Encounter undefined symbol: _ZN11nvrtcHelper4loadEb using Wav2LipPredictor

[rsandx]

The Wav2LipPredictor usually works fine on my system, but sometimes it fails with certain audio file, because of the following error: could not load library libcudnn_cnn_infer.so.8. Error: /lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8: undefined symbol: _ZN11nvrtcHelper4loadEb, version libcudnn_ops_infer.so.8 I searched the web but couldn't find the same error reported, and I'm sure the cuda installation is fine on my system (Ubuntu 20.04.4 LTS) since Wav2LipPredictor works correctly for most cases, what could be the cause of above error and how can I solve it?

[drdsgvo]

Have the same issue when trying to use AutoPipelineForText2Image from diffusers: #load model, works pipeline_text2image = AutoPipelineForText2Image.from_pretrained( model_local, torch_dtype=torch.float16, variant="fp16", use_safetensors=True ).to("cuda")

#generate, here comes the error image = pipeline_text2image(prompt="any prompt").images[0]

Any other cuda usage works: whisper, text generation, cuda checks...

[LokeZhou]

You can come here to pick up iuses https://github.com/PaddlePaddle/PaddleMIX/issues