search

KoljaB / RealtimeSTT

A robust, efficient, low-latency speech-to-text library with advanced voice activity detection, wake word activation and instant transcription.
MIT License
2.09k stars 190 forks source link

[W NNPACK.cpp:64] Could not initialize NNPACK! Reason: Unsupported hardware. #21

Closed kkkjackyboy closed 1 week ago

kkkjackyboy commented 10 months ago

I have tried to install & reinstall in ubuntu nvidia graphic driver and cuda tool kit. But I can't get rid of this problem.

kkkjackyboy commented 10 months ago

I have tried to install on two different computers with different types of nvidia graphic cards. got the same error:

[W NNPACK.cpp:64] Could not initialize NNPACK! Reason: Unsupported hardware. 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 Please make sure libcudnn_ops_infer.so.8 is in your library path! Aborted (core dumped)

KoljaB commented 7 months ago

Sorry for late feedback. I'm bad at linux, but the error message includes "libcudnn_ops_infer", so I'd guess maybe the cuDNN installation is missing or somehow corrupted?

I'd try to verify cuDNN installation. If there is everything ok, I'd pip install any library mentioned in requirements.txt one by one to find the one that causes this. Then we can look into their github issues if there's anything mentioned regarding this.

IlyaShkurenko commented 2 months ago

If anyone struggling the same issue use this https://github.com/SYSTRAN/faster-whisper/issues/516#issuecomment-1902687567

I've used this command: export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$VIRTUAL_ENV/lib64/python3.11/site-packages/nvidia/cublas/lib:$VIRTUAL_ENV/lib64/python3.11/site-packages/nvidia/cudnn/lib

KoljaB commented 2 months ago

Please try

sudo apt-get install libcudnn8

I'm guess the nnpack error is a pytorch issue. It says "unsupported hardware". nnpack can run on x86-64, ARMv7, or ARM64 architecture CPUs. So maybe the used hardware does not meet this requirements? I'd assume with this nnpack error anything else using torch wouldn't work then too?