Weighted Random Sampling with CUDA

I am writing an algorithm that involves

• Having a list of values on the GPU
• Sampling from these values based on probabilities.

For example, the probabilities are calculated directly from the values on the GPU.

However, I want to avoid cloning the values on the GPU, as they are quite large.

Is there a way to implement something like numpy.random.choice with weights on a GPU using CUDA?