Brodcasting error when saving recording after deep interpolation

Hi,
I'm having issues with storing the recording object after applying deepinterpolation.
When saving an interpolated recording, I get error messages like:

ValueError: could not broadcast input array from shape (29886,768) into shape (30000,768)

I assume that is because the convolutional network removes samples at the edges of the signal.
What is the recommended way of dealing with this? Should I pad the traces before applying deepinterpolation? If so, is there a function for this? Or would it be better to create a new object for storing the interpolated recording?

Below is a code example. I'm using spikeinterface version 0.100.6 because deepinterpolation requires python <= 3.8

from pathlib import Path
import numpy as np
import spikeinterface.extractors as se
from spikeinterface.preprocessing import common_reference, deepinterpolate, zero_channel_pad

root = Path(__file__).parent.parent.absolute()
model_path = list((root/"model").glob("*"))[0]

recording = se.read_spikeglx( folder_path=root/"data"/"neuropixels")

# pad so dimensions are compatible with model
recording = zero_channel_pad(recording, num_channels=384*2) 

# create preprocessing pipeline
recording_cmr = common_reference(recording=recording, operator="median")
recording_deepint = deepinterpolate(recording=recording_cmr, model_path=str(model_path), use_gpu=False)

# process and save
recording_deepint.save()

Thanks in advance!

Start with the deepinterpolate(...) preprocessing entry point and the recording_deepint.save() call in the supplied example. Reproduce the shape mismatch with the shown padding and inspect how the interpolated recording handles sample counts; done means the save path either handles the shortened traces or clearly documents the required recording workflow.