-
Notifications
You must be signed in to change notification settings - Fork 3.6k
Description
Describe the bug
I migrated my code from onnxruntime 1.4.0 to 1.5.2 version. When i load ONNX model with Microsoft.ML.OnnxRuntime 1.5.2, all works, but if i load model with Microsoft.ML.OnnxRuntime.Gpu 1.5.2, i've got error:
Unable to load DLL 'onnxruntime': The specified module could not be found. (Exception from HRESULT: 0x8007007E)
When i use Microsoft.ML.OnnxRuntime.Gpu 1.4.0, all works.
System information
OS Platform and Distribution: Windows 10 Pro 20H2
ONNX Runtime installed from (source or binary): nuget and binary
ONNX Runtime version: 1.5.2
Visual Studio version (if applicable): 2019
CUDA/cuDNN version: CUDA 10.2 / cudnn-10.2-windows10-x64-v7.6.5.32
GPU model and memory: NVIDIA GeForce GTX 1080Ti 8Gb
To Reproduce
var so = SessionOptions.MakeSessionOptionWithCudaProvider(0);
so.GraphOptimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL;
var session = new InferenceSession(modelPath, so);
Or
var session = new InferenceSession(modelPath); // same error
Expected behavior
Loading the model without errors.
Example model: https://yadi.sk/d/e5EhowZzndcXTg
I completely uninstalled CUDA 10.1, and install CUDA 10.2 after that. But it didn't help. I.ve got the same error:
Unable to load DLL 'onnxruntime': The specified module could not be found. (Exception from HRESULT: 0x8007007E)
I tried replace onnxruntime.dll in build folder on onnxruntime from release:
https://github.com/Microsoft/onnxruntime/releases/tag/v1.5.2
Doens't help.