Error return type mismatch in onnx_helper.py

In line 86 in quickstart/IntroNotebooks/onnx_helper.py, in the method predict err gets assigned the result of cudart.cudaMemcpyAsync which as per the documentation should be of type cudart.cudaError_t.
The same goes for cudart.cudaStreamSynchronize in line 107

However the actual calls return a tuple int the form (err, ) which leads to the following error being thrown when the method is called

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[5], line 1
----> 1 predictions = trt_model.predict(dummy_input_batch)

File [/data/gsg/onnx_helper.py:90](http://192.168.2.115:8888/lab/tree/data/gsg/data/gsg/onnx_helper.py#line=89), in ONNXClassifierWrapper.predict(self, batch)
     86 err = cudart.cudaMemcpyAsync(
     87     self.d_input, batch.ctypes.data, batch.nbytes, cudart.cudaMemcpyKind.cudaMemcpyHostToDevice, self.stream
     88 )
     89 if err != cudart.cudaError_t.cudaSuccess:
---> 90     raise RuntimeError(f"Failed to copy input to device: {cudart.cudaGetErrorString(err)}")
     92 # execute model
     93 self.context.execute_async_v3(self.stream)

File [/usr/local/lib/python3.10/dist-packages/cuda/cudart.pyx:13554](http://192.168.2.115:8888/lab/tree/data/gsg/usr/local/lib/python3.10/dist-packages/cuda/cudart.pyx#line=13553), in cuda.cudart.cudaGetErrorString()

AttributeError: 'tuple' object has no attribute 'value'

This can be remedied by checking whether err is a tuple and proceeding accordingly. Another variation may be to directly unpack the returned tuple.

The code is run under jetson-containers run $(autotag l4t-ml) with L4T 36.4.4 on Jetson Orin Nano 8gb Developer Kit

!pip list | grep cuda
cuda-python                  12.6.0+0.gb9f40f6.dirty   /usr/local/lib/python3.10/dist-packages pip
jax-cuda12-pjrt              0.6.0.dev20250414         /usr/local/lib/python3.10/dist-packages pip
jax-cuda12-plugin            0.6.0.dev20250414         /usr/local/lib/python3.10/dist-packages pip
pycuda                       2024.1.2                  /usr/local/lib/python3.10/dist-packages pip

Start in quickstart/IntroNotebooks/onnx_helper.py at the predict method, especially lines 86 and 107, and reproduce the failure by calling predict with the notebook's dummy_input_batch. Check the return values from cudaMemcpyAsync and cudaStreamSynchronize against the documented types. Done means predict handles the returned values without the tuple-related AttributeError on the reported Jetson environment.