install warp-rnnt Error (M_LN2 is undefined)
If you encounter the error message "M_LN2 is undefined" when installing the warp-rnnt
package using pip, it is likely due to a missing or incomplete installation of CUDA or cuDNN.
Here are some potential solutions to try:
Install CUDA and cuDNN: Make sure that you have installed the latest versions of CUDA and cuDNN, and that they are properly configured in your environment. You can follow the instructions provided by NVIDIA for installing and configuring CUDA and cuDNN on your system.
Check the CUDA and cuDNN versions: Check that the versions of CUDA and cuDNN that you have installed are compatible with the version of
warp-rnnt
you are trying to install. You can find the compatible versions in theREADME.md
file of thewarp-rnnt
GitHub repository.Set the CUDA_HOME and LD_LIBRARY_PATH environment variables: Ensure that the
CUDA_HOME
andLD_LIBRARY_PATH
environment variables are set to the correct paths for your installation of CUDA. These variables are needed for thewarp-rnnt
package to link against the CUDA libraries. You can set these variables using the following commands:bashexport CUDA_HOME=/usr/local/cuda
export LD_LIBRARY_PATH=${CUDA_HOME}/lib64:$LD_LIBRARY_PATH
Try a different version of
warp-rnnt
: If none of the above solutions work, you may need to try installing a different version ofwarp-rnnt
that is compatible with your system.Build
warp-rnnt
from source: Alternatively, you can try building thewarp-rnnt
package from source by cloning the GitHub repository and running the build script. This will allow you to configure the build settings for your specific environment.bashgit clone https://github.com/HawkAaron/warp-transducer.git
cd warp-transducer/pytorch_binding
python setup.py install
I hope one of these solutions works for you!