ONNX GPT Inference on two devices

[11721206]

I export ONNX model and inference, I found " CUDA kernel not found in registries for Op type: ScatterND" and Memcpy From Host and to Host, I have issue in onnxruntime repo here ：microsoft/onnxruntime#21148 , and still do not solve it. In python InferenceSession, I print the sess.get_providers() result, and show ["CUDAExecutionProvider", "CPUExecutionProvider"], it mean some ops work on CPU. Is there any sugeestion for my problems? Thanks.