[GPU] Device selection issue

Hello,
I'm currently working based on the commit of [f0ae822], I'm using several GPUs on my server, so I need to select the device.
I have both explored C++ and python versions:
in C++, when I do:
Oga::SetCurrentGpuDeviceId(3);
then I got:

2025-06-11 21:13:01.366170206 [E:onnxruntime:, inference_session.cc:2117 operator()] Exception during initialization: /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:129 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudnnStatus_t; bool THRW = true; SUCCTYPE = cudnnStatus_t; std::conditional_t<THRW, void, common::Status> = void] /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:121 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudnnStatus_t; bool THRW = true; SUCCTYPE = cudnnStatus_t; std::conditional_t<THRW, void, common::Status> = void] CUDNN failure 2007: CUDNN_STATUS_BAD_PARAM_STREAM_MISMATCH ; GPU=0 ; hostname=gpumixt ; file=/onnxruntime_src/onnxruntime/core/providers/cuda/cuda_execution_provider.cc ; line=183 ; expr=cudnnSetStream(cudnn_handle_, stream);

terminate called after throwing an instance of 'std::runtime_error'
what(): Exception during initialization: /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:129 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudnnStatus_t; bool THRW = true; SUCCTYPE = cudnnStatus_t; std::conditional_t<THRW, void, common::Status> = void] /onnxruntime_src/onnxruntime/core/providers/cuda/cuda_call.cc:121 std::conditional_t<THRW, void, onnxruntime::common::Status> onnxruntime::CudaCall(ERRTYPE, const char*, const char*, SUCCTYPE, const char*, const char*, int) [with ERRTYPE = cudnnStatus_t; bool THRW = true; SUCCTYPE = cudnnStatus_t; std::conditional_t<THRW, void, common::Status> = void] CUDNN failure 2007: CUDNN_STATUS_BAD_PARAM_STREAM_MISMATCH ; GPU=0 ; hostname=gpumixt ; file=/onnxruntime_src/onnxruntime/core/providers/cuda/cuda_execution_provider.cc ; line=183 ; expr=cudnnSetStream(cudnn_handle_, stream);

Otherwise, if I use python:
import onnxruntime_genai_cuda as og
og.set_current_gpu_device_id(1)

I get the segmentation fault directly.

My code works very well, if I don't select the GPU number. But I need to absolutely use this function.

Anyone has any idea?
Thank you

Start from Oga::SetCurrentGpuDeviceId(3) and og.set_current_gpu_device_id(1), using commit f0ae822 and the reported CUDNN error in cuda_execution_provider.cc. Reproduce both the C++ initialization failure and Python segmentation fault, then trace GPU selection through initialization. Done means both bindings select the requested GPU without the stream-mismatch error or a crash.