[Feature Request] prebuilt package with cudnn 9 support about onnxruntime HOT 6 CLOSED

we plan to publish 1.18.1 gpu packages built against CUDA 12.x and CuDNN 9.x later this month.
(in addition to the default gpu packages built against CUDA 11.8 + CuDNN 8.x)

from onnxruntime.

@jywu-msft Thx.
Everything went great when I test this whl in a docker env nvcr.io/nvidia/pytorch:24.05-py3 (cuda12.4, tensorrt 10.0.1, cudnn 9.1), using some simple models from torchvision.
Will report isssues if encountered.

from onnxruntime.

Can you help test out these Release Candidate Packages which support CUDA 12.x , TensorRT 10 and CuDNN 9.x?
https://aiinfra.visualstudio.com/PublicPackages/_artifacts/feed/onnxruntime-cuda-12/NuGet/Microsoft.ML.OnnxRuntime.Gpu.Linux/overview/1.18.1-rc
https://aiinfra.visualstudio.com/PublicPackages/_artifacts/feed/onnxruntime-cuda-12/NuGet/Microsoft.ML.OnnxRuntime.Gpu.Windows/overview/1.18.1-rc

from onnxruntime.

@jywu-msft Not so sure how to use this package. Is there a pypi whl package？

from onnxruntime.

no, those were nuget packages.
here are whl's
https://aiinfra.visualstudio.com/PublicPackages/_artifacts/feed/onnxruntime-cuda-12/PyPI/ort-nightly-gpu/overview/1.18.1.dev20240624003

from onnxruntime.

@jywu-msft Thx. Everything went great when I test this whl in a docker env nvcr.io/nvidia/pytorch:24.05-py3 (cuda12.4, tensorrt 10.0.1, cudnn 9.1), using some simple models from torchvision. Will report isssues if encountered.

Thank you!

from onnxruntime.