Is it possible to run inference on a Mamba2 model using only the CPU?

[daisy98cc]

Hi,

I have tried training a model by simplifying an example model with the Mamba2 block. I am now attempting to run inference using only the CPU. During my tests, I noticed that the model relies on prebuilt selective_scan_cuda and causal_conv1d_cuda. Is it possible to modify the model to use CPU instead of CUDA for these operations?

Thank you for your help!