speaker-diarization-3.0 won't load, ONNX error

[sedol1339]

Tested versions

pyannote-audio=3.3.2

System information

Ubuntu 24.04, CUDA 12.4, CuDNN 9.6.0, onnxruntime-gpu 1.20.1, Nvidia Driver Version 550.120

Issue description

Hello! I encounter a problem: a speaker-diarization-3.0 pipeline don't load in my environment (3.1 version does load):

from pyannote.audio import Pipeline
import torch
import os

pipeline = Pipeline.from_pretrained(
    'pyannote/speaker-diarization-3.0',
    use_auth_token=os.getenv('HF_TOKEN')
).to(torch.device('cuda'))

Output:

INFO:speechbrain.utils.quirks:Applied quirks (see speechbrain.utils.quirks): [disable_jit_profiling, allow_tf32]
INFO:speechbrain.utils.quirks:Excluded quirks specified by the SB_DISABLE_QUIRKS environment (comma-separated list): []
2024-12-15 14:28:01.224680625 [E:onnxruntime:Default, provider_bridge_ort.cc:1862 TryGetProviderInfo_CUDA] /onnxruntime_src/onnxruntime/core/session/provider_bridge_ort.cc:1539 onnxruntime::Provider& onnxruntime::ProviderLibrary::Get() [ONNXRuntimeError] : 1 : FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: /lib/x86_64-linux-gnu/libcudnn_heuristic.so.9: undefined symbol: _ZTVN5cudnn7backend23PagedCacheLoadOperationE, version libcudnn_graph.so.9
2024-12-15 14:28:01.224692208 [W:onnxruntime:Default, onnxruntime_pybind_state.cc:993 CreateExecutionProviderInstance] Failed to create CUDAExecutionProvider. Require cuDNN 9.* and CUDA 12.*. Please install all dependencies as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

As the error message suggests, I need cuDNN 9.* and CUDA 12.*, however I already have them installed.

Does this mean that this speaker-diarization-3.0 is outdated and won't work in a new versions of CUDA or pyannote-audio?

Minimal reproduction example (MRE)

What about MRE colab notebook: in this notebook the problem does not occur. However, Colab has CUDA 12.2 which is different. And at the same time Colab does not even have onnxruntime-gpu installed, so ONNX in Colab is installed in some other way i'm not sure about, and don't know how to check the version.

[clement-pages]

Hey @sedol1339,

As mentionned in pyannote/speaker-diarization-3.1:

This pipeline is the same as pyannote/speaker-diarization-3.0 except it removes the problematic use of onnxruntime.

So using pyannote/speaker-diarization-3.1 should fix your issue with onnxruntime.

[sedol1339]

Ok thank you!