Cannot start faster-whisper

[Denis-Kazakov]

I attempted to use faster-whisper in PyCharm, created a new project with dedicated venv and followed the installation instructions given in this repo. However, when I tried to run the example script I got the following error message: 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 OS: Linux Mint

[tylerjthomas9]

We experienced this issue with torch v2.4. Downgrading to v2.3.1 fixed it for us.

[formater]

Denis, any update? Same issue here on Debian 12. Cuda 12 installed, OpenAi`s whisper works, this one not :(

[Denis-Kazakov]

We experienced this issue with torch v2.4. Downgrading to v2.3.1 fixed it for us.

Thanks. I checked torch version. It is already 2.3.1 in my case.

[Denis-Kazakov]

Denis, any update? Same issue here on Debian 12. Cuda 12 installed, OpenAi`s whisper works, this one not :(

Alas no, I have given up on it. I am just a language researcher. I hope to be able to use "slower" Whisper for my application (online speech translation) through simpler logic. Another faster option is Vosk, but Whisper is better.

[Gldkslfmsd]

Note that LD_LIBRARY_PATH must be set before launching Python.

pip install nvidia-cublas-cu12 nvidia-cudnn-cu12

export LD_LIBRARY_PATH=`python3 -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`

Did you see this in installation instructions? You need to do export LD_LIBRARY_PATH=... every time in the terminal where you are running faster-whisper.

[thomasmol]

We experienced this issue with torch v2.4. Downgrading to v2.3.1 fixed it for us.

this fixed it for me as well!

[Denis-Kazakov]

Note that LD_LIBRARY_PATH must be set before launching Python.

pip install nvidia-cublas-cu12 nvidia-cudnn-cu12

export LD_LIBRARY_PATH=`python3 -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`

Did you see this in installation instructions? You need to do export LD_LIBRARY_PATH=... every time in the terminal where you are running faster-whisper.

Hi! No, I did not see it and do not see it now at https://github.com/SYSTRAN/faster-whisper

Anyhow, I started PyCharm and ran this command in the terminal but got the same error.

[Mushy-Snugglebites-badonkadonk]

No need to downgrade. Symlink to the .so.9 version of the affected files as a temporary fix until the maintainer updates the faster-whisper repo. Annoying, I know.

https://github.com/open-webui/open-webui/discussions/5253#discussioncomment-10583222 https://github.com/open-webui/open-webui/discussions/4270#discussion-7003841

#!/bin/bash

DESIRED_LIBS=("libcudnn.so.9" "libcudnn_adv.so.9" "libcudnn_cnn.so.9" "libcudnn_ops.so.9")
SOURCE_DIR="/usr/local/lib/python3.11/site-packages/nvidia/cudnn/lib"

# Greet the user
echo "Hello! Starting the CUDA library symlink creation process."

# Check if source directory exists
if [[ ! -d "$SOURCE_DIR" ]]; then
  echo "Source directory $SOURCE_DIR does not exist. Exiting."
  exit 1
fi

# Inform the user about the source directory being used
echo "Using source directory: $SOURCE_DIR"

for lib in "${DESIRED_LIBS[@]}"; do
  case $lib in
    "libcudnn.so.9")
      base_name="libcudnn.so.8"
      ;;
    "libcudnn_adv.so.9")
      base_name="libcudnn_adv_infer.so.8"
      ;;
    "libcudnn_cnn.so.9")
      base_name="libcudnn_cnn_infer.so.8"
      ;;
    "libcudnn_ops.so.9")
      base_name="libcudnn_ops_infer.so.8"
      ;;
  esac

  # Inform the user about the current operation
  echo "Creating symlink for $lib as $base_name"

  ln -sf "$SOURCE_DIR/$lib" "$SOURCE_DIR/$base_name"
  if [[ $? -eq 0 ]]; then
    echo "Successfully created symlink for $lib"
  else
    echo "Failed to create symlink for $lib"
  fi
done

# Inform the user that the process is complete
echo "CUDA library symlink creation process completed."