levnikolaevich
commented
3 months ago Have the same problem. I've made the steps to install whisperx exactly from the repository guide
Ubuntu 20.04, wsl2
Open jessienab opened 3 months ago
levnikolaevich
commented
3 months ago Have the same problem. I've made the steps to install whisperx exactly from the repository guide
Ubuntu 20.04, wsl2
GuyPaddock
commented
3 months ago It seems like perhaps pytorch and torchaudio need to be updated to work with the latest Nvidia integration. There are a slew of other issues out there like this as of February:
https://github.com/AbdBarho/stable-diffusion-webui-docker/issues/649
Unfortunately, I don't get much better results when updating to newer versions:
RuntimeError: parallel_for failed: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the deviceRuntimeError: parallel_for failed: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the deviceRuntimeError: parallel_for failed: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the deviceCould not load library libcudnn_ops_infer.so.8. Error: libcudnn_ops_infer.so.8: cannot open shared object file: No such file or directoryGuyPaddock
commented
3 months ago Looks like the errors I was seeing with 2.1.0 through 2.3.0 are covered by https://github.com/m-bain/whisperX/issues/794 and the workaround there works for me when using this environment.yaml:
name: whisperX
channels:
- pytorch
- nvidia
- defaults
dependencies:
- python=3.10
- pip
- pytorch==2.3.1
- torchaudio==2.3.1
- pytorch-cuda=11.8
- torchvision==0.18.1
- pip:
- git+https://github.com/m-bain/whisperx.git
GUUser91
commented
3 months ago I want to give a heads up to those not using conda and using a standard virtual python environment. Install these
pip install nvidia-cudnn-cu12==8.9.7.29pip install ctranslate2Find the path of the libcudnn file
sudo find / -type f -iname '*libcudnn_ops_infer.so.8*'export it to the terminal, it should look something like this
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/user/whisperx/lib/python3.11/site-packages/nvidia/cudnn/lib/Now you it should work
whisperx --model large-v3 --language en audio.wav
totomtornado
commented
2 months ago I am running the cpu install on linux... I encountered this same error... searched around a lot.... I was able to get things running by using:
pip install -U torch torchaudio
NOTE: I also needed to add a flag to my command when I called whisperx...
--compute_type int8
this was detailed #348
electricazimuth
commented
2 months ago totomtornado solution :
pip install -U torch torchaudiofixed it for me
Pastells
commented
1 month ago
pip install -U torch torchaudioworked for me
Hello, I attempt to start a transcription using a cloud compute provider, but I encounter an error I can't find any information about online...
Genuinely unsure how to proceed. I followed install instructions (near) exactly per README:
Input audiofile is 2 channel FLAC 44.1khz/16-bit or WAV pcm_s16le.
Thank you!