search

beeldengeluid / whisper-asr-worker

MIT License
2 stars 0 forks source link

Look into Govert's Dockerfile rework #84

Open greenw0lf opened 1 week ago

greenw0lf commented 1 week ago

See if I can adapt it to the new code and if it works + has benefits compared to the current approach, otherwise drop it for now.

greenw0lf commented 1 week ago

Tried to adapt and it worked on the CPU, but as expected I had trouble running it on the GPU. I tried to install the CUDA dependencies needed (because they are definitely not present) via

pip install nvidia-cublas-cu12 nvidia-cudnn-cu12==8.9.7.29
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__))'`

(from https://github.com/SYSTRAN/faster-whisper/)

and that failed, so I also tried just installing torch. Neither approach worked (plus, the image size is larger than the currently active one) so I will leave this aside for now and focus on more important issues.

KleinRana commented 1 week ago

Discuss with @gb-beng before closing this issue