Transfer skorch models from CUDA to CPU for inference

[lukethomrichardson]

Hi all, I am training skorch models locally in a CUDA-enabled torch environment, and, if possible, I would like to transfer the entirety of the model to CPU so that they can be registered and used for inference in a CPU-only environment. Is there a best method for accomplishing this?

I'm pretty new to skorch and deep learning so I'm not sure if this is even possible, but, if so, a skorch helper method for converting a model to CPU would be a nice-to-have feature.

Edit: Just noticed a very similar (old) issue that is still open at time of posting (#553). The conversation there didn't seem to completely resolve. Let me know if should post there or if reviving this topic here would be preferable.

[BenjaminBossan]

If you train a model on GPU, save it, then load it on a machine without GPU, it should already work and be automatically transferred to CPU. Please give this a try and tell us if you encounter problems.

The thread you cited is a bit different, as it is about changing the device within the same process.